Science and Technology (Bromley) 1991 [8]

Originally Processed With FOIA(s): FOIA Number: 1998-0004-F[2]; 2005-0336-F S FOIA MARKER This is not a textual record. This is used as an administrative marker by the George Bush Presidential Library Staff. Record Group/Collection: George H.W. Bush Presidential Records Collection/Office of Origin: Chief of Staff, White House Office of Series: Sununu, John, Files Subseries: White House Offices Files OA/ID Number: 29184 Folder ID Number: 29184-008 Folder Title: Science and Technology (Bromley) (1991) [8] Stack: Row: Section: Shelf: Position: G 15 25 6 1 CONTENTS Tab A. Satellite Measurements of Global Atmospheric Temperatures Tab B. Ground-based Temperature Measurements - 100-year Data Set and Urbanization Effects Tab C. Nighttime Warming versus Daytime Warming and Effects Tab D. The Ocean Conveyor Belt Theory Tab E. The Gore 6-month Temperature Update (Gore requested these authors to send him a temperature/data update every 6 months) Tab F. Schlesinger and Jiang - "Revised Projection of Future Greenhouse Warming" Reprint Series 30 March 1990, Volume 247, pp. 1558-1562 SCIENCE Precise Monitoring of Global Temperature Trends from Satellites Ror W. SPENCER AND JOHN R. CHRISTY Copyright © 1990 by the American Association for the Advancement of Science NOSTIM S 9 01 ESAD)3541 FROM 22:01 16, 9 NNS Precise Monitoring of Global Temperature Trends from Satellites Roy W. SPENCER AND JOHN R. CHRISTY now operating and discuss data obtained from 1979 to 1988. Passive microwave radiometry from sarellites provides Methodology. In lare 1978, a scries of passive microwave more precise atmospheric temperature information than radiometers was launched aboard the TIROS-N series of National that obtained from the relatively sparse distribution of Oceanic and Aumospheric Administration (NOAA) satellites. These thermometers over the earth's surface. Accurate global radiometers, or microwave sounding units (MSUs), are Dicke-type atmospheric temperature estimates are needed for detec- radiometers designed to measure the thermal emission of radiation tion of possible greenhouse warming, evaluation of com- by atmospheric O: at four frequencies near 60 GHz (4). The puter models of climate change, and for understanding atmospheric concentration of O: is constant in both space and time important factors in the climate system. Analysis of the (5), and thus O₂ provides a stable temperature tracer. The strong first 10 years (1979 to 1988) of satellite measurements of interaction of radiation from 50 to 70 GHz with O₂ through lower atmospheric temperature changes reveals a monthly rocational energy transitions causes absorprion and emission. As the precision of 0.01°C, large temperature variability on time channel frequency of the MSU approaches the 60-GHz peak in this scales from weeks to several years, but no obvious trend absorprion complex, higher levels in the atmosphere will be mea- for the 10-year period. The warmest years, in descending sured (Fig. 1) (6). We have analyzed data from MSU channel 2, order, were 1987, 1988, 1983, and 1980. The years which measures the temperature of the middle troposphere ar 53.74 1984, 1985, and 1986 were the coolest. GHz. At 57.95 GHz, MSU channel 4 can be used to monitor temperatures of the lower stratosphere. MSU channels 1 and 3 are more difficult to interpret for climate purposes because channel 1 is too sensitive to surface effects on the earth and cloud water, whereas A CCURATE ESTIMATES OF GLOBAL ATMOSPHERIC TEMPERA- channel 3 detects radiation from a strong temperature-ransition nures are needed for evaluation of global climare models, region between the troposphere and stratosphere (the tropopause). detecrion of climate changes, and a better understanding of The four channels have traditionally been used to obrain verrical the climate system. Global temperatures have generally been esti- profiles of temperature in remote regions of the earth where weather mated from surface temperature records, but there has been much balloon data are not available. However, because the weighting debate regarding, for example, whether these dara provide evidence of recent greenhouse warming (1). The primary source of uncertain- 1 ty is the relatively sparse distribution of thermometers over the surface of the earth. Most of the earth is covered by oceans, and vast oceanic areas go unmeasured Even over land, the coverage is 2 greatest where the population is greatest; therefore, remore land areas also go unmeasured. An additional problem is that urban sites, Flg. 1. Temperature 3 which represent only a small part of the globe but where many long- weighting functions (unicless) for MSU term measurements have been made. have warmed because of hear channels 2 (mid-tropo- 5 from man-made structures, and thus these records are difficult B sphere) and + (lower interpret (2. 3). Depending upon how the thermometer data are stratosphere). Also 7 analyzed, various answers can be expected. In contrast to surface shown are the different Pressure (kPa) Ch.4 channel 2 weighting 10 (57.05 GHz) thermometers, sensors on satellite platforms can provide nearly functions for ocean and complete earth coverage in as little as one day and can obtain land surfaces. which measurements from various levels of the atmosphere. Calibration of arise because the less emissive ocean reflects 20 satellite sensors is particularly difficult, however. For climate tem- more of the downwell- perature monicoring a precision of 0.1°C is needed, a goal that has ing atmospheric radia- 30 been perceived as difficult for any earth viewing radiometer. The cion back upward. Sensi- Ch.2 difficulty arises from uncertainty about the long-term stability of tivity to the surface radi- (53.74 GHz) satellite sensors. In this article, we show that accurate long-term ation itself cannot be im- 50 Ocean global temperature measurements can be obrained from satellites plied from the magni- 70 Land rudes at the intersection R. W. Spencer. is at Marshall Space Flight Center Huntsville, Code ES43. Huntsville. of the curves with the 100 AL 35812. J. R. Christy is at Johnson Research Center, University of Alabama. surface. [Adapred from 0.0 0.2 0.4 0.5 0.8 1.0 Huntsville, AL 35899. (6)] Temperature weighting function 1558 SCIENCE, VOL 247 800 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 2:02 16. 9 NNS functions for each channel are vereically broad (Frg. 1), retrieval of 2-day intervals over both the Northern and Southern hemispheres in adequare dara on the verrical temperature structure of the atmo- 2.5°-latitude bands with cosine-latitude weighting to account for the sphere that are needed for computer modeling of the weacher has decrease in surface area of these bands toward the poles. We reduced been difficult. The temperarure measurement within a constant- contamination of the measurements by large thunderstorm com- pressure depth. however, rather than at traditionally measured plexes, which cause infrequent depressions in Tb over small areas (6, specific pressure levels, is completely adequare. even preferable in 7), by excluding any scan lines that had individual footprint some respects, for climare monitoring. measurements that deviated by more than 1.5°C from their average The MSUs are externally calibrated, after each earth scan of the relationship to both neighboring footprint measurements. The instrument, by measurement of the cosmic background radiation resulting variations in the 2-day averages are dominated by the (for our purposes, constant at 2.7 K). and a warm target in the seasonal change of temperature (8). termed the annual cycle. This instrument that has its temperature monitored with redundant annual cycle is simusoidal in shape, and a smooched cycle was platinum resistance thermometers. This calibrarion design is consid- computed for each satellite individually. Then this cycle was sub- cred to be the best available for microwave radiometers because any tracted from the original 2-day time series to arrive at the anomalies temperature changes in the instrument components are canceled in temperature, that is, the temperature deviarions from the average our The earth-viewing measurements are then calculated as a temperature for a particular time of year. A period of data considera- "brightness temperature" (Tb) by interpolation between these two bly longer than 2 years is necessary for a more representative annual reference extremes. The term "brightness temperature" acknowl- cycle (and thus, more representative anomalies), but the short-term edges that the temperature measurement is actually based upon dara are utilized only for sarellite intercomparison. Several revealing radiative brightness and is only equal to 2 thermometric temperature observations can be made from the data (Fig. 2): when the emitting body is completely "black" (nonreflective). This 1) The standard deviation (SD) of the sums of the dara from the condition is nearly true of measurements of the armosphere from two sarellites is much larger than the SD of their differences. This MSU channel 2. relation means that both satellites were measuring nearly the same Satellite intercomparison. Because two MSUs have usually been temperature variations and implies that hemispheric temperature simultancously operating on separate satellites, a comparison be. anomalies can be measured with relatively little CITOR from 3 single tween them shows how different sensors agree in their measure- sarellite. The 2-day average difference between these satellites was ments and gives an estimate of errors. The NOAA-6 and NOAA-7 about 0.05°C, and for monthly averages, the difference improved to MSUs were simultaneously operating during a period of nearly 2 about 0.011°C. Similar noise was found in monthly comparisons years (29 June 1981 to 16 April 1983). These satellites are in sun- between data acquired from sensors on (i) TIROS-N and NOAA-6 synchronous, near-polar orbins, and have constant local crossing (0.012°C), (ii) NOAA-7 and NOAA-9 (0.012°C), (iii) NOAA-9 times of 7:30 a.m. and p.m. and 2:30 2.m. and p.m., respectively. and NOAA-10 (0.006°C), and (iv) NOAA-10 and NOAA-11 The precession rates of their orbics are quite different so that in a (0.008°C), although much shorter overlap periods were available. single day there are many differences (but also many overlaps) in the Thus, we estimate that the precision of monthly satellite measure- areas of the globe sampled by the two satellites. The MSUs scan across the satellite subtrack, and thus paint out swaths of coverage 0.5° about 2000 km wide beneath the sacellites. NOAA-7 NOAA-6 We averaged the channel 2 Tb dara from the separate satellires at Globa Hemispheric anomaly (cc) 14 0.0° +0.6° as -0.0233° =0.1584° -0.5° 1962 1983 1982 1983 0.00 30° -0.6° 50 10 Northern Hemisphere +0.6° & -0.0315° a, 9.0.200 MSU channel 2' T, anomalies (C') 30° 0.00 Latitude 3+ EO EQ EQ -0.6° Southern Hemisphere 30° +0.6° as -0.0318° $=0.1771° 50° 0.0° 90° -0.6° 1982 1983 1982 1083 29 June 18 April Year 1981 1983 Fig. 3. Low-pass filtered hemispheric (cop) and zonally averaged (bottom) Fig. 2. Comparison between global MSU channel 2 Tn values from NOAA- MSU channel 2 T. anomalies during the 2-year overlap period of NOAA-6 6 and NOAA-7 during 2 nearty 2-vear period (29 June 1981 to 16 April (right) and NOAA-7 (left); "N" and "S" labels represent Northern and 1983) when both satellites had MSUs operating. Time series of global and Southern hemispheres, respectively, and the global ame series is the heavy hemispheric sarellite averages (sum divided by 2) and differences (difference line. The time series do not agree new the beginning and end of the period divided by 2) are shown by the large-variation and small-variarion curves, because of the low-pass filter and the lack of dara past the end of the period respectively. The difference time series are offset -0.4°C for legibility. The for NOAA-6 or before the beginning of the period for NOAA-7. Warm SD of the two-satcllite sums (σ₂) and differences (σₑ) are also listed. zonally averaged anomalies are srippled. 30 MARCH 1990 RESEARCH ARTICLE 1559 PAGE 004 NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 82:01 16, 9 NNS 3 +2 A Thermometer c amomaly enomaly (ec) (c) 0 person +1 7 1 B MSU Sale/Mile T anomaly (°C) o 0 -1 -1 1978 1980 1982 1984 1986 Fig. 4. Time series of monthly U.S. surface air temperature anomalies (A) and MSU channel 2 Tb anomalies -2 (B) for the period 1979 to 1987. Their scatterplor (c) has a correlarion coefficiem (r) of 0.89, and an explained variance (7) of 79 percent. March and April 1981 satellite anomalies are not included because of a high -2 0 2 percentage of missing daca. Surface T anomaly (C) ments is about =0.01°C for the globe. tors have made point comparisons between weacher balloon dara 2) The sums of the two sets of data reveal that dramatic globally and satellite measurements (10-12). The differences between indi- averaged warming and cooling events of greater than 0.5°C can vidual radiosonde and satellite measurements are generally less than occur in less than 2 weeks. The warmings, representing huge energy 1.0°C. These differences are usually attributed to (i) the isolated exchanges, are possibly associated with stormy periods when large balloon sampling compared to the large area represented by a single amounts of latent heat were released in precipitation of moisture satellite measurement (a circular footprint 110 km in diameter); (ii) previously evaporated from the sun-warmed ocean. The coolings errors in calibration of the balloon thermometer before its release; might be from formation of widespread low-level cloudiness, which (iii) the random noise of a single MSU measurement, about 0.3°C; reflects significant amounts of incoming solar radiation. and (iv) time mismatches between the satellite and balloon observa- 3) The long-term drift of one instrument relative to the other, nons. We compared the 1980 through 1988 MSU observations to seen in the difference time series in Fig. 2, is so small that it is Tb data obtained from radiosondes launched twice daily by 66 virtually unmeasurable. Any trend is less than =0.01°C for the 2- National Weather Service offices around the United States. The Tb year period. This high degree of stability was unexpected. The four data were calculated with the radiative transfer equarion for MSU other satellite overlaps mentioned above also gave no indication of channel 2, and thus the radiosondes had to reach a fairly high- drift. pressure altitude, 2 kPa or less. Only MSU data within 200 km of Further evidence that the measurements were repeatable is shown the radiosonde location and within 3 hours of its release time were by low-pass filtered (9) times series of hemispheric (and global) included. The comparisons revealed that there were biases of up to temperature anomalies from NOAA-7 and NOAA-6 dara and the = 1°C between the two data sets; the biases are most easily related to zonally averaged distribution of those anomalies (Fig. 3). The zonal the difference in time between the release time of the radiosondes averages allow examination of which latitude bands of the earth (all simultaneous at 00 and 12 GMT) and the MSU observation were responsible for the warm or cool events seen in the hemispheric times, which are sun synchronous. After correction for these biases and global time series. The zonal average patterns are nearly on 2 station-by-station basis, we found no long-rerm trend in the 9 identical between satellites. Such agreement improves our confi- years of differences berween MSU and radiosonde-calculared Tb dence that even regional areas can be studied to find the origins of values and a monthly SD of 0.068°C. the hemispheric anomalies. The MSU data also can be compared to records of temperature Comparisons with United States and global thermometer variability from near-surface thermometers. Even though they are measurements. Although the above results indicare that the sam- different variables, their common variability helps to assess how well pling provided by a single satellite is geographically excensive and coupled the near-surface temperature variations are to the deep layer that the measurements are radiometrically scable enough to be useful variations. Although the global distribution of thermometers is for monitoring climate, it still must be demonstrated that MSU suspected by many researchers as being inadequare for accurare measurements are closely related to temperature. Earlier investiga- monitoring. the distribution over the United States is widely -0.5 90°N A B Filter width = 45 anomaty (c) 60° Hemisphorte 0 0.0 30° Leitude 0" 8 " -0.5 1979 1961 1983 1985 1967 Year 30° Fig. 5. Hemispheric anomalies (A) and zonally averaged anomalies (B) about the average annual cycle of MSU channel 2 To values for the 10-vear 80° period 1979 through 1988; "5" and "N" refer to Southern and Northern 00 hemispheres, respectively. the global time scries is in 2 heavy line. Zonsider's 90°S anomalies are contoured every 0.25°; positive anomalies (warm) are solid 1979 1981 1983 1985 1987 and negative ones (cool) are dashed. Year 1560 SCIENCE, VOL 247 S00 PAGE NOSTIM S 9 01 ESAD)3541 FROM 82:01 16, 9 NNS accepted as good enough for climare work. We compared monthly agreement berween the calculations by Jones and the sarellite data is temperature anomalies from thermometers over the contiguous about 40% better than that berween the calculations by Hansen and United States for the period 1979 through 1987 (13) to our the satellite data. As might be expected for two land-dominated data monthly mid-tropospheric temperature anomalies from the satel- sets, the Jones and Hansen data are much better correlated with each lites. The resulting anomaly time series (Fig. 4, A and B) are similar, other, with an explained variance of 94%, chan either is with the as verified by 2 scarterplot (Fig. 4B) and a correlation coctficient of MSU data The much lower explained variances for the hemispheres 0.89. This correlarion agrees with those between radiosonde near- and globe (Table 1) compared to the United States are the result of surface and upper air measurements, for which monthly temperature both poor thermometer coverage over much of the earth and weak anomalies range from 0.8 for the eastern United States to greater than thermal coupling between the middle and lower troposphere over 0.9 for the western United Scares. The surface anomalies are typically much of the oceans. This latter effect was deduced from radiosonde two to three times as great as the MSU anomalies. This relation is near-surface and deep layer temperature comparisons: When probably a result of daytime solar heating and nightzime cooling of the monthly surface temperature anomalies from radiosonde are com- surface, which largely control the deeper air mass temperatures over the pared to the corresponding radiosonde-calculated channel 2 anoma- United States on monthly and seasonal time scales. lies for United Stares-controlled ocean stations, the explained Two major research groups have been responsible for hemispheric variances drops to abour 35% for the Caribbean, 0 to 20% for the calcularions of temperature anomalies from chermometer measure- tropical Pacific, and 25% for the tropical south Atlantic Although ments, and we refer to them by their leading authors names: Jones there are few high-latitude ocean radiosonde stations, the data (14, 15, 16) and Hansen (17). Their results have been sufficiendy suggest that the thermal coupling increases poleward (for example, different to spark debate in the climate community and have led to 52% in Iceland), probably because of the wider range of air mass conflicting reports in the popular press regarding global rempera- temperatures encountered there. Thus we would not expect as good nire crends. We have compared our satellite-measured hemispheric agreement between MSU anomalies and tropical ocean surface air anomalies to the thermometer-based anomalies from these two temperature anomalies as are obtained over land, even if the oceanic groups (Table 1). Again, when the separate systems are viewed as a thermometer coverage were adequare. This conclusion is substari- bivariant distribution in which we wish to determine the level of ated by the somewhat poorer agreement berween the sarellice dara common variability, these comparisons reveal that the level of and combined thermometer and sea surface temperature data from A R Fig. 6. Average MSU channel 2 Th (A) during the 10-day period 26 January to 5 February and temperature anomalies (B) for the same 10-day period during the 1983 ENSO climate anomaly. Temperatures range in top image is grav, repre- senting 235 K, to red, representing 260 K. in 1 K increments: temperature anomalics in bortom im- age change color every 0.25 K. the blue side of dark gray is colder. red side is warmer. 30 MARCH 1990 RESEARCH ARTICLE 1561 PAGE.006 NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 62:01 16. 9 NNS Table 1. Monthly and annual explained variances (in percent) between MSU that could be considered "average." This partern makes the defini- and thermometer-measured temperature anomalies for United States (U.S.), non of what is "normal" for global temperatures uncertain, for as Northern (NH) and Southern (SH) hemispheres, and the globe from 1979 shown above normal can mean either warm or cool conditions. to 1988. Thermometer-based calculations for the United States are from Karl (13); monthly and annual hernispheric and global thermometer anoma. The future. Our dara suggest that high-precision atmospheric lies are from Jones (14-16); annual anomalies are from Hansen and Lebedeff temperature monitoring is possible from satellite microwave radi- (17). and hemispheric and global anomalies, which also include sea surface omerers. Because of their demonstrated stability and the global temperatures. are from Farmer " al. (18). coverage they provide, these radiometers should be made the Source U.S. NH SH Globe standard for the monitoring of global atmospheric remperature anomalies since 1979. Their use will allow relatively precise monthly Monthly determinations of the locations and magnitudes of temperature Karl 77 Jones 33 10 35 change events. The resulting data should provide a greater focus of Farmer 26 11 28 scientific debare on why remperature anomalies occur rather than Annual whether they occur. The advanced microwave sounding units Karl 77 (AMSU) will replace the MSU on NOAA satellites in the mid- Hansen 67 27 53 1990s, and these units will allow extension of the time series into the Jones 72 66 74 next century. Various computerized climate models, which predict Farmer 77 24 69 furure changes through time-dependent equations representing physical processes, can now be evaluated with accurate global temperature measurements. These data should result in improved Farmer a al. (18) (Table 1). The poor agreement raises the specification of processes in these models, which still require important issue of whether near-surface temperatures or deep layer independent verification. These improvements should facilitate temperatures should be monitored for detecrion of climate change. more informed policy decisions concerning the effects of anthropo- Because they are often different from one another over the tropical genic greenhouse gas production. oceans. it would be best to monitor both in order to gain an understanding of how the entire troposphere behaves. Indeed, it might well be that the oceanic surface air laver is so strongly coupled REFERENCES AND NOTES to sea surface temperature variations that a deep layer mean would 1. R. A. New. Science 246. 1118 (1989). provide an earlier signal of possible greenhouse warming. 2. R. C. Balling and S. B. Idso. J. Geophys. Res. 94. 3359 (1989). 3. T. R. Karl H. E Diaz G. Kukla. J. Climate 1. 1099 (1988). Global temperature anomalics 1979 to 1988. The first 10 years 4. M. L Meeks and A E. Lilley, J. Grephys. Res. 68. 1683 (1963). of satellite data reveal large fluctuations in the hemispheric and 5. L. Machra and E. Hughes. Science 168. 1582 (1970). global temperatures (Fig. 5A). The Northern and Southern hemi- 6. N. C. Grody. J. Climate Appl. Memol. 22. 609 (1983). 7. R. W. Spencer. H. M. Goodman, R_ E. Hood. J. Atmos. Occur. Tech. 6. 254 spheres trends follow each other for the slower, interannual trends, (1989). but often oppose each other on monthly to seasonal time scales. The 8. Other, smalkr signals are also present in the measurements. These include a small warmest years, in decreasing order, were 1987, 1988. 1983, and surface temperature contribution (& percent of the total over land, 4 persont over the ocean). wind-induced roughening of the ocean surface, cloud liquid water 1980. There is no obvious long-term trend, and anomalies during effects. wher vapor variations, aca surface remperature changes, and soil moisture the first 5 years nearly balance those during the last 5 years. The changes. The effects of variations in each of these parameten on the measured To values have been theoretically evaluated, and have been determined to be small for 1988 warm event was rraced to the mid-latitudes, as was the 1980 MSC channel 2 (0.01°C or less) on a hemispheric and global basis. Larger efforts warm anomaly. Both years included summer heat waves in the could conceivably occur over small regions. In contrast, MSU channel 4 is United States. The largest warm anomaly in the Northern Hemi- exentially unaffected by any of these changes. The rediative transfer theory involved in the analysis of these effects is covered in (6). sphere for the 10-year record occurred in 1987 and 1988 (Fig. 5B). 9. Low-pass filrering of the time series allows isolation of the more slowty varying The 1987 and 1983 warm events were associated with El Niño/ temperature variacions that are of interest to climatologists. Here. the fiker retains Southern Oscillation events [ENSOs (19)]. During the 1983 50 percent of the power of cycles having 90-day periods, progressively 1 power of periods shorter than 90 days, and more power of longer periods. ENSO. transfer of heat from record-setting sca surface temperarures 10. N.C. Grody. Remore Seusing of the Atmosphere and OCEANS. A Deepak, Ed. (Academic in the eastern Pacific to the atmosphere caused major changes in Press, New York, 1980). 11. and W.C. Shen, NOAA Term. Rep. NESS #8 (1982). atmospheric flow that impacted weather conditions worldwide. 12. E. R. Westwarer, z. Wang. N. C Grady. L M. McMillin, J. Almond. Oceanic Tech. Warming locally exceeded 2°C in two Pacific anticyclones (Fig. 6) 2. 97 (1985). that straddled an equatorial zone of intense convective activity 13. T. Karl. unpublished data. 14. P.D. Jones - at., J. Climate Appl. Meterol. 25. 161 (1986). caused by the warrn water. This event caused globally averaged 15. P.D. Jones. S. C B. Raper, T. M. L. Wigley, ibid., P. 1213. temperatures to rise more in several months than what is expecred 16. P. D. Jones. J. Climate 1. 654 (1988). 17. 1. Hansen and S. Lebedelf. J. Ceopleys. Res. 92. 13.345 (1987). within several decades if enhanced greenhouse warming is occur- 18. G. Farmer. T. M. L Quigley, P. D. Jones, M. Salmon. Documenting and Explaining ring. Although the 1987 ENSO has been considered weaker than Recent Clobal Mean Temperature Changes (Climaric Research Unit. University of East the 1983 ENSO, it was associated with higher temperatures that Anglia Norwich, 1989). 19. R. 5. Quiroz, Men. Weather Rev. 111. 1685 (1983). were more uniformly spread throughour the tropics. The mid- 20. R. Jenne and D. Joseph provided the MSU data used in this study; R. Hood latirudes in both the Southern and Northern hemispheres in 1988 provided data processing and programming support: N. Grody collaborated on experienced warm conditions that appear to be coupled to the 1987 portions of this research and provided general advice: F. Wenzz provided updated sea surface emissiviry estimates; P. Olsen, F. Solis, and P. Swanson assisted us in tropical warmth. The period 1984 to 1986 was dominated by cooker obtaining technical dara on the MSU: G. Wilson helped obtain the satellite data than normal tropical air. The 10-year time series exhibits bifurcation sets: discussions with J. Dodge and R. McNider led to the present research. in that there are nine cool or warm years, and only one year (1981) 27 Ocrober 1989; accepted 23 February 1990 1562 SCIENCE, VOL. 247 200 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 08:01 16, 9 NNS Reprinted from JOURNAL OF CLIMATE, Vol. 3. No. 10. October 1990 American Meteorological Society Global Atmospheric Temperature Monitoring with Satellite Microwave Measurements: Method and Results 1979-84 ROY W. SPENCER Earth Science and Applications Division. NASA Marshall Space Flight Center. Alabama JOHN R. CHRISTY Johnson Research Center. University of Alabama in Huntsville. Huntsville, Alubama NORMAN C. GRODY NOAA/NESDIS. Washington. D.C. (Manuscript received 21 October 1989. in final form 4 May 1990) ABSTRACT A method for measuring global atmospheric temperature anomalies to a high level of precision from satellites is demonstrated. Global data from the Microwave Sounding Units (MSUs). flying on NOAA satellites since late 1978, have been analyzed to determine the extent to which these data can reveal atmospheric temperature anomalies on bidaily and longer time scales for regional and larger space scales. The global sampling provided by the MSUs is an important asset. with most of the earth sampled bidaily from each of (typically) two instruments flying concurrently on separate satellites at different solar times. The primary source of tropospheric thermal information is from the MSU 53.74 GHz channel. This channel is primarily sensitive to thermal emission from molecular oxygen in the middle troposphere. with relatively little sensitivity to water vapor. the earth's surface. and cloud (especially cirrus) variations. The long-term stability of the oxygen mixing ratio in the atmosphere makes it an ideal tracer for climate monitoring purposes. Lower stratospheric temperature anomalies are derived from the MSU 57.95 GHz channel. Comparisons between monthly MSU temperature anomalies and corresponding thermometer-measured anomalies for the United States reveal a high (0.9) correlation, but hemispheric anomalies show much lower correlations. This results from some combination of poor thermometer sampling of remote regions and weak coupling of surface and deep-tropospheric temperature anomalies in tropical areas. Analysis of data from two of the MSUs (on NOAA-6 and NOAA-7). whose operational periods overlapped by two years. reveals that hemispheric temperature anomalies measured by the separate instruments are very similar (to about 0.01°C) on monthly time scales. Their combined time series of unfiltered two-day hemispheric averages show standard deviations of their mean of 0.15°-0.20°C and standard deviations of their average difference of 0.02*-0.03°C. indicating a signal-to-noise ratio of 40 for the Southern Hernisphere and 45 for the Northern Hemisphere. The intercomparison period also reveals no evidence of calibration drift between satellites at the 0.01°C level. This was substantiated by two 15-month comparisons of NOAA-6 with rawinsonde data from 45 stations in the eastern United States. which revealed 0.013°C net difference over five years. Monthly averaged comparisons between individual rawinsonde and NOAA-6 data from 1980 through 1982 reveal a monthly standard deviation of their difference of 0.04°C. The statistical and geophysical portions of this noise are found to be about equal in magnitude. 0.03°C. The single-satellite noise due to imperfect sampling for ten- day. 2.5" gridpoint temperatures was calculated by measuring the standard deviation of the difference between two satellites with ranges from 0.2°C in the tropics to 0.4°C in middle latitudes. The period of analysis (1979-84) reveals that Northern and Southern hemispheric tropospheric temperature anomálies (from the six-year mean) are positively correlated on multiseasonal time scales but negatively correlated on shorter time scales. The 1983 ENSO dominates the record, with early 1983 zonally averaged tropical tem- peratures up to 0.6°C warmer than the average of the remaining years. These natural variations are much larger than that expected of greenhouse enhancements. and so it is likely that a considerably longer period of satellite record must accumulate for any longer-term trends to be revealed. 1. Background and theory in the last several years. The possibilities of rising ocean levels and significant regional changes in climate due a. The problem to anthropogenic greenhouse enhancements have cap- Changes in the global climate system have received tured widespread attention. Unfortunately, the poten- high level of scientific, political, and public visibility tially inadequate geographical distribution of ther- mometers has resulted in much uncertainty (and thus controversy) about whether temperature anomalies on Corresponding author address: Dr. Roy W. Spencer. ES43. NASA a hemispheric basis can even be confidently inferred Marshall Space Flight Ctr.. Huntsville. AL 35812. from conventional data. Sparsely populated regions are 800 PAGE NOSTIM S 9 01 ESAD)35411 FROM 10:31 16. 9 NNS 1112 JOURNAL OF CLIMATE VOLUME 3 either poorly measured or not measured at all. Changes radiometer output is usually converted to a "brightness in thermometer exposure due to urbanization are temperature" (Tb). The term "brightness temperature" known to result in apparent warming of well populated acknowledges that the measurement is based upon ra- locations (Karl et al. 1988; Balling and Idso 1989). diative brightness that equals a thermometric temper- Unfortunately, correction for this urbanization effect ature only when the measured object is radiating as a is difficult since the majority of land station locations black body (unit emissivity. and thus zero reflectivity). have likely experienced some sort of local increase in While Th is measured directly by the satellite, radia- exposure to man-made structures. Of course, most tive transfer theory can be used to evaluate the different oceanic areas go unmeasured. Thus, we are faced with radiation emission and scattering processes which lead the task of understanding what portion of climate to the measured Ть. Conceptually. the transfer of ra- change is to be attributed to man versus "natural" diation leading to the satellite measurement at the MSU variations, when we cannot even quantify with confi- frequencies involves three "sources": 1) a dominating dence what the background natural variability is. direct (upward) thermal emission by the atmosphere: and smaller contributions from 2) an "indirect" b. Satellites to monitor climate? (downward) atmospheric emission reflected by the In contrast to conventional measurements, satellites surface back up to the satellite. and 3) a surface emis- can provide the global coverage that is needed to mon- sion shining through the (mostly opaque) atmosphere. itor the earth's atmosphere. Unfortunately. the issue Based upon the theory for nonscattering atmospheres. of satellite instrument calibration has typically been a the Tₕ at a given frequency, Tb(v). depends on the source of great concern and uncertainty. However, vertically integrated atmospheric temperature. T(p), possibly the best calibrated instruments in earth orbit between the satellite and the earth's surface, viz., to date have direct application to the global atmo- spheric temperature monitoring issue. These are the Tb(v) = Microwave Sounding Units (MSUs), built by the Jet Propulsion Laboratory (JPL) for the National Ocean- (1) ographic and Atmospheric Administration's (NOAA) operational weather monitoring needs. Because these The integral is written in pressure coordinates. p, where instruments measure a vertically averaged atmospheric Px is the surface pressure and T, is the surface skin temperature, we feel that they have the potential for temperature. For conciseness. an "effective transmit- significantly augmenting the surface-based thermom- tance function" is defined which is virtually indepen- eter record by providing a measurement representing dent of temperature a significant depth of the troposphere, rather than just a thin near-surface layer sensitive to variable surface effects (urbanization. desertification, etc.). This paper (2) provides detailed technical aspects of the MSUs utility as a climate monitoring device. A related paper (Spen- where F.(p,) = represents the surface emis- cer and Christy 1990) summarizes the global temper- sion term in Eq. 1. es is the surface emissivity, and 8 ature anomalies observed during 1979-88. is the Earth incidence angle. Equation 2 contains the atmospheric transmittance function. T.(p), which is the exponential absorption C. Theory along the vertical path between the satellite and an The MSUs are designed to measure the thermal arbitrary pressure level. As a satellite instrument scans emission by molecular oxygen in the atmosphere at away from nadir (vertical). the increased absorption different spectral intervals in the oxygen absorption due to longer path lengths is accounted for by the sec(0) complex near 60 GHz (Meeks and Lilley 1963). Be- terms. Also, the contribution due to the reflected ra- cause the oxygen abundance in the atmosphere is very diation by the surface is contained in the term having stable in both space and time (Warnek 1988) it makes the factor (1 e,). As discussed below: a compensating an ideal tracer for radiometric atmospheric temperature surface effect is contained in the surface emission term. monitoring. As Machta and Hughes (1970) state, "all in (1). reliable oxygen data since 1910 fall in the range of In (1). Tb is expressed as a vertically (pressure) 20.945%-20.952% by volume." with the instrumental weighted atmospheric temperature. where the weight- accuracy of in situ measurements being +0.006% by ing function is defined as-drip)/ding (see Fig. 1). volume. In contrast, infrared temperature monitoring In typical applications of the four channels of MSU methods depend upon thermal emission from CO2, data. (1) is inverted to retrieve the atmospheric tem- the mixing ratio of which is much more variable in perature profile T(p). We depart from this approach space and time. by interpreting the MSU channel 2 and 4 T1, for more At microwave frequencies. radiance is directly pro- nearly what they are, a vertically averaged temperature portional to the temperature of the emitting body. The measurement of the atmosphere. However. to accu- 600'3968 NOSTIM S 9 01 ESAD)35411 FROM 23:32 16. 9 NNS OCTOBER 1990 SPENCER, CHRISTY AND GRODY 1113 1 ellites (Staelin et al. 1973, Waters et al. 1975). The MSUs are four-channel passive microwave Dicke-type Nactir (9 = on radiometers. Our calibration of the MSU output in Scan Limit (8 # 56.5°) terms of Tb is described in the Appendix. A Th mea- 2 surement is made by the MSU an average of every 0.4 S, each having an instrumental noise of 0.3°C. 3 The frequencies of MSU channels 1-4 are 50.30. 53.74, 54.96, and 57.95 GHz, respectively. Measure- ments made close in frequency to the 60 GHz absorp- 5 tion peak (e.g., MSU channel 4 at 57.95 GHz) are influenced mostly by the upper atmosphere since the 7 Pressure (kPa) absorption (and thus emission) by oxygen is so strong CH. 4 10 (57.95 Gitz) that the lower atmosphere cannot be "seen" by that channel (Fig. 1). Measurements made farther in fre- quency from the absorption peak (e.g., MSU channel 2 at 53.74 GHz) experience less atmospheric absorption 20 and can see some of the earth's surface emission and cloud water emission "shining through" the partly opaque atmosphere. Thus, the MSU channel 2 Tb 30 measurement cannot be ascribed totally to oxygen CH. 2 thermal emission, as some contributions from clouds, (53.74 GHz) 50 water vapor, and the surface exist. These effects will be E = 0.5 addressed in more detail later. MSU channel 1 expe- 70 E 1.0 riences only weak oxygen emission and is quite sen- sitive to surface and cloud effects. 100 The small amount of stratospheric influence on the a.o 0.2 0.4 0.6 0.8 1.0 tropospheric channel 2 and the tropospheric influence on the stratospheric channel 4 will be neglected here. Temperature Weighting Function Accurate correction for the overlap of these channels (Fig. 1) requires knowledge of the vertical distribution FIG. 1. Temperature weighting functions for MSU channels 2 of temperature anomalies which is not generally avail- (53.74 GHz) and 4 (57.95 GHz), for the nadir (solid) and extreme able. Therefore, in all following analyses it must be (dashed) scan positions. Also shown is the difference in the weighting functions for ocean (emissivity = 0.5) and land (emissivity - 1.0) remembered that a small portion (few percent) of a surfaces (figure adapted from Grody 1983). stratospheric temperature anomaly will be included in the tropospheric channel, and vice versa. rately represent the weighted temperature, the function The traditional use of the MSU has been daily mon- must have a unique pressure dependence for a given itoring of synoptic tropospheric temperature structure frequency. This is achieved by having channels with over data-sparse regions (especially oceans) for nu- merical weather prediction. Their utility for this pur- the following characteristics: pose has often been less than desired, due primarily to a) the absorption is predominately due to uniformly the broad vertical slice of the atmosphere sampled by mixed gases. such as oxygen, whose absorption in- each MSU channel. This characteristic has made dif- creases uniformly with pressure: ficult the retrieval of satellite "soundings" with ac- b) the absorption due to variable atmospheric con- ceptable vertical resolution through inversion of (1). stituents. such as water vapor and clouds. is very small compared to the uniformly mixed gas; and b. MSU viewing geometry c) the effects due to surface emissivity variations The MSUs are mounted underneath the NOAA TI- are very small compared to the Tn changes due to at- ROS-N series of satellites, which have sun-synchronous mospheric temperature. near-polar orbits at nominal local crossing times of 0730 and 1930 LST for the "morning" satellites and As discussed below, excluding precipitation effects, 0230 and 1430 LST for the "afternoon" satellites. To these properties are characteristic of the three highest date, these satellites include TIROS-N, NOAA-6, frequency MSU channels. NOAA-7, NOAA-8, NOAA-9, NOAA-10 and NOAA- MSU sampling characteristics 11. The period of analysis for this study (1979-1984) will utilize TIROS-N, NOAA-6, and NOAA-7. The a. Channels periods of useful data acquisition from the satellites The MSUs heritage is from previous, experimental we consider during this six-year period are illustrated instruments on NASAs Nimbus series of research sat- in Fig. 2. 010 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 28:01 16. 9 NNS 1114 JOURNAL OF CLIMATE VOLUME 3 1978 1079 1960 1981 1982 1953 1964 1945 OND JFMAMJJASOND JFMAMJJASOND JFMAMJJASOND JEMAMJJASOND JFMAMJJASOND JFMAMJJASOND JFMAMJ "AFTERNOON" "MORMING" BATELLITES SATELLITES NOAA-5 NOAA-8 TIROS-N NOAA-7 NOAA-9 FiG. 2. Periods of useful MSU data acquisition from TIROS-N. NOAA-6, NOAA-7. NOAA-K. and NOAA-9 satellites from late 1978 through early 1985. Each MSU scans across the satellite suborbital track of two day's total number of footprints falling within through nadir (Fig. 3) such that it measures eleven 2.5° grid squares (Fig. 4) shows the orbital swath pat- "footprints" having a 3 dB (half-power) beamwidth of tern and the interswath data void regions. The pattern 7.5°. Half-power beamwidth is defined by the points in Fig. 4 moves eastward on a daily basis due to preces- on the sides of the antenna main lobe where the re- sion of the orbit. at a rate which varies from satellite ceived energy is 50% of that at the lobe center. Due to to satellite. the increasingly oblique viewing off-nadir, combined with the curvature of the earth. footprints off-nadir 3. Nonoxygen effects on the MSU channels have increasing horizontal sizes. The 1-footprint scan lines are repeated every 25.6 seconds providing 170 Small geophysical signals other than thermal emis- km spacing between scan lines. The 7.5° beamwidth sion by molecular oxygen are also present in the MSU from the satellite orbital altitude of approximately 850 channel 2 (and especially channel 1) measurements. km results in an atmospheric footprint horizontal size These include thermal emission by the surface of the (spatial resolution) at nadir of about 110 km (circular), earth. cloud water. water vapor and back-scattering of increasing to about 180 km by 320 km (elliptical) at the upwelling thermal radiation by precipitation-size the scan extremes. The radiation received by the MSU ice. These processes have been generally ignored for is spatially weighted by the antenna beam pattern. MSU channel 2 because the magnitude of their effect which is dominated by the main lobe (about 2.5 times on sounding retrievals is rarely significant. Because this the width of the 3 dB beamwidth), outside of which paper emphasizes tropospheric temperature anomalies, only a small amount of radiation (2%-4%) is received. which are potentially quite small (0.1° or less). we are In stark contrast to the point measurement of a ther- more concerned with the extent to which these other mometer, the horizontal and vertical sampling of each signals can influence the satellite-observed temperature MSU measurement represents approximately 100,000 anomalies. Because our emphasis will be on anomalies km³ of atmosphere. (deviations from an average value for a given location For reasons to be explained below. only the middle and time of year). we are most interested in the vari- seven of the eleven footprints are used in this study. ability of these small contaminations rather than their An example of the resulting geographical distribution absolute magnitudes. Below. we present an analysis of 323.1 im 178.8 km 168.1 km 109.3 km 7.5° 113 km 125.4 km 150.3 tem 201.7 9.47° 47.37 1173.6 km FIG. 3. MSU scan pattern and 3 dB footprint sizes assuming a nominal orbital altitude of 833 km. Angular dimensions of 7.5° and 9.47° refer to the antenna 3 dB beamwidth and along-scan footprint spacing. respectively (figure adapted from Grody 1983). 110 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 10:33 16. 9 NNI OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1115 FIG. 4. Two days of earth coverage from a single MSU (NOAA-7). where 2.5° grid squares are shaded from black (zero coverage) to white (12 or more footprints). he relative contributions to the channel 2 signal by (1979-88) of 10-day temperature anomaly imagery (an vater vapor, surface emissions. and cloud cover example of which will be discussed below) reveals no changes. such effect. 2. Water vapor emissions b. Surface emissions L an attempt to theoretically evaluate the relative Surface emission variations can arise from either ortance of variations in water vapor alone on the temperature changes or emissivity changes. For MSU MSU channel 2 measurements, radiative transfer cal- channel 2. we find that the combined surface contri- rulations were made for the effects of varying water bution to the measured brightness temperature is 8% vapor in various atmospheres using an approximate of the total contribution over land at sea level. increas- form of (1) (Table 1). The MSU channel 2 sensitivities ing somewhat with elevation. This is a small. but not to water vapor changes in Table I are quite small. but negligible effect. We are primarily concerned with how potentially nontrivial for tropical airmasses. An inter- it might vary in magnitude and thus affect our inter- esting feature is that the In measured over ocean in- pretations of atmospheric temperature variability. creases. while land Tb decreases, with increasing vapor Most of the land surface contribution is emissive content. When superimposed upon the average differ- (assuming a typical emissivity of 0.95), and the re- ence between land and ocean Tn (typically about 1°C). maining, much smaller part is reflection of the atmo- these opposing effects would create artificial Tn anom- spheric downwelling radiation back upward. Theoret- aly gradients when a water vapor anomaly extends ically, 1°C of land surface warming without any change across land/ocean boundaries. Thus. we can examine in atmospheric temperature will cause channel 2 Tb to regional Tn anomalies for any unusual land/ocean increase by 0.1°C. Of course. any change in surface transition characteristics that would indicate a water temperature must result in an atmospheric temperature vapor anomaly effect. An examination of ten years change because they are physically connected. The issue is really whether variability in the average surface air temperature coupling for a given time of year is suffi- TABLE I. Model computed sensitivity of MSU channel 2T, (nadir cient to cause a misinterpretation of this variability as view) to a 20% increase in water vapor concentration from the average amounts found in an average polar atmospheric profile. the U.S. an atmospheric temperature anomaly. This coupling Standard Atmosphere, and an average tropical atmospheric profile. might best be put in terms of the typical low-level at- Sea surface temperatures were assumed to be 0°C. 15°C. and 30°C, mospheric lapse rate. On a hemispheric scale, it would for polar. standard. and tropical atmospheres. respectively. Land sur- be difficult to imagine a widespread anomaly in near- face emissivity was assumed to be 0.95. surface lapse rates sufficient to impact the hemispheric Land Ocean temperature anomalies, especially in the ocean-dom- inated Southern Hemisphere. Over certain regions, Polar 0.00° +0.02° however, this might well contribute to noise in our Standard -0.04° +0.02° measurements. For instance, widespread anomalous Tropic -0.09° +0.03° drought in a region might well be associated with PAGE.012 NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 10:34 16. 9 NNS 1116 JOURNAL OF CLIMATE VOLUME 3 steeper than normal lapse rates during solar heating, most, a 0.1°C channel 2 Tb anomaly. This effect is, as which would then cause positive channel 2 T₆ anom- for the water vapor effect, relatively small, but again it alies. Fortunately, this would probably be. to some de- might not be negligible for regional areas experiencing gree, offset by enhanced nocturnal cooling from large SST anomalies. As in the land case, it is the local anomalously clear skies, low water vapor contents, and variability in the average lapse rate between the surface the lower heat capacity of dry ground. and lower atmosphere that is the primary concern, Emissivity anomalies for relatively unvegetated land since an SST anomaly would influence the temperature surfaces are also a potential problem. The major factor of the lower atmosphere. controlling soil emissivity variations is water content, Ocean emissivity can also change through wind-in- primarily standing surface water. Window frequency duced roughening and foam generation (Wentz 1983). measurements from other spaceborne microwave ra- Increasing winds cause increasing emissivity, ranging diometers (the Nimbus 7 Scanning Multichannel Mi- from 0.50 for calm conditions to about 0.56 for 20 m crowave Radiometer, and the Defense Meteorological s⁻¹ wind speeds. Southern Hemisphere mean values Satellite Program's Special Sensor Microwave/Imager) of surface wind speed from European Centre for Me- show that after a particularly intense rain system passes, dium-Range Weather Forecasts (ECMWF) analyses a day or two of reduced surface emissivities by about for January and July 1988 show that no individual 10% are possible. This is equivalent to about 20% of two-day period varies by more than 0.5 m s⁻¹ from all surface area seen by the satellite being temporarily the monthly mean. This wind anomaly would translate covered by standing water (or some smaller fraction into a hemispheric average channel 2 Tb anomaly of of standing water area combined with unusually wet less than 0.005°C. soil area). This decrease in surface emissivity (from 0.95 to 0.85) can cause a channel 2 Tn anomaly of C. Hydrometeors about -0.2°C, which is comparable to the atmospheric temperature variability we are seeking. Fortunately, 1) FROZEN prolonged periods of anomalously wet ground usually cover only small land areas. Those areas which are Most cirrus clouds are essentially transparent at frequently wet are also usually heavily vegetated, and MSU frequencies. This is due to both their small ice additional moisture is not likely to reduce surface particle size compared to the wavelength of the radia- emissivities more since wetted vegetation does not have tion and low ice water contents. This is one character- a lower emissivity than dry vegetation. Nevertheless, istic of microwave observations that led us to consider at this point, we must be wary that any cold anomalies them instead of infrared sounder data. which are sig- in channel 2 T1,, especially if they cover a small geo- nificantly affected by cirrus clouds. graphical area over a short period of time. might be Relatively little is known about the distribution of due to surface wetting. The net effect of land surface microphysical variables within cirrus clouds, but some wetness anomalies on hemispheric channel 2 Tb aircraft measurements do exist. They reveal a wide va- anomalies, due to their limited geographic extent, will riety of ice particle sizes, shapes, and water contents. probably be very small, less than 0.01 °C. Desertifica- The contents usually show a decrease with altitude, as tion might be expected to cause some contamination do the sizes. Ice particles, if sufficiently large and abun- also, probably through slightly reduced surface emis- dant. will cause In depressions (Wilheit et al. 1982: sivities which are observed over desert regions. These Spencer et al. 1983; Spencer et al. 1989). This results are due to the flatter nature of the surface and. like from the nature of the refractive index of ice at micro- surface wetting. might lead to small negative biases in wave frequencies. The real part of the index of refrac- the Ть. This emissivity effect. however, has not been tion is large enough to cause a small portion of the well documented. upwelling thermal radiation from the atmosphere to Over the ocean, the sensitivity to surface temperature be scattered back downward, and the imaginary part (SST) changes for MSU channel 2 is smaller than for is so small that there is no ice-emitted radiation to take land. Because the ocean emissivity is close to 0.5, the its place. Fortunately. in the 50-60 GHz region the ice ocean emits only half as much energy as the land, and particles must become precipitation size to cause a sig- thus, also reflects more of the downwelling atmospheric nificant effect. Empirical evidence for this transparency radiation back up toward the satellite. In addition, is that the higher spatial resolution (15 km) data from ocean emissivity is temperature dependent. with the the more ice sensitive 85.5 GHz channel of the Special Tn effects of increasing SST being partly offset by de- Sensor Microwave/Imager (SSM/I) reveal no case of creasing emissivity. The net effect is that for each degree cirrus-induced Tn cooling at the 0.5°C level (Spencer increase in SST. the MSU channel 2 Tn will increase et al. 1989). If. however. the cirrus are thick and the by about 0.036°C for a tropical atmosphere over warm result of deep moist convection. larger Tₙ depressions water, decreasing slightly to about 0.030°C for a polar (in excess of 1°C) do occur. which in every case have atmosphere over cold water. Therefore, we would ex- been linked to evidence of precipitation-size particles. pect a very large SST anomaly (3°C) to result in. at The influence of precipitation-size ice in thunder- NOSTIM S 9 01 FROM 58:01 16. 9 NNS OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1117 storm cores, however, can be considerable. In rare sit- ever, clouds below about the 65 kPa level will present ons. 10°C depressions have been observed in the a warm signal in contrast to the "cold" reflective ocean U channel 2 Tb for midlatitude squall lines (Grody background. Therefore, we would expect anomalies in 1983). This is the largest atmospheric influence which low-cloud cover, as was the case for water vapor, to can contaminate the MSU channel 2 oxygen emission generally have opposite effects over land (colder Tb) signal. Fortunately, the larger and more intense ice versus over ocean (warmer Tb). Deeper, cumuliform precipitation events are infrequent and are easily clouds extending from the 65 kPa to 95 kPa level have screened out due to their isolated nature. However, we essentially zero impact on the Tb, although they do cannot rule out the possibility of an anomaly in the screen out atmospheric emission and replace it with frequency of more subtle convective events that are their own emission of a near-equal magnitude. Higher below our assumed convection screening threshold. As clouds (above about 65 kPa) will have a cooling effect we shall see, though, the major convective anomaly in over both surfaces. The smaller water contents found the eastern Pacific during the 1982-83 El Niño/ in higher altitude clouds (generally below 0.1 g m⁻³) Southern Oscillation (ENSO) event showed a large help to reduce their impact on Tb. Some of the possible warm, not cold, signature. This suggests that any ice quantitative Tb effects from cloud water variations and contamination of the temperature signal was relatively other geophysical factors are summarized in Table 2. small. The cloud values, as previously pointed out, are po- tentially quite variable depending upon assumed cloud 2) LIQUID heights, thickness, cloud water contents and areal cov- erage. Hemispheric statistics on cloud water will not Cloud liquid water effects, to our knowledge, have be affected by cloud variations nearly as much as re- always been too weak to be seen visually in MSU chan- gional statistics, which might be sufficient to influence nel 2 imagery. Radiative transfer calculations suggest, the Tb anomalies. As was the case for water vapor however, that up to a few degrees impact on Tb might anomalies, low cloud anomalies extending across land/ be expected due to cloud water variations. These effects ocean boundaries should appear as an artificial gradient are particularly difficult to address theoretically. This in Tb anomaly along the coast, which would provide is not because the theory is not well understood, but empirical evidence of their existence. ause of the many combinations of cloud depth, wa- content, altitude, and areal coverage which can exist. 4. Method Like water vapor, clouds emit energy at the atmo- sphere's temperature, and their effect is to partly screen a Calibration emission from the warmer underlying atmosphere. Thus, over land (where the surface generally emits at The MSU radiometers produce digital counts as their a brightness temperature higher than any atmospheric antennas sequentially view eleven separate earth 10- level), cloud water would always result in a negative cations, cold space, and a high-emissivity warm target. To anomaly in MSU channel 2. Over the ocean, how- The earth-viewing counts are converted to Tb by in- TABLE 2 Summary of the influence of various geophysical processes on MSU channel 2 T. measurements. Geophysical parameter MSU Channel 2 T. sensitivity (°C) Oxygen emission by the atmosphere +0.92° per +1° (land) +0.96° per +1° (ocean) Land surface temperature +0.08° per +1° Land surface emissivity (wetness) -0.1° per -0.1 emissivity change Sea surface temperature +0.036° per +1° Sea surface emissivity +0.01° per 1 ms⁻¹ incr. in wind speed Water vapor (land) 0.00°, -0.04°, and -0.09° per +20% in vapor in polar. standard. tropic atmos. Water vapor (ocean) +0.02°, +0.02*, and +0.03* per +20% in vapor in polar, standard, tropic atmos. Cirrus clouds From near 0' in thin cirrus to in excess of -1° for thunderstorm anvil cirrus near the storm core Precipitation size ice in excess of 10* for the cores of large and intense storms Cloud water Doubling of areal coverage or of typical cloud water contents over: Land Ocean Low clouds: +0.05° +0.1° to 0.2° (90-95 kPa) -0.3° -0.1* Middle clouds: (60-70 kPa) PAGE 014 NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 98:01 16. 9 NNS 1118 JOURNAL OF CLIMATE VOLUME 3 terpolation between the cold space view (for our pur- These data are then summed and averaged at two- poses spatially isotropic at 2.7 and the warm target day intervals. Most of these two-day averages are made view, whose temperature is monitored by redundant up of approximately 40 000 footprint locations, which platinum resistance thermometers. The details of this at an average diameter of 120 km each (Fig. 3) cor- calibration are contained in the Appendix. responds to an average of 110% Earth coverage in two days (Fig. 4). Every two-day average has oval gaps (Fig b. Instrumental differences 4) in the subtropics, which are filled in as the coverage pattern precesses eastward at rate of 3° to 6° per day. After calibration, small (several tenths of a degree) A small area within 400 km of the poles is never sam- systematic To differences among instruments are ob- pled by the middle seven footprints. served. This can be partly ascribed to the different crossing times of the morning and afternoon satellites. 5. MSU comparisons with conventional data But even separate satellites in the same orbit have a few tenths of a degree difference between them. This Before discussing the zonally averaged temperature is most likely attributable to antenna sidelobe pattern anomalies, it is useful to demonstrate that the MSU- differences, which result in variations in the amount inferred temperature anomalies are closely related to of energy received from colder atmospheric sources off- those measured by radiosondes and thermometers. nadir. Because of these interinstrumental differences, each instrument's data must be addressed separately. a. Radiosondes Overlap in operational periods between successively Previous investigators have made comparisons be- launched instruments is necessary for adjustment to a tween MSU measurements and Th calculated from ra- common level. We have found that about three months winsonde data using the radiative transfer equation (1). is a minimum requirement for adjustment to within Considering both land and ocean surfaces, the standard 0.01°C, although at least one year is preferable. We deviation of the differences between the satellite mea- have arbitrarily chosen to adjust all satellites to the sured and rawinsonde-calculated Tb averages 0.7°C NOAA-6 level. (Grody 1980; Grody and Shen 1982; Grody 1983; Westwater et al. 1985). This error is found to be in- c. Data screening dependent of surface type. It has contributions from Not all MSU channel 2 data are included in our the MSU instrumental noise (0.3°C), errors in the analysis. As the MSU scans off-nadir, the channel 2 Tr baseline calibration of the radiosonde (0.5°), the dif- become colder, causing the well-known "limb-dark- ference in spatial coverage between an MSU footprint ening" effect. This is primarily due to the increasingly versus a radiosonde and space and time offsets in the oblique atmospheric path viewed by the instrument, satellite/radiosonde matchups. which effectively raises the weighting function (Fig. 1) We have compared the NOAA-6 MSU channel 2 to a higher (colder) level in the troposphere. This cool- Th to Tb calculations from radiosonde data from 45 ing increases nonlinearly with scan angle from nadir stations over the eastern two-thirds of the continental toward both scan extremes. Because spacecraft attitude United States. We assumed a constant surface emis- roll errors of one degree or more might be expected at sivity of 0.95 and a ground surface temperature equal certain times, combined with the nonlinear nature of to the surface air temperature for all calculations, and the limb darkening, the attitude effect could translate then removed any station-dependent biases as de- into small negative biases in the scan-averaged Tb if all scribed below. The average of the middle three MSU eleven footprint locations were used. Thus, we have footprints Tb were compared to the radiosonde-cal- chosen to utilize only the middle 7 footprints of the culated Tb (from Eq. 1) when the MSU measurements 11-footprint scan in this study. were made within 3.0 h and 200 km of the radiosonde Each of the middle seven channel 2 Tb is checked release time and location, respectively. The standard to make sure it does not fall out of absolute bounds deviation of the differences between individual MSU- (190 K and 290 K). Then, each is checked against an measured and radiosonde-calculated Th's were found average of its two immediate neighbors within its scan to range between 0.5°C and 1.0°C. consistent with the line to make sure that it does not deviate by more than aforementioned 0.7°C reported by other investigators. 1.5°C from the average difference observed from limb When these results are averaged on monthly time darkening. If any Tb does, then the entire scan line is scales. it is found that the average bias between the rejected from further processing This procedure satellite and radiosonde measurements is radiosonde screens out deep convection and some large snowcov- station dependent. Furthermore, this bias has a seasonal ered mountain areas which also cause volume scatter- modulation of up to 1.0°C depending on the radio- ing-induced Tb depressions. If the data pass these tests, sonde station, with values of (MSU Th radiosonde then the 2.5° latitude band that each footprint falls Tn) falling in the summer and rising in the winter. into is credited with the average Th from the middle While the station location dependence can be explained seven footprints. by geographical variations in emissivity due to the PAGE. NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 98:01 16. 9 NNS OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1119 presence of lakes, rivers, vegetation. etc.: the reason for eastern United States. such as interannual variations = seasonal modulation at many of the stations is not in surface emissivity, precipitation contamination, etc., L clear. To isolate the monthly and annual precision if the satellite measurements were made often enough with which the MSU can measure tropospheric tem- to measure most of the atmospheric variability. perature anomalies, an average annual cycle of the To search for evidence of long-term drift. we com- monthly satellite radiosonde sums and differences was pared a 15-month period during 1985/86 to the same removed from each of the stations individually, thus subset of months during 1980/81 when the NOAA-6 removing these station-dependent biases. as well as the MSU was operating, a time separation of five years. signal of the average annual temperature cycle. Because The difference in the 45-station average (satellite - ra- each satellite period of record was at least two years, diosonde) between the two periods was 0.013°C, with this will not cause the inadvertent removal of any ev- the later period being warmer. This is approximately idence of interannual drift in the MSU or radiosonde the expected random noise of the estimate (0.012°C), systems. When the resulting monthly anomalies in in- based upon division of the average single-comparison dividual station (satellite - radiosonde) sums and dif- noise of 0.8° by the square root of the total number of ferences are averaged together. we obtain monthly time independent matchups (about 4500). Thus, the com- series (Fig. 5) which provide a measure of the ability parison reveals no evidence of drift over the five-year of the MSU to observe tropospheric temperature vari- period at the 0.01°C level. As discussed below, this ations. The monthly standard deviation of the NOAA- conclusion is consistent with satellite intercomparisons radiosonde differences, averaged over the 45 radio- with each other. sonde stations. during the three year period 1980-82 is about 0.04°C, while the standard deviation of the b. Surface thermometer comparisons sums is about 0.92°C. indicating a large signal-to-noise ratio. in excess of 500. The level of noise (0.04°C) is Monthly averaged MSU data on a 2.5° grid covering partly due to statistical noise (0.03°C) that is calculated the United States were compared to monthly averaged from dividing the individual comparison standard de- surface air temperature data (Tom Karl, personal viation 0.8°C by the square root of the average number communication) for the period 1979 through 1984. A of monthly comparisons (about 650). The remaining six-year average annual cycle was computed for each rtion (0.03°C) is then the estimated geophysical dataset and then subtracted from the original monthly ise of the monthly temperature anomalies over the values to obtain 72-month time series of temperature anomalies. The correlation between the MSU and thermometer U.S. averaged anomaly time series is 0.90. The surface air temperature anomalies average about +2.0 twice the magnitude of the MSU channel 2 anomalies. MSU & RAOB This level of agreement was compared to radiosonde- Sure only calculations, where monthly surface temperature Difference anomalies were correlated with calculated channel 2 +1.0 Ть, both from the radiosonde data. The correlations have a geographic dependence over the United States. Monthly Temp. Anomaly (°C) with values of 0.7 over the extreme eastern and south- eastern U.S. increasing north and westward to over +0.2 0.90 in the High Plains and Rocky Mountains, then G.O decreasing again to 0.70 along the West Coast. Thus. -0.2 it appears that the MSU versus thermometer correla- tion of 0.90 is about as high as can be expected for surface/deep troposphere comparisons. -1.0 The agreement between the monthly MSU and thermometer-based calculations diminishes when the above analysis is expanded in area to the Northern and Southern hemispheres. These results, based upon -2.0 comparisons with calculations by Jones (1988) and 1980 1981 1982 Jones et al. (1986a,b), are summarized in Table 3. YEAR Whereas 81% of the variance of the monthly MSU FIG. 5. Monthly anomalies in 53.74 GHz (MSU Tb + radiosonde anomalies can be explained by surface air temperature culated Tn) and (MSU T.- radiosonde calculated T.) during measurements over the United States, only 26% and 1>80-82. A small annual cycle in these quantities was removed from 19% can be explained by Northern and Southern each of 45 United States radiosonde stations separately, and the re- hemispheric thermometer-based calculations, respec- sulting station anomalies were then averaged together. These time series represent signal to noise ratios of 480. 780. and 430 for the tively. To investigate the reason for this poor agree- years 1980. 1981. and 1982 respectively. ment, we again performed radiosonde-only calculations 910 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 23:37 16. 9 NNS 1120 JOURNAL OF CLIMATE VOLUME 3 TABLE 3. Monthly and annual explained variances between MSU low pass filtering this cycle to provide a more repre- tropospheric and thermometer-measured surface air temperature sentative (smoother) cycle. Then the original two-day anomalies for United States, Northern and Southern hemispheres, and the globe. Thermometer-based calculations for the United States averages were subtracted from these smoothed cycles are from T. Karl ("K", personal communication): Jones ("J", 1988), to give the 2-day hemispheric anomalies. The hemi- and Jones et al. (1986a and 1986b) for monthly and annual hemi- spheric and global anomalies for NOAA-6 and NOAA- spheric and global anomalies; and Hansen and Lebedeff ("H". 1987) 7 averaged together are shown in Fig. 6, and imme- for annual anomalies. diately below them is the averaged difference in anom- U.S. NH SH Globe alies between the two satellites. Even though these separate satellites had different orbits, it can be seen Monthly 0.81 (K) 0.238 (3) 0.176 (J) 0.287 (J) that both satellites show the same 2-day hemispheric Annual 0.927 (K) temperature anomalies to within about 0.03°C. The 0.700 (3) 0.632 (J) 0.677 (J) 0.548 (H) 0.348 (H) 0.464 (H) signal to noise represented in Fig. 6 (defined as the square of the standard deviation of the mean of the satellites divided by their average difference) is 45 for of the correlation between surface and deep-layer the Northern Hemisphere, 40 for the Southern Hemi- anomalies, this time for several radiosonde stations sphere, and 60 for the entire globe. Thus, these two around the world. The results show that the best cou- satellites measure essentially the same 2-day hemi- pling (highest correlation) between the surface and spheric anomalies. The ragged character of the time deep troposphere occurs over middle latitude conti- series of 2-day averages is intriguing and shows that nental areas such as the U.S. This is presumably due jumps and dips in globally averaged temperatures of to the large signal of monthly anomalies that is asso- 0.5°C can occur in less than two weeks. Because the ciated with the extreme surface net radiation changes satellites' orbits have different precession rates and these continents go through on a seasonal basis. Low substantially different daily geographical coverage, latitude explained variances ranged from about 40% these abrupt changes are not an artifact of possible around the Caribbean, to 25% at Ascension Island in similarities in the way the two satellites insufficiently the subtropical Atlantic Ocean, to 0% to 20% from sample the earth. Hawaii south and westward across the tropical west Also, during this overlap period, neither satellite re- Pacific Ocean. High latitude maritime locations include cord drifts relative to the other to less than 0.01°C. coastal Alaska (50%) and Iceland (50%). Thus, it ap- Assuming that both MSUs are not drifting together in pears that the poor monthly hemispheric agreement exactly the same way (that is doubtful, especially in between the thermometers and the MSU cannot be light of the radiosonde comparisons above) this level attributed to only poor thermometer coverage, but also of stability is sufficient for climate monitoring. to weak thermal coupling over tropical ocean regions and the smaller seasonal variability in these tempera- NOAA-5 YE NOAA-I tures. The tropical ocean explained variances would Mobe probably increase if a lag in the atmosphere's response of several months were included (Reid et al. 1989). +0.8 & 0.0198° s 9-0.1535° The annual comparisons show better agreement than the monthlies, especially for the Jones et al. data. These a.o explained variances peak at 70% for the Northern Hemisphere data of Jones but are as low as 35% for -0.8 N. Hemisphere the Southern Hemisphere calculations of Hansen and +0.2 & 0.0282° a - 0.1894° Lebedeff (1987). These results indicate that annual Northern Hemispheric surface temperature anomalies are reasonably well related to the satellite tropospheric MSU Channel TB Anomaties (C) 6.0 measurements, but the Southern Hemisphere relation- -0.8 5. Hermisphere ships are only weakly correlated. +0.8 P G-0.1921° 6. MSU results 0.0 a Precision of MSU channel 2 Tb variations -0.8 29 June 14 April 1) HEMISPHERIC SAMPLING 1981 1983 Of special interest is a near 2-year overlap period Fic. 6. Intercomparison between global NOAA-6 and NOAA-7 when the NOAA-6 and NOAA-7 satellites simulta- MSU channel T. during the near two-year period (29 June 1981 to 16 April 1983) when both satellites had MSU's operating. Time neously observed the Earth. An annual cycle of tem- series of global and hemispheric satellite averages (sum/2) and dif- perature was calculated for each satellite separately by ferences (difference/2) are shown by the upper and lower curves in averacine the two vears' 2-dav averages together. then each panel, respectively. 210'39AP NOSTIM S 9 01 ESAD)35411 FROM 88:01 16, 9 NNS OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1121 2) ZONAL AVERAGES Whereas the zonal and hemispheric averages are not very sensitive to longitudinal gaps in satellite coverage, We applied a low-pass filter having a half power point gridpoint estimates obviously are. A 10-day averaging of 80 days to the NOAA-6 and NOAA-7 time series period (11-20 September 1982) was arbitrarily chosen during their overlap period and similarly filtered the to examine the level of agreement between the grid- anomalies on a zonal basis in 2.5° latitude bands (Fig. point measurements made by the two satellites. While 7). The zonally averaged anomalies seen by the dif- use of the middle seven footprints resulted in increased ferent satellites are nearly identical in all their features precision (decreased standard deviation of the gridpoint at the 0.1°C level, and the corresponding hemispheric differences between satellites) when compared to only averages are, again, the same to approximately 0.01°C. five footprints, the results degraded for nine or eleven Therefore, the zonally averaged anomalies. too, are footprints (full swath). Since the geographical coverage measured with considerable precision when one deals is obviously better with the wider swath, we assume with the longer time scales. that the degradation is due to the aforementioned sen- sitivity of the end footprint Tb to spacecraft attitude variations. 3. GRIDPOINT AVERAGES An optimized gridpoint assignment scheme was then Of interest to researchers examining regional climate chosen where the seven middle footprints are split into variability is the precision with which an MSU can two half-subswaths (right four and left four) to preserve measure 25° gridpoint temperature anomalies. some of the smaller horizontal scale temperature in- NOAA-7 NOAA-6 0.5° Hemispheric TB Anomaly (C) 0.0° 1982 1983 1982 1983 -0.5° 90° 60* 10 30* Latitude 0° EQI EQ. EQ 30° 50° 1982 1983 1982 1983 90° Year Year FIG. 7. Hemispheric (top) and zonally averaged (bottom) low-pass filtered (half power width - 82 days) NOAA-6 and NOAA-7 MSU channel 2 Tb anomalies during the two-year overlap period of NOAA-6 and NOAA-7. The time series do not agree near the beginning and end of the period because the low-pass filter has no data to use past the end of the period for NOAA-6 or before the period for NOAA-7. 810 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 10:39 16, 9 NNS 1122 JOURNAL OF CLIMATE VOLUME 3 Correlation coefi. formation, while assigning the half-subswath averages 0.9 0.95 1.0 55 to the full half-swath (left six or right six) footprint BOTH positions. The resulting spatial resolution being mapped onto the 2.5° grid is then about 120 km along- 60° track by 600 km cross-track. To account for the de- creasing size of 2.5° gridsquares poleward of the equa- 40° tor, it was assumed that a single footprint assignment 20° would fully influence an equatorial gridsquare, while Lattude cosine-latitude weighted assignments were made to the Eq longitudinally adjacent gridsquares as the satellite pro- gresses poleward. This preserves the areal influence of 20° the data independent of latitude. Figure 8 reveals the 40" latitudinal profile of the standard deviation of the 10- day averaged differences between the two satellites 60° gridpoint Tb measurements. This gives an estimate of the precision with which a single satellite can measure 80°S the geographic variability of 10-day averaged tropo- 55 0.2 0.4 0.6 80 90 100% spheric temperatures. The values range from 0.15°C Standard Explained over the tropics to about 0.45°C at middle latitudes. Devision Variance (C) Also shown in Fig. 8 is the explained variance between the two satellites' measurements, which is usually above FIC. 8. Standard deviation of the difference in 10-day 2.5° gridpoint MSU channel 2 Tb between NOAA-6 and NOAA-7 (left). and the 90%, except within 10° of the equator where it drops inter-satellite explained variance (right), as a function of 10* latitude to about 82%. Thus, even though the precision in terms band. of temperature sensitivity is a maximum in the tropics, MSU Channel 2 Troposphere 260 Hemispheric Means 255 N N @@ (X) N Globe S S 250 S k S N S $ N N S 245 J F M A M J J A S o N D 90° N 90° N 60° 60° 8.0 30° 30° Latitude 0° & EQ. EQ EQ EO EQ & 30° 30° 0.0 60° 50° 90° S 90° S J F M A M J J A S o N D 220 260 K Month TB (K) FIG. 9. Annual cycle in MSU channel 2 To for the globe and Northern and Southern hemispheres (top): and zonally averaged in 2.5° latitude bands (bottom left). Also shown is the year-average latitudinal profile of MSU channel 2 To (bottom right). The year-averaged global. Northern Hemispheric, and Southern Hemispheric In are 249.88 K. 250.67 K, and 610 PAGE NOSTIM S 9 01 ESAD)35411 FROM 00:40 16. 9 NNS OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1123 MSU Channel 2 . Troposphere 0.5* Hemispheric TB Anomaly (°C) 0.0° 1979 1980 1981 |1982 1983 1984 -0.5° Year 90° N 60° 30° Latitudé 0° Et 30° 0 50° 1979 1980 1981 1982 1983 1984 90° S Year FIG. 10. Hemispheric anomalies (top) and zonally averaged anomalies (bottom) about the MSU channel 2 T. annual cycle shown in Fig. 9 for the 6-year period 1979 through 1984. the level of common variability measured by the sep- C. Tropospheric temperature anomalies during 1979- arate satellites decreases there, partly due to the small 1984 signal of tropical temperature variations. These levels When the zonally averaged low-pass filtered annual of precision should be adequate for many uses requiring cycle for each 2.5 degree latitude band and each satellite relatively short term (10-day) estimates of atmospheric is subtracted from the original 2-day averaged time se- temperature variability. ries for each latitude band, we arrive at the anomalies in the zonally averaged tropospheric temperatures (Fig. b. The annual cycle in tropospheric temperatures 10). including the hemispheric averaged time series. We stress that with only six years of data these anom- The combined period of useful operation of the TI- alies might not be representative of anomalies relative ROS-N, NOAA-6, and NOAA-7 MSUs was 17 No- to a longer time series. The zonally averaged values vember 1978-18 February 1985. The zonally averaged have been cosine-latitude weighted to allow visual annual cycle of tropospheric temperature (MSU chan- evaluation of the impact of individual latitude bands nel 2) for this 6+ year period is shown in Fig. 9. The on the hemispheric values. As can be seen, the anom- annual range of temperature is seen to be largest at alies are well correlated between hemispheres for the h latitudes, especially in the Northern Hemisphere multi-seasonal anomalies over 0.1°C. These global ver 16°C in places) and only about 1°C within ten anomalies can usually be traced to the tropics. degrees latitude of the equator. Due to more efficient The strongest anomaly during this time period was conversion of solar input to sensible heating by land the 1982/3 ENSO event. A warm anomaly of around masses in the Northern Hemisphere, its average annual 0.2°C dominated the tropical latitudes during 1983 temperature is about 1.6°C warmer than the Southern with a zonally averaged peak exceeding 0.6°C in early 1983 compared with the six-year average. A videotape 020 PAGE NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 10:40 16. 9 NNS JUN 6 '91 10:41 FROM NASA/MSFC ESAD(ES41) TO G S WILSON PAGE. 021 -2 2 no. 11. MSU channel 2 T, gridpoint anomalies during 21-30 February 1983. The 2.5° gridpoint anomalies have been interpolated to 1.25° for display purposes. 0 I+ + +2 DEG. C OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1125 MSU Channel 4 Lower Stratosphere 215 7 N S 6 & Globe eg 210 1 6 k N Hemispheric Means 205 J F M A M J J A S O N D 90° N 90* N 8.0 60° 50* O.O. 30* 0.0 30* Latitude 0" CO EQ EQ. EO EO EO b 10.0 : 30° 30° 0.0 0.0 60° 60° B.O 90° S 90° S J F M A M J J A S o N D 200 220 K Month TB (K) FIG. 12 Same as Fig. 9, but for the lower stratosphere (MSU channel 4). The year-averaged global, Northern Hemispheric, and Southern Hemispheric Ts are 211.13 K. 211.42 K. and 210.84 K, respectively. of the geographic distribution of these anomalies reveals annual range of these stratospheric temperatures is that the tropical warmth began in the eastern Pacific larger than in the troposphere at high latitudes, but that rapidly peaked in intensity as a warm couplet smaller at low latitudes. A much larger annual cycle is straddling the equator (Fig. 11). This feature then seen near the South Pole than near the North Pole. weakened as the rest of the tropics became unusually warm for several more months. The zonally averaged f. Lower-stratospheric temperature anomalies during anomaly is fairly symmetric in latitude about the 1979-84 equator as are the relatively cold temperatures im- The stratospheric anomalies (Fig. 13) appear mostly mediately following in 1984. uncorrelated with the tropospheric anomalies except d. Noise estimate of the lower stratosphere measure- for a period of marked tropical warmth starting six months before the tropospheric ENSO warm anomaly ments appeared, and similarly, beginning its decline six The quality of the channel 4 sampling was evaluated months before the tropospheric anomaly decline. The in the same manner as the tropospheric measurements channel 4 anomaly during this period is impressive. above. The lower stratosphere (channel 4) signal-to- exceeding a zonally averaged value of 1.2°C centered noise computed from the sums and differences of the on the equator in late 1982 and early 1983. A videotape NOAA-6 and NOAA-7 2-day averages during the 2- of geographic anomaly imagery for this period reveals year overlap period (not shown) is even higher than that before the El Chichon volcanic eruption, warming IT the troposphere (channel 2), with a global signal- of the entire Northern and Southern Hemisphere sub- -noise value of 125. tropics simultaneously began (early 1982) and then expanded into the tropics and intensified in mid-1982. e. The annual cycle in lower-stratospheric temperatures This might have been the combined result of the erup- A mean and anomaly analysis of the MSU channel tion of El Chichon in the spring of 1982 and a tropical 4 data completely analogous to the channel 2 analysis African volcano which erupted in December 1981. The was performed. The resulting annual cycle of lower absorption of solar radiation by the aerosols would heat PAGE.022 NOSTIM S 9 01 ESAD)35411 NASA/MSFC FROM 10:41 16. 9 NNS 1126 JOURNAL OF CLIMATE VOLUME MSU Channel 4 - Lower Stratosphere 1.0° Hemispheric TB Anomaly (C) 0.0° 1979 1980 1981 1982 1983 -1.0° 11984 Year 90° N 60° 30° Latitude 0° EQ EQ 30* 60° " 1979 1980 1981 1982 1983 1984 90° $ Year FIG. 13. Same as Fig. 10, but for the lower stratosphere (MSU channel 4). fluence the measurements at 57.95 GHz. That the temperature anomalies since 1979. Evidence has been warming was probably not a manifestation of the presented which suggests that satellite microwave ENSO is supported by the fact that our analysis of the measurements of molecular oxygen thermal emission 1987 ENSO (not shown) had no associated strato- by an externally calibrated radiometer can be used to spheric warming. monitor tropospheric temperature anomalies on a As was the case for channel 2. the tropical anomalies global basis to a high level of precision. The noise as- are quite symmetric about the equator and contribute sociated with these measurements is 0.03°C for two- to a positive correlation between hemispheric anom- day hemispheric averages and 0.01°C for monthly time alies on longer time scales. However, a marked negative scales. This estimate is based upon a near two-year coupling between shorter time-scale hemispheric overlap period when the NOAA-6 and NOAA-7 MSU's anomalies exists (see the top panel in Fig. 13) and has operated simultaneously but in different earth orbits. time scales from two to six months. This "out of phase" Any long-term drift of one instrument relative to the behavior cannot be traced to the tropics but is instead other during this period of time was essentially un- occurring at high latitudes and might indicate some measurable (below 0.01°C), suggesting a high level of sort of inter-hemispheric exchange of mass or energy stability. Comparisons of the NOAA-6 MSU Ib with in the stratosphere or troposphere. Tb theoretically computed from radiosonde measure- ments between two 15-month periods separated by five 7. Summary and conclusions years also reveals no significant drift (0.013°C). These This study has introduced the use of satellite micro- results, when taken together, suggest that spaceborne wave radiometry for precise monitoring of atmospheric microwave radiometry can be used for relatively precise 028 PAGE NOSTIM S 9 01 ESAD)3541 NASA/MSFC FROM 20:42 16, 9 NNI ** TATAL ** OCTOBER 1990 SPENCER. CHRISTY AND GRODY 1127 monitoring of global tropospheric temperature varia- cold space, and warm target views) for nonlinearity of tions. instrument response: Comparisons between monthly MSU-derived 2) conversion of warm target platinum resistance hemispheric temperature anomalies and those com- thermometer measurements (digital counts) to tem- puted from surface thermometer data show very good peratures: and agreement over the United States but rather poor 3) interpolation between warm target counts and agreement for the hemispheres. especially the Southern cold space counts to earth viewing counts and. Hemisphere. The poor hemispheric agreement cannot thus, Th. be ascribed only to sparse thermometer coverage but is found to be partly attributable to weaker thermal Step 1 was accomplished by invoking the quadratic coupling between the ocean and deep troposphere than count correction equations veloped by the National that which occurs over the U.S. Annual anomalies for Environmental Satellite Data and Information Service the hemispheres show better correlations. although the (NESDIS). which were based upon instrument mea- level of agreement depends upon the data source used surements made by JPL. The correction is small and and may well change as more years are added. of the same sign for each instrument. The general form The magnitude of the hemispheric temperature of the NESDIS correction equation is anomalies (relative to a 6-year mean) were between (A1) 0.1°C and 0.3°C. with time scales of close to one year. where C is the corrected count value, Crow is the raw Zonally averaged data reveal that the anomalies' origins count value. and the different values for a, are con- could usually be traced to the tropics. Abrupt warming tained in Table Al. This correction is applied to all and cooling events of 0.5°C occur in as little as two radiometer digital counts: i.e., the cold space view weeks and are verified by separate satellites. counts. the warm target view counts. and all Earth- Theoretical evaluations of the possible contaminat- view counts. ing influence of cloud water. water vapor. and surface Step 2 involves converting warm target platinum temperature changes on the tropospheric temperature resistance thermometer (PRT) digital counts to resis- signal suggest some possible small effects. although they tances. then converting these resistances to tempera- would be expected to be much more significant over ture. Two redundant circuits in the MSU each monitor egions rather than hemispheres. the output of one thermometer in each of two warm Our analysis suggests that the year-to-year variability calibration targets. Target #1 is seen by MSU channels of global tropospheric temperatures is so large that any I and 2. while Target #2 is seen by channels 3 and 4. enhanced greenhouse warming since 1979 will need to There have been no failures of these circuits to date reach about 0.1°C before it becomes discernible in the out of seven MSUs successfully launched and operated. satellite record. Therefore. we advise the continued use Our target temperatures were computed as an average of satellite radiometer measurements to monitor global of both thermometer outputs in each target, unless climate anomalies past 1985. there was more than 2°C disagreement between the Acknowledgments. We are very grateful to Roy Jenne two PRTs in which case the entire scan line was and Dennis Joseph at the National Center for Atmo- skipped. The conversion from thermometer digital spheric Research who provided the MSU data used in counts to resistance requires the following equation: this study; to Ms. Robbie Hood for processing of the radiosonde data and library research support: to Mr. R = k₀ - Ccal Frank Wentz. Remote Sensing Systems. who provided (A2) updated sea surface emissivity estimates: to Pete Olsen. where R is resistance: ko and k, are 495.6 and 107.8. Fred Soltis. and Dr. Paul Swanson at the Jet Propulsion respectively. and the same for all MSUs: Curper is the Laboratory, who assisted us in obtaining technical data radiometer digital counts (corrected) during target on the MSU; to Dr. Gregory Wilson. who helped obtain viewing; and Ccal ni and Call to are precision reference the satellite datasets: and to Drs. Richard McNider and resistor digital counts in each of the separate redundant James Dodge. discussions with whom led to the present research: and to Dr. Richard Rosen who provided use- TABLE A1. MSU digital count nonfinearity correction coefficients ful advice for improving the manuscript. for channels ? and 4. a, ar as APPENDIX TIROS-N (ch. 2) 26.85 0.939964 0.0000172642 MSU Calibration (ch. 4) 22.2317 0.964907 0.00000965909 NOAA-6 (ch. 2) 22.0408 0.927569 0.0000167910 Calibration of the MSU radiometric output (digital (ch. 4) 36.5099 0.935564 0.000016923 counts) in terms of brightness temperature (Ts) in- NOAA-7 (ch. 2) 26.061 0.9438 0.00001626 (ch. 4) 21.736 0.9598 0.00001124 volves three general steps: NOAA-8 (ch. 2) 18.71 0.9589 0.000012800 1) correction of radiometer digital counts (earth, (ch. 4) 15.14 0.9533 0.0000146528 PAGE.024 NOSTIM S 9 01 FROM NASA/MSFC ESAD(ES41) 10:42 16. 9 NNS Precision and Radiosonde Validation of Satellite Gridpoint Temperature Anomalies, Part I: MSU Channel 2 by Roy W. Spencer ES43 Earth Science and Applications Division NASA Marshall Space Flight Center Alabama 35812 John R. Christy Johnson Research Center University of Alabama in Huntsville Huntsville, Alabama 35899 MANUSCRIPT still Review in submitted to Journal of Climate April 1991 1 ABSTRACT In Part I of this study, monthly 2.5° gridpoint anomalies in TIROS-N Microwave Sounding Unit channel 2 brightness temperatures during 1979-88 are evaluated with multiple satellites and radiosonde data for their climate temperature monitoring capability. The monthly gridpoint agreement between concurrently operating satellites is generally better than 0.1°C in the tropics and better than 0.2°C at higher latitudes. Monthly anomalies in radiosonde channel 2 brightness temperatures com- puted with the radiative transfer equation compare very closely to the MSU measured anomalies in all climate zones, with correla- tions generally from 0.94 to 0.98 and standard errors of 0.15°C in the tropics to 0.30°C at high latitudes. Simplification of these radiative transfer calculations to a static weighting profile applied to the radiosonde temperature profile leads to an average degradation of only 0.02° in the monthly skill. In terms of a more traditionally measured quantity, the MSU channel 2 anomalies match best with either the 100-20 kPa or 100-15 kPa layer anomalies. No significant spurious trends were found in the ten-year satellite dataset. Thus sequentially launched, overlapping passive microwave radiometers provide a useful system for monitoring intraseasonal to interannual climate anomalies. The Appendix includes previously unpublished details of the MSU gridpoint anomaly data set construction. Part II of this study addresses the removal from channel 2 of the temperature influence above the 30 kPa level, a layer which often experiences con- siderably different trends than those measured by radiosondes below 30 kPa. 2 1. Background The potential of externally calibrated microwave radiometers for precision monitoring of climate change was demonstrated by Spencer and Christy (1990, hereafter SC) and Spencer et al. (1990, hereafter SCG) in the context of deep tropospheric and lower stratospheric brightness temperature (Tb) anomalies. Satellite intercomparisons between simultaneously operating Microwave Sounding Units (MSUs) on morning and afternoon TIROS-N satellites revealed agreement to 0.01°C for monthly, globally averaged Tb anomalies, and instrument stability to better than 0.01°C over a two year overlap period. Whereas these results ad- dressed the precision of the satellite measurements on hemis- pheric and global scales, we now extend this evaluation to the 2.5° gridpoint scale, and incorporate radiosondes as an independ- ent tool for verifying their quality. A fundamental question often asked is what do the channel 2 Tb anomalies represent? First let us address the deep-layer na- ture of the measurement. The channel 2 weighting function (Fig. 1) is vertically broad and represents the vertical distribution (in loge coordinates) of the satellite-measured radiation over a layer extending from the surface to above 30 kPa. The bulk of this microwave radiation is thermally emitted by molecular oxygen with an intensity directly proportional to temperature. A much smaller contribution is from water vapor thermal emission. Oxygen is an ideal temperature tracer for climate monitoring be- cause it occupies a large fraction of our atmosphere (20.95% by volume), is uniformly mixed, and its concentration is very stable in time. Therefore, MSU channel 2 measurements are dominated by the vertically weighted air temperature through a deep tropos- pheric layer of air. In typical satellite sounder "retrieval" schemes, this deep layer measurement is combined with other chan- nels having overlapping weighting functions (such as MSU channels 1,3, and 4 in Fig. 1) in order to deconvolve these weighting functions into an "averaging kernel" (see Conrath, 1972, and Grody, 1980) which is considerably sharper than the individual channel weighting functions. However, we rapidly approach a 3 practical limit in our ability to sharpen the averaging kernel because of the finite number of channels, each having a certain amount of measurement noise. Therefore, based upon current theoretical and engineering limits, the satellites simply can not measure the many levels a radiosonde system measures. Thus, even though satellite temperature retrievals often relate the measure- ments to temperature at a specific level in the atmosphere, this assignment is ultimately based upon certain average statistical relationships which occur in the atmosphere which are, by their assumed time-invariant nature, irrelevant to the climate monitor- ing issue. A second issue concerns much of the satellite-measured sig- nal is contaminated by non-oxygen emission. Based upon theory (see SCG) we expect the MSU channel 2 Tb anomalies to have small contaminating influences from interannual variations in precipitation-size ice in deep convection, cloud water, water vapor, and surface emissivity. However, the magnitudes of these contaminants are difficult to quantify without knowledge of in- terannual variations in these variables. Theoretical evidence was presented by SCG to support interpretation of the Tb anomalies as dominated by thermometric temperature to about 0.01°C for globally averaged monthly anomalies, although regional areas might experience contaminating influences exceeding 0.1°c. Therefore, at this point we hypothesize that the channel 2 anomalies are dominated by thermometric temperature variations over a layer represented by the channel 2 weighting function. However, theory alone is insufficient for rigorous evaluation of the satellite temperature monitoring technique. Here we will present results of comparisons between ten years of monthly radiosonde thickness anomalies to the MSU channel 2 anomalies, in atmospheric environments ranging from tropical to polar. If it can be demonstrated that individual gridpoint anomalies in MSU channel 2 Tb closely match anomalies in radiosonde measurements, then we will have increased confidence that the full global 2.5° gridpoint dataset can be successfully utilized for intraseasonal to interannual climate studies. Furthermore, if there is 4 little or no ten year trend in the difference between the radiosonde and satellite time series, then the satellite system will be supported for long-term monitoring (e.g. of potential global warming). 2. Some limits of radiosonde precision for gridpoint climate monitoring Before gridpoint comparisons with radiosonde data are carried out, it is useful to address the capability of a single radiosonde station to measure monthly anomalies in different layers, as well as for long-term trends. a. monthly radiosonde precision Individual radiosonde temperature profiles are generally considered to have about a 0.7°C random error in the absolute calibration of the sonde. If a station makes sixty observations per month, then that station! precision in the measurement of a monthly temperature anomaly is (0.7/(60) 1/2,, or about 0.1°c. The fact that each measured profile might not accurately repre- sent the larger scale environment will produce an additional source of error for climate monitoring which is related to tem- perature variability on the micro- and mesoscale. Let us assume this error component to be about 0.2°C. Assuming these errors to be uncorrelated, we arrive at an RMS error of somewhat over 0.2°C for a single station's ability to measure monthly temperature anomalies. We have chosen two pairs of tropical western Pacific Ocean stations which are closely spaced to see what level of agreement is experienced for monthly temperature anomalies. These are Yap and Koror Islands; and Truk and Ponape Islands. The tropical en- vironments were chosen to avoid significant horizontal gradients in the monthly anomalies. The distances between these station pairs are 450 km and 700 km, respectively. In Figure 2a we have plotted the agreement between these station pairs in terms of correlations and standard errors of estimate for different layers' thicknesses. We see that the agreement between stations is about 0.2°C, which is consistent with our a priori estimate. 5 b. ten year trends When we plot the ten-year trends of the layer temperatures for these station pairs (Fig. 2b) we see disturbingly large dif- ferences of 0.2° to 0.5°C for the period 1979-88. If we also ex- amine two other western Pacific stations, Guam and Majuro, we find somewhat less disagreement and a tendency towards zero trend. Whatever the reasons for such large variability in single station trends, it appears that decadal trends calculated from a single radiosonde station should be viewed with extreme caution. If this station-to-station variability is random in nature, then many-station averages might be more confidently related to real atmospheric temperature trends (as we will see below). In any event, any disagreement we later find in the satellite com- parisons to decadal radiosonde trends at the 0.1° or 0.2° level probably falls in the realm of noise. 3. Satellite gridpoint dataset construction and precision a. limb correction and gridpoint assignment The MSU measures at eleven scan angles as it scans across the satellite subtrack every 9.5° from 47° left to 47° right of nadir (see SCG for more information regarding the MSU sampling geometry). Each scan position measurement was calibrated as a Tb according to the procedure in SCG. After calibration, regression-derived limb correction equations were applied to all ten non-nadir scan positions data to correct for the off-nadir Tb being an average amount colder than the nadir Tb. This relative coldness is due to the off-nadir weighting functions being slightly higher in the troposphere than the nadir weighting func- tion caused by the increased atmospheric path length of the radiation off-nadir. These limb correction equations have the form TbL = aTb - b, (1) 6 where TbL is the limb corrected Tb for an individual MSU. scan position, and the regression coefficients constants "a" and "b" were derived as a function of satellite, 10° latitude band, and scan position. Approximately 6,700 scans were chosen from each of twelve calendar months of the separate satellites for certain years, i.e. TIROS-N (1979), NOAA-6 and NOAA-7 (1982) ; NOAA-9 (1986) ; NOAA-10 and NOAA-11 (1990). For it to be included, each scan had to exhibit little evidence of horizontal (synoptic scale) temperature gradients, with the difference in the Tb from opposite ends of the scan line not exceeding a few degrees. For each satellite, all twelve months of data (80,000 scan lines) were partitioned according to which 10° latitude band each oc- cupied (assigned by the nadir scan position). Both northern and southern latitude bands were combined together to avoid averaging biases arising from a single scan position usually being on the warmer (tropical) or colder (polar) side of the nadir footprint. The resulting limb corrected TbL for the first (#1) and last (#11) scan positions were quite noisy, so they were not included in any further processing of the data. Next, each TbL was binned into the appropriate 2.5° grid square according to the latitudes and longitudes provided with the raw data by NESDIS. This was done on a daily basis for each satellite separately and for ascending and descending satellite passes separately. At the end of each day, a simple (1/distance) interpolation was performed to empty 2.5° grid squares for as- cending and descending grid fields separately with the nearest non-zero grid data in the east and west directions, out to a max- imum distance of fifteen grids. These interpolated ascending and descending TbL fields were then averaged together to provide a single daily TbL field for each satellite. At the end of each month, a time interpolation was performed (+/- 2 days) to any remaining empty grid squares. The daily fields were then averaged together within each month. Further references to the averaged fields will always assume the limb correction procedure was applied to the calibrated data (TbL), so for brevity we will now refer to the data as simply Tb. 7 b. gridpoint anomaly calculation Production of monthly gridpoint Tb anomalies would be a simple procedure if a single satellite had been operating for the full period of the satellite record. This would have involved computing, say, a ten-year average for each of the twelve calen- dar months at every gridpoint and then subtracting this average from individual months within the ten year period. Unfor- tunately, there are two complications which must be taken into account. First, the satellite record is made up of seven satellites (six of which we use, see Fig. A1), all of which have calibration biases relative to each other by up to several tenths of a degree. The second complication is the existence of not one, but two separate orbit times alternating throughout the satellite record. Because MSU channel 2 has a small surface sensitivity to diurnal variations in land surface heating, the gridpoint averages are dependent upon the satellite orbit time (2:30 or 7:30) and the local strength of the diurnal heating cycle, which is a function of geography and time of year. The procedure for the merging of all six satellites' data records into a single time series of anomalies is described in the Appendix. This is the same procedure which led to the global time series reported by SC and SCG, but some details of which have been previously undocumented. C. gridpoint precision The precision of the satellite Tb anomaly measurements at the gridpoint level can be readily demonstrated without the satellite merging procedure described in the Appendix. Because the TIROS-N satellite system occasionally had two satellites simultaneously operating in separate orbits (7:30 and 2:30) for more than one year, we can compare their measurements of interan- nual variability in two successive years. (This, of course, provides an interannual anomaly measurement since any multi-year mean assigned for those two years cancels out in the difference.) Then we can compare the results from the separate satellites to evaluate how well a single satellite can measure monthly grid- 8 point anomalies in Tb. If this procedure is followed for the month of February from 1982 and 1983, we find remarkable agree- ment between the NOAA-6 and NOAA-7 satellites (Fig. 3a and 3b, respectively). For this single month, the gridpoint correlation between satellites in each 2.5° latitude band ranges from 0.95 in the tropics to over 0.99 at the higher latitudes, with the stan- dard deviation of the satellite difference ranging from generally less than 0.1°C in the tropics to generally less than 0.2°C at middle and high latitudes. Note that this level of gridpoint agreement between satellites is as good or better than the ac- curacy provided by adjacent tropical radiosonde stations, as pre- viously discussed. Further, since these are single-satellite precision, the error is reduced even further (by 0.7) when two satellites are operating. If we utilize the full gridpoint dataset (assembled accord- ing to the procedures in the Appendix) we can evaluate monthly gridpoint precision for many more months when the satellite operating periods overlap. A map of the standard deviation of the monthly inter-satellite differences between NOAA-6 and NOAA-7 (Fig. 4) shows the level of monthly anomaly agreement for an overlap period of twenty months. Note that most areas have grid- point Tb agreement to better than 0.1°C, mainly in the tropics, with almost all other non-tropical areas exhibiting better than 0.2°C agreement. Degradation of a few hundredths of a degree is seen over land areas, noise which is probably introduced by the diurnal variations in the surface as seen by the morning and af- ternoon satellites. A few small areas of mountainous terrain (the Andes, Himilayas, Tibetan Plateau, and portions of Greenland and Antarctica) show poorer agreement between satellites due to enhancement of this surface influence as the terrain protrudes up into the channel 2 weighting function. 4. Radiosonde comparisons While the above results demonstrate there is great precision in the ability of a single MSU to precisely monitor monthly grid- point anomalies in channel 2 Tb, it still remains to be 9 demonstrated that these anomalies are dominated by a temperature signal. We will address three kinds of radiosonde comparisons to MSU measured anomalies: 1) channel 2 Tb computed from the radiosonde profiles of temperature and humidity with the radia- tive transfer equation (see Grody, 1983), 2) a much simpler method where an assumed constant weighting function is applied to the radiosonde temperature profile only, and 3) comparisons to traditionally measured thickness anomalies for various standard atmospheric layers. a. radiosonde data processing The channel 2 Tb anomalies from the merged multi-satellite dataset at specific 2.5° gridsquares were compared to radiosonde measurements for individual stations within those grid squares. The radiosonde data utilized for our intercomparisons is a quality controlled dataset provided by NCAR. The stations are United States (U.S.) -controlled and include, in addition to the mainland United States, scattered Pacific, Caribbean basin, and Alaska locations. of approximately 120 stations which have fairly complete ten year records, here we will present the results from six stations representing various climate regimes (Table 1). Also, comparisons to several stations from each of these six regimes are composited together to provide regionally averaged comparisons. In the first type of radiosonde comparison, the radiative transfer equation (Grody, 1983) was used to compute a Tb with the temperature and water vapor information provided by the radiosonde. All radiosonde ascents were included which went above 10 kPa, with the most recent radiosonde data above 2 kPa (usually not more than one or two days old) being inserted if the current profile did not go above 2 kPa. Each sounding was capped by an artificial temperature of 250 K at 0.01 kPa level. These Tb computations throughout the 10-year period were averaged on a monthly basis. The ten-year (1979-88) monthly averages were then subtracted from the individual monthly averages to arrive at a 10-year time series of monthly anomalies. 10 Secondly, a simpler method of computing Tb was used which avoids the computational intensity of the radiative transfer equation. Because the temperature dependence of the oxygen ab- sorption and the water vapor absorption are both small, a static weighting function can be applied to just the temperature profile without risking great error. The channel 2 weighting function for a U.S. Standard Atmosphere (Fig. 1) was applied to the radiosonde temperature profiles to provide a vertically weighted Tb estimate, i.e. Îb = wiTᵢloge(P₁/P₂) (2) wiloge(P₁/P₂) where the ith layer of N radiosonde profile layers has an average temperature (Ti), an average weight (wi) from Fig. 1, and bound- ing pressures (P₁ and P₂). Third and lastly, comparisons were made to all combinations of thickness layers having bases at 100 kPa or 85 kPa, and tops at 70, 50, 40, 30, 25, 20, 15, and 10 kPa. Because virtually all of the correlations were somewhat lower for layers having bases at 85 kPa, only the results for the layers having 100 kPa bases are presented here. b. single station gridpoint correlations The levels of agreement between the satellite measurements and each of the six stations are presented in Figures 5 through 10 in terms of correlation coefficients (R) and standard error of estimate (SE) of the radiosonde temperature anomalies. The first half of these figures (a) shows the time series of comparisons to radiosonde-computed channel 2 Tb, while the second half (b) shows R and SE for various thickness layer temperature anomalies. In Cold Bay, Alaska (Fig. 5a) we find a monthly correlation of 0.98 between the satellite-measured and the radiosonde- calculated Tb, with a corresponding SE of 0.24°C. The yearly anomaly correlation is just under 0.99. Use of the simplified Tb computations with Eq. 2 results in R=0.96 and SE=0.28 C, which is only a slight degradation in performance for the monthly 11 anomalies. The best monthly agreement between the satellite and a traditionally measured thickness layer (Fig. 5b) is with the 100-20 kPa layer (R=0.94) which is not quite as good as with the other measures. Most of the stations we examined (of over 100 in number) indicated the 100-20 kPa layer as having the best agree- ment of all traditionally measured thickness layers. This should not be surprising considering the channel 2 weighting function distribution in Fig. 1. In St. Cloud, Minnesota (Fig. 6a) we find similar results to those of Cold Bay, with a monthly correlation of 0.98 and SE of 0.24°C. Use of Eq. 2 gives only slightly degraded results with R=0.97 and SE=0.27°C. These figures are slightly better than the correlation with the 100-15 kPa layer (R=0.96) in Fig. 6b. The comparisons to radiosonde-computed channel 2 for Oakland, California (Fig. 7a) reveal R=0.95 and SE=0.25°c. Interestingly, the results with Eq. 2 are slightly better than those obtained with the radiative transfer equation, with SE=0.23°c. As with the other stations, these are both slightly better than with the thickness layer of best agreement (100-20 kPa) for which R=0.94 (Fig. 7b). At Lihue, Hawaii (Fig. 8a) the radiative transfer calcula- tions with the radiosonde data show a 0.96 correlation and SE=0.16°. With Eq. 2 these measures become R=0.95 and SE=0.18°C, again revealing the utility of a simple vertically weighted average of the temperature profile with a constant weighting function. Compared to a traditionally measured layer, best agreement is seen in the 100-20 kPa layer, with R=0.95 and SE=0.23°C (Fig. 8b). Note that we are now reaching the limit of agreement (0.20°C) between adjacent tropical radiosonde stations measuring the same layers shown in Fig. 2. In fact, comparison between the Lihue and Hilo, Hawaii radiosonde stations, despite their relatively close proximity to each other, yields poorer agreement (the circles in Fig. 8b) than the satellite shows with Lihue. This casts some doubt on the regional representativeness of isolated radiosonde stations when used to construct "global" temperature anomalies. of some interest are the low correlations 12 between the MSU channel 2 anomalies and the lower tropospheric temperature anomalies at Lihue. This illustrates very little coupling between the lower and middle troposphere in this oceanic region. The agreement with the satellite anomalies at San Juan, Puerto Rico (Fig. 9a) is about the same as was found in Hawaii, with R=0.95 and SE=0.15°c. Using Eq. 2 for the radiosonde com- putations, we once again obtain only slightly poorer performance, with R=0.93 and SE=0.20°c. In the comparisons to thickness anomalies (Fig. 9b), for the 100-25 kPa layer R=0.95 and SE=0.21°C. Again, these levels of agreement approach the preci- sion inherent in the sampling provided by a single radiosonde stations. Guam, in the tropical western Pacific (Fig. 10a) reveals a somewhat lower correlation (R=0.85), but an excellent standard error (SE=0.14°C) This reflects the very weak signal of monthly temperature anomalies in this region of the world. Note that the correlation is no worse than that between the adjacent radiosonde stations in Fig. 2. The constant weighting method (Eq. 2) gives R=0.81 and SE=0.17°C, again only a slight degradation from using the full radiative transfer equation. Considerably worse results are revealed for the 100-15 kPa layer anomalies, where R=0.74 and SE=0.25°C This degradation is the result of low natural inter- layer correlations within the troposphere which makes comparison with the correct vertical weighting profiles very important. C. composite station correlations When satellite-radiosonde comparisons are grouped by region, the levels of agreement improve further (Table 2). We see the largest improvement in correlation for the tropical western Pacific, for which Guam was 0.85, and now a six station composite correlation increases to 0.94, with the SE improving to 0.09°C. These station composites thus reveal that the radiosonde (and satellite) noise at individual gridpoints is reduced when several gridpoints are averaged together. 13 d. trends Table 3 shows the ten year trends for these station com- posites as well as the satellite measured trends. Note that the differences are well within the levels of disagreement seen in Fig. 2b between closely spaced tropical Pacific radiosonde sta- tions. Thus the ten year trends in the multi-satellite record of monthly anomalies, when compared to regionally averaged radiosonde station measurements, show no obvious evidence of drift. Unfortunately, these radiosonde station groupings are probably not sufficient to measure drifts to better than 0.1°c. 5. Conclusions Radiosonde comparisons support the use of satellite passive microwave temperature measurements for precise deep layer tem- perature monitoring, even at the 2.5° gridpoint scale. Inter- satellite comparisons of monthly TIROS-N MSU channel 2 Tb anomalies at the 2.5° gridpoint level reveal remarkable agree- ment, generally better than 0.1°C in the tropics and better than 0.2° at high latitudes. Since these are single-satellite preci- sions, the precision of the gridpoint values with two satellites is improved even further. Comparisons between MSU-observed and radiosonde calculated monthly anomalies in channel 2 Tb reveal correlations generally around 0.95 and standard errors from 0.15°C in the tropics to 0.25°C at high latitudes. This perfor- mance degrades by an average of only 0.02°C if a simple vertical weighting is applied to the radiosonde temperature profiles, in- stead of using radiative transfer calculations. This simple method is recommended for quantitative comparisons to other mul- tiple layer temperature datasets. Compositing of four to fifteen radiosonde stations together in specific regions improves the satellite-radiosonde agreement to 0.10°C in the tropics to about 0.20°C at high latitudes. Thus, the satellite precision ap- proaches that of individual radiosonde stations in their ability to measure monthly temperature anomalies, which we estimate to be 0.2°C from intercomparisons of closely spaced tropical oceanic stations in the Pacific. In terms of traditionally measured 14 layers, the highest satellite correlations exist for the 100-20 kPa layer or the 100-15 kPa layer. These correlations are only slightly poorer than for the channel 2 weighting profile, except over the tropical Pacific where natural inter-layer correlations are low, making the correct weighting of the radiosonde profile very important. We find no evidence of instrumental drift in the ten years of MSU channel 2 measurements (1979-88) to our ability to measure it with multiple-station radiosonde data. Acknowledgments We thank Roy Jenne and Dennis Joseph at NCAR for provision of the raw MSU data sets and radiosonde data. Robbie Hood assisted in the processing of the radiosonde data and radiative transfer modeling. Norm Grody (NOAA/NESDIS) and Shelby Tilford (NASA Headquarters) provided useful advice during the course of this study. 15 APPENDIX Merging of Multi-Satellite MSU Data for Climate Monitoring The TIROS-N series of sun-synchronous polar orbiting satel- lites have provided a continuous monitoring of the Earth since late 1978. Each satellite has a operational lifetime of one to three years. Nominally two satellites are operated at the same time, one in a morning orbit (7:30) and one in an afternoon orbit (2:30), although a few periods have existed when only a single satellite has been available. Launches of new satellites alter- nate between morning and afternoon. If a single satellite had always been operated (and its sen- sors remained sufficiently stable), then computation of monthly brightness temperature anomalies from this single satellite would have been a simple process of averaging all data into monthly bins for some multi-year period sufficient to observe a represen- tative annual cycle (say 1979-88), and then subtracting this average from the individual single-month averages to arrive at a time series of monthly anomalies. Unfortunately, we must deal with multiple satellites with slightly different sensors in dif- ferent orbits. As a consequence, merging of the data from these overlapping satellites into a continuous time series requires the removal of two basic sources of differences between satellites. First is the absolute calibration differences between different MSUs, for which we find a very small latitudinal dependence in our two primary channels of interest (2 and 4). The second source arises from the different solar times of the morning and afternoon orbits. MSU channel 2 has some sensitivity to the sur- face, and so the 2:30 afternoon channel 2 Tb observations over land can be considerably warmer than either the 7:30 a.m. or 7:30 p.m. Tb. This effect becomes stronger as one moves toward the equator and higher elevations, e.g. 5-10°C over the United States, 10-15°C over the Sahara Desert. Therefore, removal of an annual cycle for anomaly computation is not only a function of the different satellite orbits, but also of geographical region. 16 Following are the specific steps performed in the time series construction: We first compute an average annual cycle from the monthly grid fields described in Section 3 for a single overlap period between two concurrently operating satellites. We chose 1982 when there was a full-year overlap between NOAA-6 (morning) and NOAA-7 (afternoon), using NOAA-6 as a base period due to its long (7+ year) lifespan. The morning and afternoon satellite annual cycles were then used as bases from which anomalies for all the other satellites were computed, a simple subtraction of the gridpoint annual cycles the from the monthly averages obtained from the other morning and afternoon satellites. The anomalies at this point are based upon only the 1982 annual cycle of either NOAA-6 or NOAA-7, without any correction for inter-satellite differences. To make the inter-satellite correction, the anomalies from different satellites were compared during all periods of overlap between satellites (Fig. A1) to find the time- and gridpoint- averaged offsets relative to NOAA-6, which were then individually removed from each satellite's anomalies. At this point, we have anomalies which are intercalibrated between satellites, but are still relative to only a single year's annual cycle (1982). Finally, the average of 12 years of each month's anomalies was forced to zero by adding an appropriate offset to each anomaly, thus removing the 12-year mean annual cycle. Table Al lists statistics of the inter-satellite comparisons and adjustments, including the route of adjustment to a common level (NOAA-6) ; the dates of overlap between satellites used in the adjustments; the daily standard deviation of the difference (single satellite "error") between the overlapping satellites globally averaged channel 2 Tb anomalies; the daily signal to noise ratio (the sum of the overlapping satellites daily ch. 2 Tb anomaly divided by their daily difference, squared) ; 30-day statistics; the globally-averaged adjustments necessary to bring that satellite to the NOAA-6 level; and the net trend in the time 17 series of the daily differences between overlapping satellites Tb anomalies. NOAA-8 had many operational problems and was not used in any of our statistics. References Grody, N.C., 1980: Analysis of satellite-based microwave retrievals of temperature and thermal winds: Effects of channel selection and a-priori mean on retrieval accuracy. Remote Sens- ing of the Atmosphere and Oceans, A. Deepak, Ed., Academic Press, 381-410. Grody, N.C., 1983: Severe storm observations with the Microwave Sounding Unit. J. Climate Appl. Meteor., 22, 609-625. Spencer, R.W. and J.R. Christy, 1990: Precise monitoring of global temperature trends from satellites. Science, 247, 1558- 1562. Spencer, R.W., J.R. Christy, and N.C. Grody, 1990: Global atmos- pheric temperature monitoring with satellite microwave measure- ments: Method and results, 1979-84. J. Climate., 3, 1111-1128. 18 List of Figures Figure 1. Weighting functions for MSU channels 1 (50.3 GHz) ; 2 (53.74 GHz) ; 3 (54.96 GHz) ; and 4 (57.95 GHz) for a 22° view angle through a U.S. Standard Atmosphere. Figure 2. Monthly temperature anomaly agreement (a) and decadal trend agreement (b) between closely spaced tropical Pacific radiosonde stations for different thickness layers. Figure 3. Gridpoint MSU channel 2 Tb anomaly differences for February 1983 minus February 1982, as measured separately by NOAA-6 and NOAA-7. Figure 4. The standard deviation of the MSU channel 2 gridpoint anomaly differences (°C) between NOAA-6 and NOAA-7 for a twenty month overlap period, resulting from the satellite intercalibra- tion procedure outlined in the Appendix. Figure 5. Time series (a) of MSU channel 2 anomalies and radiosonde-calculated channel 2 anomalies; and correlation and standard errors (b) with radiosonde thickness temperature anomalies for Cold Bay, Alaska during 1979 through 1988. In (a) the solid line is MSU-measured, the dashed line is radiosonde- measured, and the correlations (R) are for-the monthly and yearly anomalies, respectively. In (b) the solid line shows the best agreement possible based upon MSU channel 2 Tb being computed from the radiosonde data, and the dashed line shows the satellite-measured agreement with the radiosonde thickness tem- perature anomalies. Figure 6. As in Fig. 5, except St. Cloud, Minnesota. Figure 7. As in Fig. 5, except Oakland, California. 19 Figure 8. As in Fig. 5, except Lihue, Hawaii. Also shown in (b) is the level of agreement between Hilo, Hawaii and Lihue, Hawaii. Figure 9. As in Fig. 5, except San Juan, Puerto Rico. Figure 10. As in Fig. 5, except Guam. Figure Al. Periods of operation and overlap between MSUs during 1979-90. Gaps indicate one or more days of less than 85% global data. The authors do not yet have March 1989 data in their dataset. 20 Table 1. Radiosonde stations from various climatic regimes chosen for comparison to the satellite gridpoint temperature anomaly dataset. Station Climatic Type Location Cold Bay, Alaska Polar 55.2°N 162.7°W St. Cloud, Minnesota Continental 45.6°N 94.2°W Oakland, California Mid-latitude 37.7°N 122.2°W Lihue, Hawaii Sub-tropical 22.0°N 159.3°W San Juan, Puerto Rico Sub-tropical 18.4°N 66.0°W Guam Tropical 13.6°N 144.8°E Table 2. Station composite correlations: and standard errors of satellite versus radiosonde monthly anomalies in channel 2. Num- bers in parentheses are the corresponding single-station values from the stations represented in Figs. 5-10. Region R SE Alaska (15) .98 (.97) .17° (.24°) Great Lakes (6) .98 (.98) .22° (.24°) West Coast (4) .97 (.95) .18° (.25°) Caribbean (6) .97 (.95) .10° (.15°) West Pacific(7) .94(.85) .09° (.14°) Alaska: (15) Anette, Yakutat, Kodiak, King Salmon, Cold Bay, Adak, St. Paul Is- land, Anchorage, Fairbanks, McGrath, Bethel, Kotzebue, Nome, Barter Island, Barrow. Great Lakes: (6) Sault Ste. Marie, MI; International Falls, MN; Green Bay, WI; St. Cloud, MN, Huron, SD. U.S. West Coast: (4) Quillayute, WA; Salem, OR; Oakland, CA; San Diego, CA. Tropical West Pacific: (7) Koror, Yap, Majuro, Wake, Guam, Truk, Ponape. Caribbean: (6) Christ Church, Barbados; Santo Domingo, Dominican Republic; Piarco, Trinidad; San Juan, Puerto Rico; Curacao, Netherland Antilles; Roberts, Cayman Islands 21 Table 3. Ten year trends (°c, for the decade 1979-1988) in MSU- observed and radiosonde-computed channel 2 Tb anomalies for the regional station composites listed in Table 2. The number of stations is each group is shown in parentheses. Radiosonde MSU Region Ch. 2 Trend Ch. 2 Trend (Raob-MSU) Alaska (15) -0.67° -0.64° -0.03° Great Lakes (6) +0.33° +0.43 O -0.10° U.S. West Coast (4) -0.03° +0.18° -0.21° Caribbean (6) -0.01° +0.10 O -0.11° West Pacific( (7) +0.14 O +0.08 +0.06° Weighted Average -0.06° 22 Table Al. Global statistics of all satellite overlaps between MSUs for MSU channel 2 data. Satellite Method Overlap Dates #days daily daily 30-day 30-day Global 1-yr error S/N error S/N Adj. trend TIROS-N to NOAA-6 7/1/79- 164 .033° 24.4 .013 o 48.9 +.068° +.074 1/19/80 NOAA-6 BASE NOAA-7 to NOAA-6 6/26/81- 596 .028° 36.7 .006° 577.4 -.002° +.007° 4/16/83 NOAA-9 to NOAA-6 10/30/85- 284 .028° 71.4 .006° 183.0 +.067 +.015 11/4/86 NOAA-10 to NOAA-9 11/26/86- 98 .030° 47.4 .006° 361.7 -.281 -.068° 3/7/87 NOAA-11 to NOAA-10 10/23/88- 641 .030° 29.2 .011° 114.0 +.240 +.004 O 10/31/90 * Due to slightly different crossing times in afternoon satellites, there seems to be a very slight difference in annual cycles between TIROS-N versus NOAA-7. This is also seen in NOAA-7 versus NOAA-11, however, the two year overlap between NOAA-10 and NOAA-11 eliminates any trend due to an annual cycle in the satellite difference time series. PRESSURE (kPa) 100 80 09 50 40 30 20 10 8 9 5 4 3 2 1 0.1 Troposphere Stratosphere Fig 1 NORMALIZED WEIGHT 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ch. 1 Ch. 2 Ch. 3 Ch. 4 I .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 A 100-10 100-10 100-15 100-15 THICKNESS LAYER (kPa) 100-20 100-20 KOROR vs. YAP TRUK vs. PONAPE 100-25 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 H .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) B Ponape Koror Truk Majuro Guam Yap 100-10 100-15 THICKNESS LAYER (kPa) 100-20 100-25 100-30 100-40 100-50 100-70 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 DECADAL TREND (°C) Fig. 2a,b -5 -4 -3 3 -2 -1 I 0 +1 +2 +3 +4 +5 DEG NOAA- - 6 -5 5 -4 -3 3 -2 -1 0 +1 +2 +3 +4 +5 DEG NOAA- 7 Fig. 3 a,b (overlay) YOU 88 BM EQ 0 J Fig. 4 COLD BAY, ALASKA Tb ANOMALY (°C) -2 - -1 O 3 2 1 SE=0.24°(0.28°) -3 79 80 81 82 83 84 85 86 87 88 R = 0.970 0.986 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 100-15 COLD BAY THICKNESS LAYER (kPa) 100-20 best possible 100-20 MSU Ch. 2 100-25 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 5 ST. CLOUD, MINNESOTA 3 2 Tb ANOMALY (°C) 1 0 -1 -2 SE=0.24°(0.27°) -3 - 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.980 0.962 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 100-15 ST. CLOUD THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 H 45678910 0.0 0.5 1.0 CORRELATION STANDARD ERROR (C) Fig. 6 OAKLAND, CALIFORNIA 3 2 Tb ANOMALY (°C) 1 M 0 -1 -2 SE=0.25°(0.23°) -3 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.951 0.841 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 OAKLAND 100-15 THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5.6.7.8.9 10 0.0 0.5 1.0 CORRELATION STANDARD ERROR (C) Fig. 7 LIHUE, HAWAII Tb ANOMALY (°C) -3 -2 -1 O 3 2 1 SE=0.16º(0.18°) 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.955 0.961 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 a LIHUE Lihue vs. Hilo 100-15 THICKNESS LAYER (kPa) 100-20 0 of best possible 100-20 100-25 MSU Ch. 2 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 8 SAN JUAN, PUERTO RICO Tb ANOMALY (°C) -1 3 O 2 2 1 SE=0.15°(0.20°) -3 79 80 81 82 83 84 85 86 87 88 R = 0.950 0.972 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 SAN JUAN 100-15 THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 4.5.67.8 .9 10 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 9 GUAM, PACIFIC ISLAND Tb ANOMALY (°C) -1 3 O 2 1 -2 SE=0.14º(0.17°) -3 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.847 0.835 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 + 100-15 GUAM THICKNESS LAYER (kPa) 100-20 best possible + 100-20 100-25 MSU Ch. 2 # 100-25 100-30 100-30 + TRUK VS. PONAPE 100-40 100-40 KOROR vs. YAP 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 10 TIROSN NOAA- 6 NOAA- 7 NOAA-8 X NOAA- 9 NOAA- 10 NOAA- 11 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 YEAR Fig. A1 Precision and Radiosonde Validation of Satellite Gridpoint Temperature Anomalies, Part II: A Tropospheric Retrieval and Trends during 1979-90 by Roy W. Spencer ES43 Earth Science and Applications Division NASA Marshall Space Flight Center Alabama 35812 John R. Christy Johnson Research Center University of Alabama in Huntsville Huntsville, Alabama 35899 submitted to Journal of Climate April 1991 1 ABSTRACT TIROS-N satellite Microwave Sounding Unit (MSU) channel 2 data from different view angles across the MSU scan swath are combined to remove the influence of the lower stratosphere and much of the upper troposphere on the measured brightness tempera- tures. The retrieval provides a sharper averaging kernel than the raw channel 2 weighting function, with a peak lowered from 50 kPa to 70 kPa and with only slightly more surface influence than raw channel 2. Monthly 2.5° gridpoint anomalies of this tropos- pheric retrieval compared between simultaneously operating satel- lites indicates close agreement. The agreement is not as close as with raw channel 2 anomalies because synoptic scale tempera- ture gradients across the 2000 km swath of the MSU are averaged out in the retrieval procedure, and because the retrieval in- volves the magnification of a small difference between two large numbers. Single gridpoint monthly anomaly correlations between the satellite measurements and the radiosonde calculations range from around 0.95 at high latitudes to below 0.8 in the tropical west Pacific, with standard errors of estimate of 0.16°C at Guam to around 0.50°C at high latitude continental stations. Calcula- tion of radiosonde temperatures with a static weighting function instead of the radiative transfer equation degrades the standard errors by an average of less than 0.04°C. of various standard tropospheric layers, the channel 2 retrieval anomalies correlate best with radiosonde 100-50 or 100-40 kPa thickness anomalies. A comparison between global and hemispheric anomalies computed for raw channel 2 data versus the tropospheric retrieval show a cor- rection in the 1979-1990 time series for the stratospheric warm- ing of 1982-83, which was independently observed by MSU channel 4. This correction leads to a slightly greater tropospheric warming trend in the twelve year time series (1979-1990) for the tropospheric retrieval (+0.048°C/decade) than for channel 2 alone (+0.015° C/decade). 2 1. Background In Part I we showed the excellent agreement between separate TIROS-N satellites when independently measuring monthly anomalies of MSU channel 2 Tb at the 2.5° gridpoint level. The average gridpoint agreement between satellites was better than 0.1°C over the tropical oceans, and better than 0.2°C virtually everywhere else on the Earth. This level of agreement was shown to be at least as good as adjacent tropical oceanic radiosonde stations provide. When the satellite measurements were compared to radiosonde measurements of ten years of monthly anomalies, nearly as good agreement was found with many correlations over 0.95 and standard errors of estimate of 0.15°C in the tropics to 0.25°C at high latitudes. In terms of conventionally measured atmospheric layers, the channel 2 anomalies correlated best with 100-15 kPa thickness anomalies. Decadal trends in both the satellite and radiosonde time series showed no significant trend in their dif ference, indicating stability of MSU channel 2 long-term monitor- ing. However, the possibility of cooling in the lower strato sphere coupled with significant energy being received from above this level by MSU channel 2 (Fig. 1), suggests that MSU channel 2 trends and anomalies might not accurately reflect temperature trends in the lower and middle troposphere. Examination of ten year radiosonde temperature trends for different layers of the atmosphere (Table 1) reveals that different trends can be ex- perienced depending upon the layer in question, even within the troposphere. Most of these regions, with the exception of the tropical Pacific, experienced considerable difference in trends above versus below the 30 kPa level. Note also that these ten- year trends are quite variable from region to region, making the lack of radiosonde data in remote regions a significant source of uncertainty in monitoring "global" temperature trends with radiosonde data. These differences suggest the need to remove as much of the upper troposphere and lower stratosphere as possible from the satellite measurements if maximum utility of the satel- 3 lite is to be made for monitoring a single layer which is ex- pected to behave more uniformly (e.g. the lower and middle troposphere). We can examine the global temperature variability in the lower stratosphere during the period 1979-90 by applying the same procedures outlined in Part I of this study to MSU channel 4. All inter-satellite statistics on monthly agreement and long-term stability of MSU channel 4 (not shown) are even better than those reported for channel 2 in Part I, making channel 4 a superb resource for monitoring global lower stratospheric temperature trends. When we compute the channel 4 temperature anomalies for the period 1979-90 (Fig. 2) we find a relatively intense warm event during 1982-83. This appears to be superimposed upon an average cooling trend, although this cooling appears to have leveled off since 1985. The 1982-83 warming is generally at- tributed to solar heating of volcanic dust injected into the stratosphere by El Chichon volcano in early 1982 and by an equatorial African volcano in December 1981. This warm event happened to occur during the tropospheric EI Nino Southern Oscil lation (ENSO) warm event as seen by MSU channel 2 (also in Fig. 2). Since MSU channel 2 was viewing the tropospheric warmth "through" the stratospheric warmth, to the extent that the MSU channel 2 and 4 weighting functions in Fig. 1 overlap, the chan- nel 2 warming might be partially due to the stratospheric event. We say "might be" because it is not proven that the warming seen by either channel 2 or 4 was contributed to by the overlap region of the weighting functions (although this would seem unlikely). Correction for any stratospheric influence would be increasingly important at high latitudes where the base of the stratosphere is much lower than in the tropics. Fortunately, it is possible to combine MSU channel 2 data from various view angles in such a way as to cancel out any stratospheric contribution, and achieve a vertical weighting dis- tribution which is sharper than a single channel can offer. Here we describe a method for removing most of the stratospheric in- fluence on channel 2 based upon the differencing of MSU channel 2 4 near-nadir and limb measurements through the "averaging kernel" concept. This concept was originally described in the context of atmospheric temperature retrievals by Conrath (1972), where weighting functions can be linearly combined to achieve different vertical distributions of weighting. This is the fundamental basis for any satellite temperature retrieval procedures, and is in effect a deconvolution of overlapping weighting functions. This technique was discussed more recently by Grody (1980). Also, updated tropospheric time series of both channel 2 and the tropospheric retrieval are presented for the period 1979-90. 2. Multi-Angular retrievals of deep-layer temperature a. retrieval philosophy Retrieval of temperature information from the MSU (or other satellite sounding radiometers) is usually performed for specific pressure levels through various retrieval procedures which in ef fect deconvolve overlapping weighting functions into a sharper averaging kernel. The retrieval of specific pressure level in- formation has provided conformity with traditional radiosonde measurements, and with numerical weather prediction models which have been tailored to accept radiosonde data. Unfortunately (as discussed in Part I) even though deconvolution of these weighting functions can achieve an averaging kernel which is sharper than the original weighting functions, it is not possible to compress the depth of the deconvolved layer beyond certain limits imposed by the finite number of channels and the existence of noise in those channels. Thus, a retrieval for a specific pressure level will always use information gathered from a relatively deep at- mospheric layer. These attempts to retrieve a temperature for a specific pressure level exploit average statistical relationships which naturally exist between temperatures in different atmos- pheric layers. Because of their assumed time-invariant nature, these average statistical relationships have little use in climate monitoring. 5 Therefore, we assume at the outset that we wish to retrieve a deep-layer temperature measurement from the satellite (since it is not possible to reduce the thickness of the satellite measured layer beyond certain limits anyway). For climate monitoring this seems preferable since it would be unlikely that different climate trends would exist in adjacent, very thin layers of the troposphere. The individual channel weighting functions can in principle be used by themselves (as we saw in Part I), provided that their vertical weighting distribution is considered useful. Removal of the stratospheric influence on MSU channel 2 might be accomplished through some deconvolution from MSU channel 3, which has information from the layer whose influence we wish to remove (Fig. 1). Unfortunately, our analysis of twelve years (1979-90) of MSU channel 3 data reveals that one or more MSUs show some sizeable calibration drifts in channel 3, evidenced by a trend in their difference (Fig. 3) While this drift is ac tually the difference in drifts between the two sensors, we have seen evidence to suggest that it is the NOAA-6 channel 3 which is drifting, which over the period represented in Fig 3 would amount to 0.3°c. (The step function character of the drift in channel 3 is a result of the removal of a one year annual cycle (1982) which was not detrended. The drift by itself seems to be continuous and linear in time.) This amount of instrumental drift is too large to accept for climate monitoring purposes. Channel 3 drift problems were found for at least one other satel- lite. We have been unsuccessful in quantifying these drifts either through radiosonde comparisons or satellite intercom- parisons. As a consequence of these drift problems, use of MSU channel 3 for any type of long-term climate monitoring should be done with extreme caution. b. the tropospheric retrieval and its global precision Fortunately, it turns out that if the goal is the removal of the stratospheric influence on MSU channel 2, then a combination of channel 2 data from different view angles provides better stratospheric canceling than any combination of channels 2 and 3. This is shown in Fig. 4, where the view angles corresponding to 6 MSU scan positions (#1,2,10,11) and (#3,4,8,9) can be combined with appropriate weighting to yield a new weighting function which removes most of the stratospheric influence, lowers and sharpens the peak of the averaging kernel, and results in only slightly more sensitivity to the surface than even the channel 2 nadir scan position ( # 6 ) provides. (SCG addressed the con- taminating non-thermometric influence of surface processes on MSU channel 2). The use of eight scan positions was chosen to mini- mize measurement retrieval noise, which is greater than raw chan- nel 2 noise because the retrieval involves the magnification of small differences between two large numbers. Separate retrievals for the two halves of the swath were attempted, but showed poor results. This lower tropospheric retrieval (which for brevity we will call channel 2R) wipes out any horizontal temperature gradients across the 2000 km wide MSU swath. This gradient information had been retained in our channel 2 study (Part I) through the limb correction procedure. Thus, we can not expect channel 2R to provide very useful synoptic scale temperature information on short time scales. For our simple linear retrieval, after a ratio of weights is chosen which provides stratospheric canceling of the individual scan angle weighting functions in Fig. 4, the weights can then be adjusted SO that their sum equals 1. This results in the retrieval being a brightness temperature, accurate for the averaging kernel distribution of weighting. The small amount of negative area under the channel 2R averaging kernel in Fig. 4, although probably insignificant in comparison to uncer- tainties in the oxygen absorption theory, would lead to a mini- scule warming in the channel 2R Tb if stratospheric cooling oc- curs, and vice versa. The same procedures for gridpoint assignment, satellite in- tercalibration, and anomaly calculation at both the 2.5° grid- point and zonal levels were used as described for raw channel 2 data in Part I. However, since the full swath of data is re- quired for the retrieval, no limb corrections are applied to the individual MSU scan positions. The channel 2R Tb were assigned 7 to only the middle three MSU scan position locations. From this point onward all daily gridpoint assignment and interpolation to empty gridsquares, monthly averaging, satellite intercalibration, and annual cycle removal follow the same procedures as were described in Part I for channel 2. 3. Gridpoint Channel 2R Validation a. satellite intercomparisons Despite the expected poorer performance of channel 2R for gridpoint comparisons, hemispheric and global averaging of the data again reveals excellent agreement between concurrently operating satellites during overlap periods (Table 2). It can be seen through comparison of this table with Table Al of Part I that the monthly global signal to noise is is not as good for the tropospheric retrieval anomalies as it is for raw channel 2 anomalies. The standard deviation of the difference between satellites for single-satellite monthly global anomalies is often closer to 0.02°C compared to 0.01°C for channel 2 anomalies. This is, however, sufficiently good for the multi-month overlap periods to allow accurate intercalibration between satellites for long-term monitoring on at least a global basis. It should be emphasized that when two satellites are operating our final anomaly fields are an average of the two satellites fields, thus further reducing the noise by 20.5. As we did in Part I for channel 2 anomalies, we present a simple example of the precision of the monthly gridpoint retrieval anomalies by computing a difference between February 1983 and February 1982 channel 2R Tb fields (Fig. 5). Again we see that these monthly anomaly patterns measured separately by NOAA-6 and NOAA-7 are very similar, although the similarity is not quite as striking as the channel 2 anomaly fields. Any dif- ference between the two satellites for channel 2R is almost al- ways one of anomaly magnitude, and not of sign. The retrieval anomalies are larger at high latitudes than are the channel 2 anomalies from Fig. 3 of Part I. This is due to both removal of the stratospheric influence (which is usually negatively corre- 8 lated with tropospheric temperature variations) and to the greater variability of temperature variations at lower tropos- pheric levels. The tropical retrieval anomaly magnitudes are not, however, greater than the channel 2 anomalies. This is due to both the tropical stratosphere being much higher, and the fact that tropical oceanic temperature anomalies are often slightly weaker in the lower troposphere than they are in the middle troposphere. A notable exception to the lower troposphere/stronger anomaly rule at high latitudes is seen in the cold anomaly over Western Europe in Fig. 5. Comparison with the corresponding channel 2 cold pattern (Fig. 2 in Part I) reveals that, if any- thing, the retrieval anomaly is actually weaker than the channel 2 anomaly. The reason for this is seen in an image of the MSU channel 4 (lower stratosphere) difference fields for the February 1983-82 periods (Fig. 5c). Here we find that the channel 2 cold anomaly over Europe was contributed to by a very cold anomaly in the lower stratosphere. The retrieval removed this stratospheric influence, which by itself would have resulted in a weaker chan nel 2R cold anomaly, but combined with the typically stronger cold anomaly in the lower troposphere, the channel 2R anomaly ended up being not much different from the channel 2 anomaly. This gives empirical evidence that stratospheric effects are indeed being removed by the retrieval. b. single-station radiosonde comparisons The monthly gridpoint channel 2R anomalies were compared to radiosonde data with the same three methods used for the channel 2 anomalies in Part I. Radiosonde computations of channel 2R were performed with (1) the radiative transfer equation, and (2) a constant (standard atmosphere) weighting profile (Fig. 4) in combination with Eq. 2 from Part I. A third method involved com- parisons to standard pressure level thicknesses. As in Part I, we computed thickness anomalies for all combinations of layers having bases at either 100 or 85 kPa, and tops ranging from 70 to 10 kPa. Consistent with Part I, we present comparisons between channel 2R and the radiosonde channel 2R and thickness tempera- 9 ture anomalies at Cold Bay, Alaska (Fig. 6) ; St. Cloud, Minnesota (Fig. 7) ; Oakland, California (Fig. 8) ; Lihue, Hawaii (Fig. 9) ; San Juan, Puerto Rico (Fig. 10) ; and Guam (Fig. 11). At Cold Bay (Fig. 6), we find a monthly correlation of 0.94 and SE=0.50°C between the MSU measured anomalies and the radiosonde anomalies calculated with the radiative transfer equa- tion. When the static weighting method is applied to the radiosonde temperature profile, the SE degrades by only 0.02°C to 0.52°C. When compared to the standard layers measured by radiosondes, the layers starting at 100 kPa, rather than 85 kPa, had the best correlations with the MSU. This is to be expected from the weighting profile shown in Fig. 4. For Cold Bay, as well as all other stations, we find that the best correlations to the retrieval exist with either the 100-50 kPa or 100-40 kPa layers. At St. Cloud, Minnesota (Fig. 7) R=0.97 and SE=0.48°C, results which are similar to Cold Bay, Alaska. The static weighting method resulted in only 0.01°C degradation in the stan- dard error. The 100-40 kPa layer had the best correlation of all conventionally measured radiosonde layers, with R=0.95 and SE=0.54°c. At Oakland, California (Fig. 8) R=0.91 and SE=0.54°C. Use of the static weighting function method leads to SE=0.58°C. Highest correlation with a standard layer is the 100-50 kPa layer with R=0.91 and SE=0.70°c. In Lihue, Hawaii (Fig. 9) R=0.89 and SE=0.25°c. Use of the static weighting function yields SE=0.32°C. Highest correlation with a standard radiosonde layer is 0.88 and SE=0.31°c. At San Juan, Puerto Rico (Fig. 10) R=0.84 and SE=0.29°C. Slight degradation to SE=0.33°C is seen when the standard atmos- phere weighting function is applied to the radiosonde temperature profile. Correlation to the 100-40 kPa thickness anomalies is 0.84 also, with SE=0.35°c. The worst correlations are seen for Guam (Fig. 11, R=0.74), but as was the case for the channel 2 comparisons, this is be- cause the monthly anomaly signal is so small that it approaches 10 the noise of the radiosonde and satellite systems. In fact, the correlation with the satellite is about the same as between ad- jacent tropical radiosonde stations for the 100-40 kPa layer (0.76), while the standard error of estimate the satellite provides (0.16°C) is at the limit of the radiosondes ability to measure monthly anomalies anyway, as discussed in Part I. Use of the static weighting function method gives SE=0.20°C. The layer of highest correlation is 100-40 kPa with R=0.66 and SE=0.21°c. C. composite station radiosonde comparisons As was done for channel 2 in Part I, we computed statistics for radiosonde station grouping from several regions (Table 2). We find that these station grouping improve the levels of agree- ment considerably, with standard errors ranging from 0.11° over the tropical west Pacific to around 0.35°C at high latitudes and correlations sometimes exceeding 0.95 at high latitudes. These results suggest that, while individual gridpoint satellite data are good, the full 10,000 gridpoint dataset should allow low- noise estimates of global temperature anomalies even on a monthly time scale. 4. Retrieved Tropospheric Trends 1979-90 The good quality of the channel 2R gridpoint anomalies, com- bined with the small (0.02°C) monthly disagreement between simul- taneously operating satellites (Table 1) suggests that hemis- pheric and global temperature trends in the lower half of the tropospheric can be monitored with this retrieval. The tropos- pheric retrieval temperature anomalies for 1979-90 are shown in Fig. 12 superimposed upon the channel 2 anomalies. The largest difference between channels 2 and 2R is seen in 1982-83, where the stratospheric warm event's influence has now been removed. Other differences can also be seen throughout the time series which, since apparently not stratospheric adjustments, probably indicate variations in globally-averaged tropospheric stability. That most of these differences between channel 2 and 2R are real 11 is supported by their large magnitude (0.1°C - 0.2°C) relative to the monthly level of channel 2R disagreement between simul- taneously operating satellites in Table 1 (0.01° to 0.02°C). of some importance is the finding that the long-term trend of the data is somewhat warmer for the retrieval time series (+0.032°C) than it is for the raw channel 2 time series (+0.015°C). This was a combination of a Northern Hemisphere trend of +0.058.°C, and a Southern Hemisphere trend of -0.026°C. The removal of the stratospheric influence on MSU channel 2 during 1982-83 is probably the major reason for the difference in long-term trends between channel 2 and 2R since it occurred in the first half of the twelve year record. 5. Conclusions A procedure for removal of the stratospheric influence on MSU channel 2 for more useful long-term tropospheric temperature monitoring has been applied to MSU channel 2 data from 1979-90, and tested with radiosonde data during 1979-88. Single gridpoint comparisons with individual radiosonde stations give correlations ranging from 0.97 (SE=0.48°C) at St. Cloud, Minnesota where the monthly tropospheric temperature anomaly signal is strong, to 0.74 (SE=0.16°C) at Guam where the signal is very weak compared to the limiting accuracy of both the radiosonde and satellite systems. Groupings of stations improve the standard errors to 0.11°C over the west Pacific to SE=0.35°C at high latitudes, with correlations ranging from 0.83 to 0.98, respectively. As was the case for the channel 2 results in Part I, the tropical standard errors reach the levels of agreement displayed by adjacent radiosonde stations. Best agreement with standard layers which radiosondes traditionally measure is with either 100-50 or 100-40 kPa layer anomalies. For tropospheric temperature trends, the tropospheric retrieval shows some differences with raw channel 2 data, par- ticularly during 1982-83 when stratospheric warming exaggerated the MSU channel 2 warm signal of the 1982-83 ENSO event. Other differences between channel 2 and 2R variations are probably re- 12 lated to variations in tropospheric stability. The satellite record trends for the retrieval differ from the channel 2 trends by only 0.01°C, with a tropospheric retrieval trend of +0.032°/decade and a channel 2 trend of +0.015°C/decade for the period 1979-90. 6. References Conrath, B.J., 1972: Vertical resolution of temperature profiles obtained from remote radiation measurements. J. Atmos. Sci., 29, 1262-1271. Grody, N.C., 1980: Analysis of satellite-based microwave retrievals of temperature and thermal winds: Effects of channel selection and a-priori mean on retrieval accuracy. Remote Sensing of the Atmosphere and Oceans, A. Deepak, Ed., Academic Press, 381-410. Spencer, R.W. and J.R. Christy, 1990: Precise monitoring of global temperature trends from satellites. Science, 247, 1558-1562. Spencer, R.W., J.R. Christy, and N.C. Grody, 1990: Global atmospheric temperature monitoring with satellite microwave measurements: Method and results, 1979-84. J. Climate., 3, 1111-1128. Acknowledgments We thank Norman Grody (NOAA/NESDIS) for helpful advice during this study. 13 List of Figures Figure 1. MSU channel 2, 3 and 4 weighting functions for a nadir view through a U.S. Standard Atmosphere. Figure 2. MSU channel 4 (stratosphere) and channel 2 (troposphere) temperature anomalies during 1979-90, assembled by the method described in the Appendix of Part I. Anomalies have been low-pass filtered. Figure 3. Time series of globally averaged NOAA-6 and NOAA-7 sums (widely varying line) and differences (slightly varying line) of daily MSU channels 2, 3, and 4 Tb anomalies. Drift in channel 3 is evident in the difference time series. Figure 4. Scan-angle averaged MSU channel 2 weighting functions for scan positions #1,2,10,11, #3,4,8,9, and a retrieval averag- ing kernel (channel 2R) computed through a linear combination of functions the other two weighting functions. Figure 5. Gridpoint MSU channel 2 Tb anomaly differences for February 1983 minus February 1982, as measured separately by (a) NOAA-6, and (b) NOAA-7; and channel 4 anomaly differences measured by NOAA-6. Figure 6. Time series (a) of MSU channel 2R anomalies and radiosonde-calculated channel 2R anomalies; and correlations and standard errors (b) with radiosonde thickness temperature anomalies for Cold Bay, Alaska during 1979 through 1988. In (a) the solid line is MSU-measured, the dashed line is radiosonde- measured, and the correlations (R) are for the monthly and yearly anomalies, respectively. In (b) the solid line shows the best agreement possible based upon MSU channel 2R Tb being computed 14 from the radiosonde data, and the dashed line shows the satellite-measured agreement with the radiosonde thickness tem- perature anomalies. Figure 7. As in Fig. 6, except St. Cloud, Minnesota. Figure 8. As in Fig. 6, except Oakland, California. Figure 9. As in Fig. 6, except Lihue, Hawaii. Figure 10. As in Fig. 6, except San Juan, Puerto Rico. Figure 11. As in Fig. 6, except Guam. Figure 12. MSU channel 2 (thin line) and tropospheric retrieval (thick line) Tb time series for the globe, Northern Hemisphere, and Southern Hemisphere during 1979-90. Series have been low pass filtered. 15 Table 1. Composited radiosonde layer average temperature trends for the ten year period 1979-88 for different regions. Layer Florida 1 NW U.S. 2 NE U.S. 3 Great Caribbean 5 West Lakes 4 Pacific 6 30-15 kPa -0.48° -1.28° -0.53° -0.92° -0.29 +0.14 O 50-30 kPa -0.06° +1.34 O +0.85 O +1.14 O -0.02° +0.10 o 70-50 kPa -0.05° +1.37° +0.80 O +1.31° +0.04 O +0.07° 85-70 kPa 0.00 O +1.37° +0.84 O +1.67° +0.19° +0.25 1 Apalachicola, Tampa, Miami, Key West 2 Salem, Oregon; Quillayute, Washington 3 Caribou, Maine; Chatham, Massachusetts; Buffalo, New York; Al- bany, New York; Portland, Maine 4 Flint, Michigan; Peoria, Illinois; Green Bay, Wisconsin; Inter- national Falls, Minnesota; St. Cloud, Minnesota; Huron, South Dakota 5 Christ Church, Barbados; Santo Domingo, Dominican Republic; Piarco, Trinidad; San Juan, Puerto Rico; Curacao, Netherland An- tilles 6 Yap Island; Koror Island; Ponape Island; Truk Island; Majuro Is- land; Guam; Wake Island 16 Table 2. Global statistics of all overlaps between MSUs for the channel 2 tropospheric temperature retrieval. Satellite route Overlap Dates #days daily daily 30-day 30-day to NOAA-6 error S/N error S/N (1 sat/2 sat) (1 sat/2 sat) TIROS-N to NOAA-6 7/1/79- 156 .093°/.066° 7.0 .024°/.017° 20.4 1/19/80 NOAA-6 BASE NOAA-7 to NOAA-6 6/26/81- 566 .087°/.061° 8.4 .015°/.011° 124.8 4/16/83 NOAA-9 to NOAA-6 10/30/85- 228 .087°/.061° 17.7 .015°/.011° 74.1 11/4/86 NOAA-10 thru NOAA-9 11/26/86- 96 .092°/.065° 11.9 .013°/.009° 300.5 3/7/87 NOAA-11 thru NOAA-10 10/23/88- 678 .088°/.062° 8.3 .029°/.020° 40.7 12/31/90 Table 3. Station composite correlations and standard errors of satellite versus radiosonde monthly anomalies in channel 2R. Numbers in parentheses are the corresponding single-station values from the stations represented in Figs. 5-10. Region R SE Alaska (15) .97 (.94) .31° (.50°) Great Lakes (6) .98 (.97) .37° (.48°) West Coast (4) .95 (.91) .35° (.54°) Caribbean (6) .91 (.84) .16° (.29°) West Pacific (7) .83 (.74) .11° (.16°) Alaska: (15) Anette, Yakutat, Kodiak, King Salmon, Cold Bay, Adak, St. Paul Is- land, Anchorage, Fairbanks, McGrath, Bethel, Kotzebue, Nome, Barter Island, Barrow. Great Lakes: (6) Sault Ste. Marie, MI; International Falls, MN; Green Bay, WI; St. Cloud, MN, Huron, SD. U.S. West Coast: (4) Quillayute, WA; Salem, OR; Oakland, CA; San Diego, CA. Tropical West Pacific: (7) Koror, Yap, Majuro, Wake, Guam, Truk, Ponape. Caribbean: (6) Christ Church, Barbados; Santo Domingo, Dominican Republic; Piarco, Trinidad; San Juan, Puerto Rico; Curacao, Netherland Antilles; Roberts, Cayman Islands 18 PRESSURE (kPa) 100 80 09 50 40 30 20 10 8 9 5 4 3 2 1 0.1 Troposphere Stratosphere Fig. - NORMALIZED WEIGHT 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ch. 2 Ch. 3 Ch. 4 Fig. 2 Year -1.0 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 MSU Ch. 2 O'I + -1.4 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 Global Tb Anomaly (°C) summary MSU Ch. 4 I.F + Daily Global MSU Tb Anomaly NOAA-6 & NOAA-7 Sum/2 & Difference/2 (C) -1.0 +1.0 + -1.0 +1.0 + + 26 June 1981 -1.4 +1.4 DRIFT Fig. 3 16 April 1983 Ch. 2 Ch. 3 Ch. 4 0.1 1 2 PRESSURE (kPa) 3 4 our 5 6 8 10 Stratosphere Ch. 2 (footprints #1,2,10,11) (footprints #3,4,8,9) 20 30 40 50 60 Troposphere Retrieval Averaging Kernel 4(#3,4,8,9) - 3(#1,2,10,11) 80 100 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 NORMALIZED WEIGHT Fig. 4 П -5 -4 -3 3 -2 -1 0 +1 +2 +3 +4 +5 DEG. NOAA-6 -5 5 -4 -3 -2 -1 0 +1 +2 +3 +4 +5 DEG NOAA- 7 Fig. 5 a,b,c (overlay) +5 DEG h+ E+ Z+ I+ 0 -1 Z- E- H. -4 - 5 S - COLD BAY, ALASKA 5 4 3 Tb ANOMALY (°C) 2 1 0 -1 -2 -3 - 4 SE=0.50°(0.52°) -5 - 79 80 81 82 83 84 85 86 87 88 R = 0.940 0.941 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 COLD BAY 100-15 THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2R 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 6 ST. CLOUD, MINNESOTA 5 Tb ANOMALY (°C) - -1 3 O 4 2 3 2 4 1 - SE=0.48º(0.49°) - -5 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.969 0.982 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 ST. CLOUD 100-15 THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2R 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (C) Fig. 7 OAKLAND, CALIFORNIA 5 4 3 Tb ANOMALY (°C) 2 1 O -1 - 2 - 3 - 4 SE=0.54°(0.58°) - -5 79 80 81 82 83 84 85 86 87 88 R = 0.907 0.887 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 100-15 OAKLAND THICKNESS LAYER (kPa) 100-20 best possible 100-20 MSU Ch. 2R 100-25 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 H .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 8 LIHUE, HAWAII Tb ANOMALY (°C) -1 0 2 3 1 -2 SE=0.25°(0.32°) -3 79 80 81 82 83 84 85 86 87 88 R = 0.893 0.945 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 LIHUE 100-15 THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2R 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 9 SAN JUAN, PUERTO RICO Tb ANOMALY (°C) -2 -1 0 3 2 1 SE=0.29°(0.33°) -3 79 80 81 82 83 84 85 86 87 88 YEAR R = 0.844 0.880 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 100-15 SAN JUAN THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2R 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 4.5 .6 .7 .89 10 0.0 0.5 1.0 CORRELATION STANDARD ERROR (C) Fig. 10 GUAM, PACIFIC ISLAND 3 2 Tb ANOMALY (°C) 1 O -1 -2 SE=0.16º(0.20°) -3 79 80 81 82 83 84 85 86 87 88 R = 0.741 0.802 YEAR .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 100-10 100-10 100-15 100-15 GUAM THICKNESS LAYER (kPa) 100-20 best possible 100-20 100-25 MSU Ch. 2R 100-25 100-30 100-30 100-40 100-40 100-50 100-50 100-70 100-70 .4 .5 .6 .7 .8 .9 1.0 0.0 0.5 1.0 CORRELATION STANDARD ERROR (°C) Fig. 11 +1.0 Global Mr muhr home 1979 1980 1981 1982 1983 1984 1985 I 1986 1987 1988 1989 1990 -1.0 +1.0 Northern Hemisphere Tb Anomaly (°C) MVM Mahaqun -1.0 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 +1.0 Southern Hemisphere MMM W~~~ 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 -1.0 Year Fig. 12 JUN 07 '91 17:44 =CDC ASHVL NC Reprinted from Nature, Vol. 347, No. 6289, PP. 169-172, 13th September. 1990 P.7 © Macmilian Magazines Ltd., 1990 Western USSR ssessment of urbanization a effects In time series of 70 surface air temperature over land P. D. Jones*, P. Ya. Groismant, M. Coughlant, N. Plummer W-C. Wang! & T. R. Karill 60 * Climatic Research Unit, School of Environmental Sciences. University of East Anglia, Norwich NR4 7TJ, UK 1 State Hydrological Institute, Leningrad, USSR $ Bureau of Meteorology, Melbourne, Australia 1 Atmospheric Sciences Research Center, State University of New York, 50 + Albany, New York 12205, USA H National Climate Date Center, Asheville, North Carolina 28801, USA RECORDS of hemispheric average temperatures from land regions for the past 100 years provide crucial input to the debate over global warming Despite careful use of the basic station data 40 30 40 60 60 70 80 in some of these compilations of hemispheric there have been suggestions? that a proportion of the 0.5 °C b Eastern Australia warming seen on a century timescale may be related to urbanization influences-local warming caused by the effects of urban develop- + ment. We examine here an extensive set of rural-station tem- perature data for three regions of the world: European parts of 20 the Soviet Union, eastern Australia and eastern China. When combined with similar analyses for the contiguous United States9.10, the results are representative of 20% of the land area of the Northern Hemisphere and 10% of the Southern Hemisphere. The results show that the urbanization influence in two of the most widely used hemispheric data sets¹,²,⁴ is, at most, an order of 30 agnitude less than the warming seen on a century timescale. Significant urbanization effects have been noted at many cities".11. Two factors must be considered, however, when com- paring individual city "heat island' magnitudes with hemispheric 40 + warming trends. First, many of the extreme urban blases that have been quoted are the largest daily occurrences, perhaps happening during still evenings or intense inversions. The effect + on station monthly mean temperatures is sure to be considerably 130 140 150 160 smaller. Second, in any gridded temperature data set, a single affected station is unlikely to have a large influence on the time C series of the nearest grid-point, because this is generally a Eastern China 50 weighted average of between 5 and 20 station records1,2. The most comprehensive attempt to assess the significance of urbanization on large-scale temperature trends has been made for the contiguous United States by Karl et al.¹². In this analysis the Historical Climate Network (HCN)" of 1,219 stations was 40 + used to assign stations to pairs of rural and urban sites. Karl et al. showed that an urbanization influence could be detected in many records, with the urban bias being a nonlinear function of population. To assess the influence of urbanization on regional averages developed from the gridded data sets¹⁻³, the 30 average for the contiguous United States derived from the HCN data set of principally rural stations was compared with the gridded series. The gridded series make use of some of the + + temperature data that have been assembled over the present century from sources such as World Weather Records and Monthly Climatic Data for the World (see ref. 5 for details of 20 90 100 110 120 130 these sources). Some of these records come from large cities and could be affected by urbanization-related trends. Com- FIG. 1 Study regions, showing the rural stations (*) and the grid points (+). parison of the gridded and the HCN series revealed that the a Western USSR: D, Eastern Australia: 6 Eastern China. Details of the rural gridded time series shows a warming of 0.1 °C over the period station networks are included in the text. The grid points are taken from 1901-84°. It can be argued¹⁰, however, that the rate of urbaniz- the data set of refs 1 and 2, which interpolate station data onto & 5° latitude ation growth in the United States is atypical compared with by 10° longitude grid. Station data used in the gridding extend 2.5° of latitude many other parts of the world. We have therefore attempted to and 5° of longitude away from each grid point. For the western USSR, 60 assess the urbanization influence in other regions of the world station records were used to construct the grid-point series. There were using specially developed rural-station temperature series. 25 stations in operation by 1901 and 32 were operating in 1987. For eastern Western Soviet Union. For the western part of the Soviet Union Australia 20 stations were used, 7 of which were operating during the 1930s selected a network of 38 stations from alson in included with Death For China 88 stations MAM - THANK we areas with loog stations, resords (Fig. 1a). The vites instude DOWARKI 3 grid And ENTIRE 2 JUN 07 '91 17:45 =CDC ASHVL. P.8 settlements. The largest populated sites are nine towns with populations of the order of 10,000 people. All nine towns are TABLE 1 Comparison of temperature trends located at least 80 km away from major cities. All the site records Standard Linear trend were assessed for artefacts due to factors such as site moves or deviation (°C over changing methods used to calculate monthly mean temperatures. Series Period ("C) period) At twelve sites the observing station was moved slightly. Com- Western USSR parisons with neighbouring sites were made before and after RUSSR (rural) 1901-1987 0.82 0.38 each change, and where necessary, corrections were made to JUSSR 1901-1987 0.79 0.38 ensure homogeneity of the rural-station record. No corrections VUSSR 1901-1987 0.81 0.31 were deemed necessary for the remaining 26 stations, where no RUSSR (rural) 1930-1987 0.84 -0.21 station moves were reported. JUSSR 1930-1987 0.82 -0.09 Data from the 38 stations were then converted to anomalies VUSSR 1930-1987 0.85 -0.20 from the 1951-75 average and combined using the inverse- Eastern Australia distance weighting gridding-acheme-uscd-in-ref 1: Dy-using-the- RAUS (reamt) 1998-1900 0.01 0.58* same averaging scheme as for the gridded data, any differences JAUS 1930-1988 0.31 0.60* between this average and the gridded average will be the result VAUS 1930-1987 0.34 0.55* of data differences and not due to differing interpolation Eastern Chine methods. The resulting series (RUSSR, Fig. 2a) is an average RCHI (rural) 1954-1983 0.30 0.23 for the European part of the Soviet Union, parts of western JCHI 1954-1983 0.27 0.19 Siberia and Kazakhstan. VCHI 1954-1983 0.37 0.13 Using the gridded data from ref. 1, we developed a regional UCHI 1954-1983 0.32 0.39* time series (JUSSR) for the region using 22 gridpoints (see Fig. Contiguous United States 1a). Similarly, we developed a regional time series (VUSSR) Rural 1901-1984 0.42 0.16 for this region using the hemispheric data set from ref. 4. An Grid*10 1901-1984 0.39 0.31 optimum interpolation procedure was employed by Vinnikov et al* to produce their annual hemispheric average time series, Significant trend at the 5% level. using about one-third the number of stations that were used in ref. 1. In this procedure, interpolation of station data to grid trends for these series shows that the 1930-88 warming is greater points was not undertaken, the published data being available for the minimum temperatures (a 0.81 °C rise) than for the only as annual hemispheric averages. For our study, regional maximum temperatures (a 0.30 °C rise). This result indicates a time series were constructed from the data of ref. 4 using this reduction in the daily temperature range from this region, a same optimum interpolation scheme. The stations used extend feature noted for the United States by Karl et al.14. It is unfortu- beyond the bounds of the rural network, unlike those used in nate that separate maximum and minimum temperature data ref. 1. sets are not more widely available. Many countries calculate The rural-network average was compared with the two gridded mean monthly temperatures using observations from a number data series over two periods, 1901-87 and 1930-87. Intercorrela- of fixed hours per day, without recording the daily maximum tion between the three series over these two periods is exceed. and minimum temperatures. ingly high, with all correlations between 0.98 and 0.99. Com- 2.0 parisons of the standard deviation and linear trend are given in Table 1. The similarity between the statistics for all three series 1.0 over both periods is remarkable. Over the 1930-87 period, a cooling of -0.2 °C in RUSSR is observed. This cooling is about 0.0 0.1 °C smaller in JUSSR, but there are no statistically significant -1.0 differences between the two series. Given that all three data sets make use of different station networks, slight differences between -2.0 Western USSR the time series and their trends are to be expected. Eastern Australia. We assembled a network of 49 stations in the eastern half of Australia, from the states of Queensland, Temperature anormaly 1°C) 0.5 New South Wales, Victoria and the southeastern quarter of 0.0 South Australia (Fig. 1b). All stations are rural or small village sites and all encompass the period 1930-88. The average populs- -0.5 tion of the stations is 5,775; the maximum population is 33,368 Eastern Australia and there were neven lightheuse sites which were assumed to have a negligible population. The series from all stations were reduced to anomalies from the 1951-80 period before being 0.5 combined into a regional series (RAUS, Fig. 2b) using the same inverse-distance gridding scheme. 0.0 A regional time series (JAUS) was developed for this region -0.5 from ref. 2. using grid points (Fig. 1b). We also used the data Eastern China in ref. 4 to derive a regional series (VAUS) for the region, which incorporated between 7 and 27 stations. 1900 1920 1940 1960 1980 2000 The correlation coefficients between the RAUS, JAUS and FIG. 2 Time series of annual temperature anomalles for the three regions: VAUS series over the 1930-88 period are all between 0.95 and a Western USSR (1901-58), b. Eastern Australia (1930-88), a Eastern China 0.96. Comparisons of the standard deviation and linear trend (1954-83). The smooth curves were obtained using a gaussian filter desig- are shown in Table 1. The results indicate that neither JAUS ned to suppress variations on timescales of less than 10 years. The base nor VAUS differ significantly from the rural time series RAUS. periods for the three regions are 1951-75. 1951-80 and 1954-83, respec- For VAUS, the warming over the region is similar to that from tively. For the Chinese region all the rural stations are available for every year. For Eastern Australia, 3% of the annual station values are missing. the rural series. In all three series, warming over the 59-year For the Soviet series, there were only 20 of the 38 stations contributing in period is statistically significant at the 4% level 1004 The rural daea not and for Austintia can be 957112 based on maximum JUN 07 '91 17:46 =CDC ASHVL.NC P.93 tern China. We assembled a network of 42 station pairs of a city or town emphasizes the need to assess the homogeneity 1 and urban sites in the eastern half of China (Fig. 1c). The of individual site records data cover the period 1954-83. The 84 stations were selected In none of the three regions studied here is there any indication from a 260-station temperature set recently compiled under the of significant urban influence in either of the two gridded US Department of Energy and People's Republic of China series1.2,4 relative to the rural series. Earlier work on the con- Academy of Sciences Joint Project on the Greenhouse Effect¹⁵. tiguous United States showed an urban influence of 0.15 °C The stations were selected on the basis of station history: we over the period 1901-84. (The results of this work are summar- chose those with few, if any, changes in instrumentation, location ized in Table 1.) The United States result therefore does seem or observation times. All 84 records were complete for the to be somewhat stypical compared with other industrialized 30-year period. The urban stations were in regions with popula- regions of the world. The results from the United States clearly tions of over 0.5 million, whereas for the rural stations popula- represent an upper limit to the urban influence on hemispheric tions mostly less than 0.1 million (according to 1984 population temperature trends. statistics). From the 42-station rural network, we formed an In total, the three regions and the contiguous United States average (RCHI, Fig. 2c) for eastern China using the inverse- encompass 82 grid points (22 over the western USSR, 14 over weighting gridding scheme. The 42-station urban network eastern Australia, 16 over eastern China and 30 over the con- (UCHI) was averaged in the same way Using the gridded data tiguous United States) on a resolution of 4° latirude hy 10° from ref. 1, a regional time series (JCHI) was developed for the longitude. The three Northern Hemisphere regions in total com- region encompassing 15 grid points (see Fig. 1c). An average prise about 20% of the landmass of the hemisphere, and the series (VCHI) for the region, making use of 32 stations, was eastern Australian region comprises 10% of the landmass in the developed from the data in ref. 4. Southern Hemisphere. Thus there seems to be little urbanization The correlation between the RCHI and JCHI series over the influence in three regions of the world that, when taken together, 1954-83 period is 0.90, and there is a lower correlation of 0.81 are twice the size of the contiguous United States. between RCHI and VCHI. Comparisons of the standard devi- It is unlikely that the remaining unsampled areas of the ation and linear trend are shown in Table 1. Although the results developing countries in tropical climates, or other highly popu- indicate that both gridded series show a slight cooling relative lated parts of Europe, could significantly increase the overall to the Chinese rural network (RCHI) over the 30-year period, urban bias above 0.05 °C during the twentieth century. A bias differences between all three series are not statistically sig- of this order is an order of magnitude smaller than the hemi- nificant. Slight differences are to be expected given that the spheric and global-scale warming trend observed over the last various series make use of different sets of raw station data. The 100 years. The bias will be further halved in hemispheric and warming in UCHI is 0.39 °C, considerably higher than that in global temperature estimates that incorporate marine as well as RCHI. For this region, UCHI is the only series for which land temperatures. rming is statistically significant. The Chinese results may, at We emphasize, however, that our results do not imply that glance, seem somewhat surprising, as 24 of the stations urban warming influences in observations of local temperature M in ref. 1 are among the 42 urban sites. In refs 1 and 4, will remain inconsequential in global averages in the future. average temperatures for these sites generally come from airport Indeed, careful selection, inspection and monitoring for urbaniz- sites, which may be in rural areas. In many parts of the world, ation influences in the climate record will be required. This city-centre observatories were relocated to airport sites during concern could be greatly lessened by an international effort to the 1950s and 1960s. That temperature records from many sites monitor and place observing stations outside urban areas. are a combination of information from various locations within Received 5 April: accepted 24 July 1990. 11. Oke, T. R. (ed.) Proceedings of the Technical Conference: Urban Climatology and its applications 1. Jones. P. D. et at. 1 Clim. appl. Met 25, 101-179 (1986). with special regard to tropical areas WMO-No. 652 (World Meteorological Organization. Geneva. 2. Jones. P. D., Raper, S. c. B. & Wigley, T. M. L. 1 Clim. appl. Met. 26, 1213-1230 (1996). 1984). 3. Hansen. J. & Lebedeff, 8. J. geophys. Res. $2, 13345-13372 (1987). 12. Karl, T. R., Disz. H. F, & Kukla, a. 1 Clim. 1, 1099-1123 (1988). 4. Vinnikov, K. Ya, Graisman, P.Ya. & Lugina, K. M. & Clim. a. 662-677 (1990). 13. Quinian, F. T., Karl, T. R. & Williams. c. N. Jr United States Historical Climatology Network (HCN) 5. Jones, P. D. et al. A girld point surface air temperature date set for the Northern Hemisphere Temperature and Presipitation Date NOP-019 (CDIAC. Oak Ridge, 1987). Tech, Rep. TRO22 (US Department of Energy, Washington, DC, 1985). 14. Karl, T. R., Kukla, d. & Gavin, J. A Clim. appl. Met. 23, 1489-1504 (1984). 6. Jones, P. D., Raper, 8. C. B., Goodess. c. M., Cherry, B. 8. G. & Wigley, T. M.L. A grid point surface 15. Koomenoff, F. et at Bull Am. met Soo. 60, 1301-1308 (1988). air temperature date ast for the Southern Hemiaphere Tech. Rep. TRO27 (Us Department of Energy. Washington DC. 1986). 7. KIKIA G., Gavin, I & KArl, T R 1 Clim nont Met. 26, 1265-1270 (1986). ACKNOWLEDGEMENTS. This work was funded partly by the us Department of Energy. Carbon Dioxide 6. Wood, F. B. Clim. Change 12, 297-312 (1988). Research Division and through NOAA/DOE Interagency Agreement. The dollaboration of the many 9. Kari, T.R. & Jones. P. D. Bull Am. met Soc. TO, 265-270 (1959). Individuals in this work was initiated by a meeting of one of the subgroups of Working Group 1 (WG1) 10. Jones, P. D., Kelly, P. M., Goodess. C. M. & Karl, T. R. & Clim. 2, 285-290 (1989). of the Intergovernmental Panel on Climatic Change (IPCC) in December 1989. JUN 07 '91 17:46 =CDC ASHVL. NC P.10 Global Warming: Evidence for Asymmetric Diurnal Trends T.N REVIEW 11 "SCIENCE" Thomas R. Karl' George Kukla² Vyacheslav N. Razuvayev3 Michael J. Changery¹ Robert G. Quayle' Richard R. Heim, Jr.¹ MAY 1991 1 NOAA/NESDIS/National Climatic Data Center Global Climate Laboratory Federal Building Asheville, NC 28801 USA 2 Lamont/Doherty Geological Observatory of Columbia University Palisades, NY 10964 USA 3 All-Union Research Institute of Hydrometeorological Information 6, Korolev St. Obninsk, Kaluga, 249020, USSR JUN 07 '91 17:47 =CDC ASHVL. NC P.11 Global Warming: Evidence for Asymmetric Diurnal Trends ABSTRACT Analyses of the surface thermometric record over three large countries in the Northern Hemisphere (the United states, the Soviet Union, and the People's Republic of China) indicate that virtually all of the warming which has occurred in these regions over the past four decades can be attributed to an increase of mean daily minimum (nighttime) temperatures. Mean daily maximum (daytime) temperatures display little or no warming. In the USA and the USSR (no data available in China) similar characteristics are also reflected in the trends of extreme seasonal temperatures, e.g., positive trends in extreme minimum temperatures and little or no trend in extreme maximum temperatures. Nighttime warming is particularly strong in the USA and the USSR during the warm season, but it is most pronounced in China during winter. The cause(s) of the asymmetric trends are uncertain and may be directly or indirectly related to greenhouse effects. Evidence suggests that changes in cloud cover likely play a direct role (where increases in cloudiness result in reduced maximum and higher minimum temperatures), but these changes may stem from a variety of causes, e.g. greenhouse forcing, so, emissions, natural climate fluctuations. 2 JUN WV (U1 17:40 -000 1'.1 JUN 07 '91 17:40 =CDC ASHVL.NC P.2 Reprint Series 1 March 1991, Volume 251, PP. 1058-1061 SCIENCE The Greenhouse Effect in Central North America: If Not Now, When? THOMAS R. KARL, RICHARD R. HEIM, JR., AND ROBERT G. QUAYLE Copyright © 1991 by the American Association for the Advancement of Science JUN 07 '91 17:41 =CDC ASHVL NC P.3 adapting. In this report, we use observation- al records and purely statistical methods to review temperature and precipitation trends over the past century and assess the likeli- hood that climate projections will be verified by current climate monitoring techniques if contemporary climate model predictions are basically correct. Areally averaged seasonal mean tempera- ture and precipitation over the IPCC Cen- tral region (Fig. 1A) bounded by the lati- tudes 35° to 50°N and longitudes 80° to 105°W (less a small part of land over the Canadian Provinces) were used to deter- mine past changes (5). This is one of the regions where projections of temperature and precipitation change have been issued (3). The data are derived from the climate division averages of temperature and precip- itation (6). A slightly different region re- ferred to as "Central" (Fig. 1B) was used to examine past changes in the seasonal mean daily maximum and mínimum temperatures (7, 8), for which projections were not made by the IPCC. The seasonal mean daily max- imum and minimum temperatures are area averages over the period 1901 to 1987 The Greenhouse Effect in Central North America: derived from 147 stations (8). We analyzed changes of seasonal mean If Not Now, When? temperature and precipitation over the IPCC Central region using tests of signifi- HOMAS R. KARL, RICHARD R. HEIM, JR., ROBERT G. QUAYLE cance of linear trends, the nonparametric Wilcoxon sign-rank test, and the two-phase Climate models with enhanced greenhouse gas concentrations have projected temper- regression test described by Solow (9). The ature increases of 2° to 4°C, winter precipitation increases of up to 15 percent, and two-phase regression can be used to test for summer precipitation decreases of 5 to 10 percent in the central United States by the a change point in the linear trends. Use of year 2030. An analysis of the climate record over the past 95 years for this region was the two-phase regression helps ensure undertaken in order to evaluate these projections. Results indicate that temperature against undetected nonlinear trends. Similar has increased and precipitation decreased both during winter and summer, and that the tests were also applied for changes of the ratio of winter-to-summer precipitation has decreased. The signs of some trends are seasonal mean daily maximum and mini- consistent with the projections whereas others are not, but none of the changes is mum temperatures. statistically significant except for maximum and minimum temperatures, which were not among the parameters predicted by the models. Statistical models indicate that the greenhouse winter and summer precipitation signal could have been masked by natural climate variability, whereas the increase in the ratio of winter-to-summer precipitation and the higher rates of temperature change probably should have already been detected. If the models are correct it will likely take at least another 40 years before statistically significant precipitation changes are detected and another decade or two to detect the projected changes of temperature. A DEQUATE SUPPLIES OF WATER FOR hanced greenhouse world is the decrease of A crops and livestock are of primary summertime precipitation and an increase of importance to the central North temperature across central North America America region (1). Severe droughts of the (3). The model predictions admittedly suffer 20th century have affected both the biophys- from potentially significant limitations. For ical and socioeconomic systems of the re- example, some regional phenomena such as gion (2). One of the more ominous scenar- the El Niño-Southern Oscillation (ENSO) ios that has been suggested on the basis of are not adequately simulated by these mod- everal climate model simulations in an en- cls, but do appear to have linkages with regional surface temperature and precipita- tion variations in the United States (4). B Global Climate Laboratory, National Climatic Data Cen- ter, National Environmental Satellite. Data, and Infor- Nonetheless, if the projected climate occurs, mation Service, National Oceanic and Atmospheric Ad- it would certainly have a deleterious impact, ministracion, Department of Commerce, Federal Pig. 1. (A) The "IPCC Central" North American Building Asheville, NC 28801. and the region would have a difficult time region; (B) the "Central" United States region. JUN 07 '91 17:41 =CDC ASHVL. NC We developed autoregressive moving-av- trends in 1000 simulations of the ARMA Table 2. Observed trends of seasonal mean models without trends imposed. temperature (Temp) and total precipitation crage (ARMA) models using the seasonal (Precip) from 1895 to 1989. No trends are series of the precipitation and temper- No significant trends, even at the 0.10 statistically significant at the 0.10 level. Both a records for the IPCC Central region. significance level, could be found in the parametric test and a nonparametric test were data were transformed to standardized seasonal means of temperature, precipita- used. Slopes are percent per 100 years for departures for temperature and precipitation tion, or the ratio of winter-to-summer pre- precipitation and degrees centigrade per 100 by use of the normal and gamma statistical cipitation over the 1895 to 1989 time peri- years for temperature. distributions, respectively. Five models were od. We looked for both linear trends and Parameter Season Slope developed. Two were models for seasonal trends with one change point (9) (Table 2). temperature (one for winter and one for No significant trends were observed even Precip Winter -4.2 Precip Summer -0.7 summer) and two were for precipitation. though substantial changes of temperature Temp Winter 0.23 The fifth model pertained to the ratio of and precipitation over this region could Temp Summer 0.43 winter to following summer precipitation. have been anticipated on the basis of the Precip ratio Winter/summer -3.7 The models can be written as model projections. For example, winter pre- $ cipitation is expected to increase by as much Y, = Σ as 15% (Table 1), but it has fallen at a rate j-1 slightly greater than 4% per century (Table calculated the probability that those trends where Y, is the value of the time scries at 2). Summer precipitation has shown a slight would not be detected by our tests. The time ⑆ Φ, is the ith autoregressive (AR) decrease, but the magnitude is small, less magnitudes of the trends imposed were 7, coefficient; e, is the /th moving average than 1% per century (Table 2), despite 14, and 28% of the total projected changes (MA) coefficient; and a, is random noise (or projections of a decrease in precipitation by (Table 1). The rationale for these magni- a shock) at a time t. The order of the model 5 to 10% by 2030. The ratio of winter-to- tudes is that global temperatures have risen is expressed as the sum of p and q and summer precipitation is projected to in- by nearly 0.5°C since the 19th century (3), represented as ARMA (p. q). We assumed crease over this region, but it has decreased and this is 28% of the projected warming that a, had a normal distribution with mean at a rate of nearly 4% per century. At least (1.8°C) expected by the year 2030 (3). The zero and standard deviation σₐ. Most sta- the direction of temperature change in both 7 and 14% changes correspond to potential tionary processes can be fitted by an ARMA winter and summer is consistent with the undetectable greenhouse gas-induced model (10). projections, increasing by 0.23° and 0.43°C changes that are only a fraction (0.25 and The Akaike information criterion (AIC) per century, respectively. None of the 0.50) of that corresponding to the 0.5°C and the Bayesian information criterion changes, however, are statistically signifi- global mean temperature increase. The C) were both considered in the selection cant. The hypothesis that all of the time IPCC concluded that the observed warming he appropriate model order. Each model series are stationary ARMA processes can- of nearly 0.5°C is already at the low end of 48 tested up to order 12. After inspection not be rejected, and thus ARMA models are the sensitivity of current climate models to of the residual mean squares of each of the appropriate tools for further analysis of greenhouse-induced warming. models, a decision was made to use the AIC there data. For the projected rates of summer (three in the one instance where the two informa- We also considered the possibility that a of three models) and winter (two of three rion criteria provided conflicting advice (the greenhouse gas signal was masked by the models) warming. it is unlikely, but still ratio of winter-to-summer precipitation). natural climate variability in this region. To possible (probability between 0.05 and The AIC is known to somewhat overspecify answer this question we imposed trends of 0.25, depending on the model used), that a model (11), but in our analysis only one of various magnitudes on the 1000 simulations the changes projected would have been un- the models had an order as high as two, and representing the years 1895 to 1989 for detected if the rates of change in IPCC none of the models deviated greatly from each of the five parameters of interest and Central were proportional to the rate of white noise process. Four of the models are global warming (28% of the value projected ARMA(0,1), whereas the ratio of winter-to- by the year 2030, Fig. 2A). Only for the summer precipitation is an ARMA(1,1). Table 1. Estimates of changes in areal means of surface air temperature (Temp) and winter projections from the GFDL (3) mod- The time series generated from these precipitation (Precip) over central North el (2°C warming by 2030) did we find it models (if the models are perfectly specified) America (35° to 50°N, 80° to 105°W) from probable that changes could have been reproduce data that have all the characteris- preindustrial times to 2030, for the IPCC masked by natural variability. If climate tics of the observations. For each model, "business as usual" (3) emission scenario. These model sensitivity were reduced by a factor of 1000 time series were generated of length projections have been scaled to correspond to a global mean warming of 1.8°C by 2030. Model 2, we would likely (probability greater than 41 years (regarded as spanning the years abbreviations are the Canadian Climate Center 0.50) be unable to detect any of the green- 1990 to 2030) and 95 years (regarded as (CCC), Geophysics Fluid Dynamic Laboratory- house signals in the 95 years of thermomet- spanning the years 1895 to 1989). Linear USA NOAA (GFDL), and the United ric records. trends (a convenient simplification used in Kingdom Meteorological Office (UKMO) as For winter and summer total precipita- the absence of any compelling reasons to provided by IPCC (3). tion (Fig. 2B), the natural variability of the suggest otherwise) were superimposed on Model climate record makes it unlikely that we the output from these models. The trends Parameter Season would be able to detect the changes project- consisted of various percentages of the pro- CCC GFDL UKMO ed by the climate models, even if they were ed changes given in Table 1. The first Temp (°C) Winter 4 2 4 correct. In contrast, for the projected change in which a significant trend could be Temp (°C) Summer 2 2 3 in the ratio of winter-to-summer precipita- detected was calculated by use of the empir- Precip (%) Winter 0 15 10 tion, at least for two of the models in which ically based 0.05 significance level. This sta- Precip (%) Summer -5 -5 -10 Precip (%) Winter/ 5 21 22 large changes in that ratio are projected, we tistical significance level was derived from ratio should have already detected the changes. summer the distribution the clouse of the liness Once again this conclusion is atrictly valid 1 MARCH 1901 JUN 07 '91 17:43 =CDC ASHVL. NC P.5 1.00 only if the changes were proceeding at a rate Table 3. Observed trends of the scasonal mean Winter 0.80 in concert with the increase of global mean daily maximum and minimum temperatures over temperature (and if the increase of global the central region of the United States for the Probability 0.60 Summer GFDL period 1901 to 1987. Trends that are mean temperature is indeed greenhouse- statistically significant at the 0.10 significance 0.40 induced), that is, if we have experienced level are indicated by an asterisk (paranetric about 28% of the projected greenhouse- test). None were significant according to the 0.20 CCC. GFDL UKMO induced changes in temperature and precip- nonparametric test. The years of significant A CCC, UKMO 0.00 itation as of 1990. change points are listed along with the sign of 0.0 0.2 the associated shift in the time rate of change. 0.4 0.6 0.8 1,0 1.2 Assuming that all of the IPCC projections Temperature (°C per 100 yr) are correct and that the observed record to Slope Year of date is trend-free, we calculated when detec- Parameter Season (°C/100 change CCC. GFDL CCC UKMO 1.00 tion of the projected changes during the years) point 0.80 next 41 years would be likely. Another 15 to Maximum Winter 0.20 None UKMO GFDL 20 years would be required for the detection Minimum Winter 0.59 None Probability 0.60 Summer Winter of summer temperature changes (Fig. 3A). Maximum Spring 0.38 None Between 15 and 30 years would be required Minimum Spring *0.24 None 0.40 CCC Maximum Summer 0.45 for the detection of winter changes, depend- 1934(-) Minimum Summer *0.98 1935(-) 0.20 Winter-to-summer GFDL. UKMO ing on the projection and whether the Maximum Fall -0.55 None ratio B warming that has already occurred is consid- Minimum Fall 0.35 None 0.00 -2 o 2 4 6 cred. Detection of the projected change for Precipitation (percent per 100 yr) precipitation is less likely than for tempera- Fig. 2. Probability that trends of various magni- ture during the next four decades (Fig. 3B) ed change in the ratio of winter-to-summer tudes would have gone underected over the IPCC because of the large year-to-year variability. precipitation, even if all the changes began Central North American region during the period If the precipitation changes were about to occur in 1990, it is unlikely that we will 1895 to 1989: (A) temperature and (B) precipi- twice as large as those projected, we proba- be able to detect these signals over the next ration. Large dots indicate the trends that are bly (probability >0.5) could detect the few decades. Recently, Briffa et al. (13) have consistent with projections from each of three climate models when these projections are re- changes by about 2020. For the expected made similar conclusions for temperature duced to the fraction of global warming already change in the ratio of winter-to-summer changes in the Fennoscandia regions. observed compared with that projected in the year precipitation even though we should have These analyses leave planners and policy 2030 (reduced, that is, to 28% of rotal projec- already detected the larger rates of change makers in an unenviable position. They tions). Abbreviations are defined in Table 1. projected by the GFDL and UKMO models must decide to use or not use the IPCC during the past 95 years, it would not be projections without a clear rejection or ac- 2030 until around 2005 that we would be able to ceptance of many of those projections and 2025 detect the projected changes if they were to only a fuzzy notion that the projections may GFDL begin in 1990 and proceed at an approxi- be overly sensitive to greenhouse increases 2020 mately linear rate. when compared to the observed climate Years 2015 CCC, UKMO Although the general circulation model record in this region. To compound the 2010 results we used do not project changes in the dilemma, significant changes have now tak- Winter 2005 CCC, GFDL extremes of temperature, such extremes are en place with respect to increases of the 2000 UKMO useful diagnostic tools. Seasonal mean daily seasonal mean daily minimum temperatures A Summer minimum temperature has increased in the (which appear to be related to increased 1995 o 1 2 3 4 5 6 7 8 central United States (Fig. 1B and Table 3) cloud cover), but we do not yet have a Temperature (°C per 41 yr) during all seasons and at a significant rate sound understanding of how these changes during spring and summer. The rise of the may or may not be related to the greenhouse Summer Winter minimum is stronger than that of the max- effect. 2030 CCC imum in all seasons. Differential changes in REFERENCES AND NOTES 2025 UKMO UKMO the seasonal mean maximum and minimum CCC, GFDL 1. W. E. Ricbsame, S. A. Changnon, T. R. Karl, 2020 CCC temperatures indicate that because of a focus Drought and Natural Resources Management In the Years 2015 Summer Winter on changes of seasonal mean temperature United States: Impacts and Implications of the 1987- GFDL, UKMO alone, many important aspects of climate 1989 Drought (Westview, Boulder, CO. 1990); A. 2010 D. Hecht. J. Clim. Appl. Meteorol 22, 51 (1983). change are being overlooked. Recent work 2. R. W. Kates. 1. H. Ausubel. M. Barberian, Eds., 3006 Winter to summer B ratio (12) implies that the differential changes of Climate Impact Assessment (Wiley, New York, 1985). 2000 5 15 maximum and minimum are related to in- 3. Intergovernmental Panel on Climate Change -25 -15 -5 25 35 45 (IPCC), Scientific Assessment of Climate Change creases of cloudiness. Precipitation (Percent per 41 yr) (World Meteorological Organization. United Na- These results suggest that it is difficult to tions Environment Programme. Geneva, Switzer- Flg. 3. The year in which detecting trends (at the land. 1990). 0.05 significance level) of various magnitudes dismiss outright many of the IPCC projec- 4. K. E. Trenberth, G. W. Branstator, P. A. Arkin, becomes more likely than not (probability >0.5). tions as inconsistent with the observations, Science 242. 1640 (1988); K. E. Trenberth, Bull. (A) Temperature; (B) precipitation. Trends begin except for the change in the ratio of winter- Am. Meteoral. Soc. 71, 988 (1990). 5. Seasons are defined as follows: winter is December. in the year 1990. The solid dots represent the to-summer precipitation and the rates of January, and February: spring is March, April, and magnitudes that would be equivalent to the pro- icctions of CCC, GFDL, and UKMO from the summer temperature increase as projected May; summer is June, July, and August: fall is by the UKMO model. If these latter projec- September, October. and November. beginning of the industrial revolution to 2030 as 6. T. R. Karl. Clim. Change 12. 179 (1988), given by the IPCC (3). The open circles represent tions were correct, we probably should have and R. W. Knight. Historical Climatology Series 3-14 similar magnitudes after consideration of the already detected some statistically significant (National Climaric Data Center, Asheville, NC, changes already observed since 1895. 1985); T. R. Karl. L. K. Mercaif. M. L. Nicodemus. changes. Furthermore, except for the expect- R. G. Quayle. ibid. 6.1 National Climatic Dara 1060 JUN 07 '91 17:44 =CDC ASHVL. NC P.6 Center, Asheville, NC, 1983). 11. R. W. Katz, J. Atmos. Sci. 39, 1446 (1982). 7. T. R. Karl and C. N. Williams, Jr., J. Clim. Appl. 12. M. S. Plantico, T. R. Karl, G. Kukla, J. Gavin, Meteoral. 26, 1744 (1987). J. Geophys. Res. 95, 16617 (1990). 8. T.R. Karl, R. G. Baldwin, M. G. Burgin, Historical 13. K. R. Briffa et al., Nature 346, 434 (1990). Climatology Series 4-5 (National Climatic Dara Cen- 14. This work was partially supported by the Depart- ter, Asheville, NC, 1989). ment of Energy National Occanographic and Atmo- 9. A. R. Solow, ]. Clim. Appl. Meteoral. 27, 1401 apheric Administration Interagency Agreement DE- (1987). A106-90ER60952. 10. M. B. Priestly, Speciral Analysis and Time Series (Academic Press, Orlando, FL, 1981). 9 October 1990; accepted 28 December 1990 I MARCH The Great Global Conveyor Wallace S. Broecker Lamont-Doherty Geological Observatory of Columbia University Palisades, New York 10964 Submitted to Oceanography May 1991 INTRODUCTION A diagram depicting the ocean's "conveyor belt" has been widely adopted as a logo for the Global Change research initiative. This diagram (see figure 1) first appeared as an illustration in an article about the Younger Dryas event which appeared in the November 1987 issue of Natural History. It was designed as a cartoon to help the largely lay readership of this magazine to comprehend one of the elements of the deep sea's circulation system. Had I suspected that it would be widely adopted as a logo, I would have tried to "improve" its accuracy. In hindsight such repairs would likely have ruined the diagram both for the readers of Natural History and for use as a logo. The lure of this logo is that it symbolizes the importance of linkages between realms of the Earth's climate system. The ocean's conveyor is driven by the salt left behind as the result of the transport of water vapor through the atmosphere from the Atlantic Basin to the Pacific Basin. A byproduct of its operation is the heat which maintains the anomolously warm winter air temperatures enjoyed by northern Europe. A millinneum of very cold conditions known as the Younger Dryas appears to have been the result of a temporary shutdown of the conveyor. Thus the conveyor logo portrays the concern which led to the launching of the Global Change research initiatives; namely devilishly complex interconnections among the elements of our Earth's climate system will greatly complicate our task of predicting the consequences of global pollution. The objective of this paper is to provide a readable summary, from my prospective, of the conveyor's operation present and past. IT'S PATH The main problem with the logo is that it implies that if one were to inject a tracer substance into one of the conveyor's segments it would travel around the loop as a neat package eventually returning to its starting point. As we all know this is hardly the case. 1 Other circulation "loops" exist in the ocean and mixing occurs among the waters traveling along these intersecting pathways. The logo symbolizes a far more complex situation. To understand the logo's message, let's start at the point of origin of its lower limb and work our way around the ocean. Waters in the vicinity of Iceland are cooled through contact with the cold winter air masses which sweep in from the Canadian Arctic. The cooling densifies the surface water to the point where it can sink to the abyss and flow southward forming the conveyor's lower limb. In the logo this flow is depicted as a ribbon of water which jets its way through the deep Atlantic from the vicinity of Iceland to the tip of Africa. In reality it is a sluggish mass which fills most of the deep Atlantic. Its flow is more akin to that in a slow moving river than to that in a fast moving mountain stream. This water mass, known to oceanographers as the North Atlantic Deep Water (NADW), stands out as a tongue of high salinity, low nutrient content and high 14C/¹²C ratio water in sections drawn along the Atlantic's length (see Figure 2). Its only competitor for space in the deep Atlantic is a wedge of Antarctic Bottom Water (AABW) which underrides the NADW mass. This intruding water is mixed upward into the southward flowing NADW, increasing the transport by the conveyor's lower limb. Southward of 30°S the lower limb of the conveyor joins a rapidly moving deep current which encircles the Antarctic continent. This current serves as the great mix-master of the world ocean. It blends the NADW exiting the Atlantic with new deep water generated along the perimeter of the Antarctic continent and also with old deep waters recirculated back into the Antarctic from the deep Pacific and Indian Oceans. So efficient is this blending that the NADW entering from the Atlantic loses its identity before it passes even one half of a revolution around the Antarctic! 2 A rough quantification of the contribution of NADW to the deep waters of world ocean is provided by a property called PO₄ (Broecker, et al., 1991) which is defined as follows where PO₄ and O₂ are the measured phosphate and dissolved oxygen gas concentrations in a given water sample. The coefficient, 175 is the global Redfield coefficient relating O₂ consumption to PO4 release during respiration (Broecker, et al., 1985) and the coefficient 1.95 is arbitrarily introduced in order to bring the values of PO₄* into the range of deep water PO₄ concentrations. To the extent that the respiration coefficient is a constant, PO₄ constitutes a conservative property of any given deep water parcel; the increase in PO4 due to the oxidation of organic material is exactly balanced by the decrease in O₂/175. PO₄* is attractive as an indicator of the contribution of NADW because deep waters formed in the northern Atlantic have much lower PO₄ values than those formed in the Southern Ocean. Further the range of PO₄ values for the northern source waters (0.73 ± 0.03) and for southern source waters (1.67 ± 0.10) is small compared to the difference between the means for these end member values (1.67 0.73 = 0.94). The nomogram in figure 3 permits the conversion of deep sea PO₄ values into percentage contribution of NADW. As can be seen in the map in figure 4, at a depth of 3 kilometers the contribution of NADW to the deep water mix remains strong throughout the Atlantic, but after the conveyor's lower limb passes around the southern tip of Africa into the Antarctic it rapidly becomes blended with the high PO4* deep water generated along the edge of the Antarctic continent. In this way an ambient deep water mix with a PO* value of 1.37 is produced (see histograms in figure 4). This blend which consists of 1 part deep water produced in the northern Atlantic with about 2 parts of deep water produced in the Antarctic floods the deep Pacific and Indian Oceans. 3 A more complete picture of the geometry of the blending process is given by the PO₄ sections in figure 5. As can be seen, the more dense and higher PO₄ waters produced in the Antarctic mix upward and northward into the less dense and lower PO₄* waters entering from the Atlantic. By the time the water reaches the deep Indian, the deep Pacific and the Drake Passage, these endmember waters have been well blended. As depicted in the logo the lower limb water returns to the surface in the northern Indian and Pacific Oceans. In reality this upwelling is widely spread with a large amount taking place in the Antarctic. The logo also suggests that the major route for return flow to the Atlantic (i.e the conveyor's upper limb) is through the Indonesian archipelago and around the tip of Africa. This view was impressed on my brain by enthusiastic presentations by our society's president who stressed the role of the Agulhas current in global circulation (Gordon and Piola, 1983; Gordon, 1985 and 1986). As discussed below, this route probably accounts for only about one quarter of the return flow. A more important pathway is that through the Antarctic via the Drake Passage into the South Atlantic. These additional upwelling and return flow pathways are portrayed in figure 6. ITS FLUX It is my view that the magnitude of transport by the conveyor is best constrained by radiocarbon measurements on samples of deep water from the Atlantic Ocean. Some physical oceanographers might dispute this claim and opt instead for estimates derived from a combination of current meter measurements and geostrophic flow calculations. Fortunately the two approaches yield similar answers. The radiocarbon based estimate of the flux of NADW into the deep Atlantic is obtained by dividing the volume of water contained in the deep Atlantic by the radiodecay based mean residence time. This estimate must be corrected for the contribution made by 4 AABW to the conveyor's lower limb. It must also be corrected for the impact of temporal changes in the 14C/C ratio for atmospheric CO₂. The major obstacle to calculation from radiocarbon measurements of residence times for water in the deep Atlantic is the determination of the initial 14C/C ratio for each parcel. The reason is that all waters in the deep Atlantic are mixtures of northern component water with a comparatively high 14C/C ratio (4C = -68%o) and of southern component water with a comparatively low 14C/C ratio (414C = -158%o). Because of the large difference in the A14C values for these end members much of the variation in 14C/C ratio within the deep Atlantic is created by differences in the end member blend. As shown by Broecker et al. (1991a), PO₄ provides a quite accurate means of establishing the proportions of northern and southern component water in the sample analyzed for radiocarbon. The measured radiocarbon concentration is then subtracted from the initial concentration calculated for the mixture yielding the deficiency attributable to radiodecay. An example of this calculation is shown in Table 1. The radiocarbon measurements used in this analysis were made in the laboratories of Gote Ostlund at the University of Miami and Minze Stuiver at the University of Washington. Water column averages for radiodecay deficiencies are shown in figure 7 for all the stations occupied during the GEOSECS, TTO, TAS and SAVE expeditions. Little information is lost by this vertical averaging because significant trends with depth are not found for any of the stations. The vertically averaged deficiencies do however show a pronounced geographic trend. The lowest values (<10%) are found along the western margin of the Atlantic and the highest values (>30%) along the eastern margin. As radiocarbon decays by 1%o in 8.27 years, the isolation times corresponding to these radiocarbon deficiencies range from near zero for the western boundary in North Atlantic to as high as 300 years along the eastern boundary. This suggests rapid ventilation from both 5 ends of the Atlantic along the western boundary coupled with more leisurely dispersion into the interior. The radiocarbon deficiency for the entire deep Atlantic averages about 22%o. This corresponds to a residence time of about 180 years. The volume of the deep Atlantic reservoir is 1.55 X 1017 m³ (i.e., 2500m mean thickness with an area of 6.2 x 10¹³ m²). Hence to achieve this residence time requires a ventilation flux of 8.6 X 1014 m³/yr or 27 Sverdrups (1 Sverdrup = 106m³/sec). As the flux of AABW is about 4 Sverdrups, the flux of NADW is estimated to be 23 Sverdrups. This calculation assumes the system to be at steady state. While we have no way to know whether this is true for the water fluxes, we do know that the atmosphere's 14C/C ratio has changed with time. When these changes are taken into account, the flux has to be reduced by a factor of about 0.88 (Broecker et al. 1991b). Hence, we get a flux of close to 20 Sverdrups for the northern component (i.e., NADW). It is difficult to assess the error in this estimate but it is probably on the order of 25% (i.e., ± 5 Sverdrups). To appreciate the immense magnitude of this flux it is important to be reminded that it is 20 times the combined flow of all the world's river and somewhat larger than the rainfall over the entire globe! ITS DRIVE My contention is that the conveyor is driven by the excess salt left behind in the Atlantic as the result of vapor export (Broecker, et al., 1985). As-can be seen from the map in figure 8, the surface waters of the Atlantic are on the average 1 gm/liter higher in salt content than those in the Pacific. For sea water with temperatures in the range of those constituting the NADW mass (i.e. 2 to 4°C) one gram per liter extra salt has the same impact on the water's density as a cooling of 2 to 3°C. The salinity contrast between surface waters in the northern Atlantic and those at comparable latitudes in the northern Pacific is even larger, ranging from 2 to 3 grams per liter. This difference is so large that 6 surface waters in the northern Pacific even when cooled to their freezing point (i.e -1.8°C), sink to a depth of only a few hundred meters before reaching their buoyancy limit. Hence no deep water can form in the northern Pacific. Three means are available by which the magnitude of the vapor export flux can be estimated. The first approach is based on the water budget for the Atlantic Ocean and its continental drainage basin (see figure 9). Baumgartner and Reichel (1975) have constructed a water budget based on estimates of rainfall and evaporation over the Atlantic Ocean and runoff from its drainage basin. Their result is that vapor is being lost from the Atlantic basin at a rate averaging 0.45 Sverdrups. A second approach is to estimate the vapor export necessary to maintain the salinity differences in the sea against mixing among the ocean's water masses which tends to homogenize the sea's salt. If the mixing rates within can be determined and incorporated into an ocean mixing model, the fresh water budget for any region of the ocean can be determined. Broecker et al., (1990a) adopted this approach. Using a radiocarbon calibrated ocean box model they obtain a flux of 0.25 Sverdrups. The Princeton ocean GCM yields a water vapor loss of 0.45 Sverdrups (Manabe and Stauffer, 1988) and the Hamburg ocean GCM 0.20 Sverdrups (Ernst Maier-Reimer, personal communication). The third approach involves the determination of the net fluxes of water vapor across segments of boundary separating the Atlantic's drainage basin from the remainder of the world. To do this, Zaucker and Broecker (in press) have used wind and humidity data summarized by Oort (1983) to calculate vertically integrated and annually averaged vapor transports for all positions on the globe (using the 4° X 5° grid 11 level eggcrate geometry employed by GISS). They obtain in this way a net vapor loss from the Atlantic of 0.32 Sverdrups. In addition to providing an estimate of the magnitude of the vapor loss, this approach also yields the routes for this loss. As is shown in figure 10, vapor loss occurs 7 both in the belt of northern hemispheres westerlies and in the belt of tropical easterlies. For the westerlies substantially more vapor is exported across Eurasia than is imported across the North American cordilera. For the easterlies substantially more vapor is exported across Central America than is imported across Africa. Based on these results we estimate that the rate of vapor loss from the Atlantic basin is 0.35 ± 0.12 Sverdrups. To put this flux in context it helps to be reminded that it is about twice that for the Amazon River. Over the course of a year vapor export removes an amount of water equal to that in a 15 cm. thick layer covering the entire Atlantic! ITS SALT BUDGET Adopting a 20 Sverdrup export rate for lower limb water and an 0.35 Sverdrup fresh water loss from the Atlantic, it is of interest to see what combination of return flow water could balance the Atlantic's salt budget. As the outgoing lower limb water has a salinity of about 34.9% and the outgoing water vapor a salinity of 0.0%o, the aggregate salinity of the return flow water must be about 34.3% (see figure 11) or 0.6% lower than that of the outflowing lower limb water. Thus the salinity contrast between sea water and water vapor is about 60 times larger than the salinity contrast between the waters being traded between the Atlantic and the remainder of the ocean! It is for this reason that a measly 0.35 Sverdrup vapor loss can drive a mighty 20 Sverdrup ocean current! It should be kept in mind in this regard that were the salt buildup to go uncompensated, the salinity of the entire Atlantic would increase at the rate of about 1.4 gm/liter per millennium. As we shall see below, the conveyor appears to have been running more or less as it does today for the last 9000 years. Had the salt buildup not been compensated, the Atlantic's salinity would have increased during that time by a staggering 13%o. Clearly this can not have been the case. Rather, on the average over this period of time the export of salt via the conveyor's lower limb must have balanced the enrichment of salt by vapor loss. 8 The "remainder water," which feeds the upper limb of the Atlantic, has three components; Antarctic surface waters passing through the Drake Passage (S ≡ 33.8%o), Indian surface water passing around the tip of Africa via the Agulhas Current (S ≡ 35.1%o) and intermediate waters formed at the northern perimeter of the Atlantic segment of the Antarctic (S ≡ 34.3%). The salinity of the intermediate water matches that required to achieve salt balance. However this salinity could also be achieved by mixing 1.6 parts Drake Passage surface water with 1 part Agulhas water. So, based on salinity alone it is not possible to say how much of the remainder water enters the South Atlantic at intermediate depth and how much enters at the surface. A rough idea of the proportions of intermediate water on one hand and the Drake- Agulhas mix on the other can be obtained by invoking another constraint, namely, that the element phosphorus be conserved within the Atlantic. The reason is that the residence time of phosphorus in the ocean (several tens of thousands of years) is roughly 100 times the residence time of water in the Atlantic (several hundreds of years). Hence, there can be no significant gain or loss of phosphorus during a single ventilation cycle; in other words, the phosphorus contained in the exported water must match that contained in the imported water. Before employing this constraint I must point out that it has a possible flaw. While we assume in our calculations that phosphorus is transported only in inorganic form (i.e., PO4), the finding by Suzuki, et 1985, that sizable amounts of organic nitrogen are contained in low nutrient surface waters, opens the possibility that phosphorus is contained in the large molecules or fine particles thought to carry the organic nitrogen. If phosphorus is being shuttled back and forth between organic and inorganic forms, then the constraint suggested here becomes invalid. The phosphorus constraint stems from the fact that only Agulhas Water carries less PO4 than outgoing lower limb water; the other contributors to the return flow have 20 to 9 50% higher phosphate concentrations. Agulhas water carries no PO4 and Drake Passage water carries about 1.5 µm/kg PO₄; hence the Agulhas-Drake Passage mix required to yield the right salinity (1 part to 1.6 parts) carries about 0.9 µm/kg. Intermediate water carries about 1.8 µm/kg of PO₄. Hence to match the 1.3 µm/kg of PO₄ carried by the outgoing lower limb water requires a mix of about 56% Agulhas-Drake Passage component and about 44% intermediate water. This then suggests that the 15.35 Sverdrups of remainder water is made up of 6.7 Sverdrups intermediate water, 5.3 Sverdrups of Drake Passage water and 3.3. Sverdrups of Agulhas water. However because of the many simplifications required in order to make this calculation, these fluxes must be taken with a grain of salt (and perhaps also a dram of phosphorus). ITS BENEFITS The benefit provided by the conveyor is the heat it releases to the atmosphere over the northern Atlantic. This heat is responsible for Europe's surprisingly mild winters. The amount of heat released to the atmosphere is given by the product of the conveyor's flux and the temperature change required to convert upper limb water to lower limb water (i.e., to create NADW). The temperature of NADW averages about 3°C. As will be shown below, the temperature of the upper limb water averages about 10°C. Thus each cubic centimeter of upper limb water releases 7 calories of heat to the atmosphere during its conversion to deep water. At an average flux of 20 Sverdrups this totals 4 X 1021 calories each year, an amount of heat equal to 35% of that received from Sun by the Atlantic north of 40° latitude! Manabe and Stauffer (1988) have shown that indeed the thermohaline circulation of the Atlantic maintains high surface water temperatures in the northern Atlantic. Using the GFDL ocean GCM, they demonstrate that circulation in the Atlantic can assume two quite different modes; one with a strong thermohaline component akin to the conveyor and one 10 with no thermohaline circulation. The difference between surface water temperatures maintained by these modes is shown in figure 12. As can be seen when the conveyor is operative, the temperature of surface waters in the northern Atlantic average 5°C warmer than when it is off. Considering that strength of the thermohaline circulation in the Manabe and Stauffers' conveyor-on mode is only 12 Sverdrups, this warming should be even greater in the real ocean with its 20 Sverdrup thermohaline circulation. Rind et al., (1986) have used the GISS atmospheric GCM to estimate the geographical pattern of the winter air temperature change supported by the conveyors heat output. They adopted for the surface water temperature difference between the conveyor- on and conveyor-off modes that reconstructed by the CLIMAP group for glacial surface water relative to today's. As shown in figure 13, the air temperature anomaly obtained in this way extends across Europe into Siberia. As we will discuss below this geographic pattern matches very nicely that for the Younger Dryas cooling. Now let us return to the problem of estimating the temperature of the upper limb water supplying deep water formation in the northern Atlantic. The major supplier of water to the region of deep water formation is the Gulf Stream. As the roots of this great current extend to a depth of 1500 or so meters, the properties of the water supplied to the NADW source region depend strongly on the depth at which the major transport takes place. A large contrast exists between the water from depths of less than 500 meters and water from depths of greater 800 meters (see figure 14). For example, the upper waters have too high a salinity and the deep waters too low a salinity to match that needed to generate NADW. As summarized in figure 15, when the inputs into the northern region of low salinity Bering Sea water and of fresh water (through river runoff and the excess of precipitation over evaporation) are taken into account the salinity of the aggregate upper limb supply water must be about 35.6%o. This value is intermediate between those for upper and lower 11 Gulf Stream water. A similar situation exists for the nutrient constituents phosphate and silica. The deeper portion of the Gulf Stream carries too high concentrations and the upper portion too low concentrations to match that in outgoing NADW. Thus it appears that a mixture of upper and lower Gulf Stream water is required to create the salinity and nutrient content of new NADW. Radiocarbon provides a crosscheck on this conclusion (see Broecker, in press a). The prenuclear 14C value for new NADW was close to -66%o. The feed water for NADW production should have a value 6%o more negative than this (i.e., -66 6 = -72%)). The reason is that CO₂ exchange with the atmosphere adds more radiocarbon to the source region than is lost by radiodecay during residence in the source region. CO₂ exchange raises the 14C value by 11% while radiodecay lowers it by only 5%o. The prenuclear 14C value for Gulf Stream water was about -50%o; and that for lower Gulf stream water about -80%. In order to obtain the desired input value of -72% requires that about 1 part upper Gulf Stream water be mixed with 2 parts thermocline water carried beneath the Gulf Stream. These proportions match reasonably well those required to create the phosphate and silicate contents of new NADW. While the proportions of upper and lower Gulf Stream water necessary to achieve the salinity nutrient and radiocarbon contents of new NADW are not well defined, the temperature of the input water should lie between that of about 18°C for the upper water and that of about 6°C for the lower water. A 1 parts upper - 2 parts lower water mixture would have a temperature of 10°C. 12 ITS ACHILLES HEEL The addition of fresh water to the northern Atlantic poses a constant threat to the conveyor. Northward of 40°N in the Atlantic precipitation and continental runoff exceed evaporation by about 0.30 Sverdrups (Baumgartner and Reichel, 1975). In addition, the 1 Sverdrup of low salinity (S ≡ 33.0%) water entering the Arctic arm of the Atlantic through the Bering Straits contributes the equivalent of 0.06 Sverdrups of fresh water bringing the total to 0.36 Sverdrups. When the conveyor is running at its current strength of 20 Sverdrups, this fresh water is efficiently swept away causing only an 0.63% reduction in the conveyor water's salinity as it passes through the northern Atlantic. If the conveyor were to progressively weaken, this salinity reduction would grow. It would be 0.94% at 15 Sverdrups, 1.26% at 10 Sverdrups... At some point the salinity reduction would become so large that deep water could no longer form. The conveyor would shut down. Were this to happen, fresh water would pool at the surface of the northern Atlantic much as it currently does in the northern Pacific creating a severe barrier to deep water formation. Ocean GCM simulations by Manabe and Stauffer (1988) clearly demonstrates the role of this fresh water input. As shown in figure 12, for the conveyor-off mode pooling of fresh waters reduces the salinity surface waters in the northern Atlantic by about 3%o. Maier Reimer and Mikolajewicz, 1989, using the Hamburg ocean GCM show that a modest dose of excess fresh water to the source region of NADW can kill the model's thermohaline circulation. Furthermore, the demise is abrupt occurring on the time scale of a few decades. The rapidity of this response is not surprising for it depends on the residence time of water in the source region. At a flushing rate of 20 Sverdrups the entire volume of water contained in the Atlantic north of 45°N can be replaced in two decades! We know of no ocean GCM experiment which shows how the conveyor circulation might be restarted. Because of the strong barrier created by the pooling of fresh water, this 13 may prove to be a tricky task. Microprocesses such as densification through brine formation beneath sea ice may have to be invoked. ITS HISTORY The best indicator of the past operation of the conveyor is the air temperature in the northern Atlantic basin. The reason is that as we have already shown turning on and off the conveyor causes 5 to 8°C changes in the air temperatures over Greenland and Europe. The most detailed record of air temperature in the northern Atlantic basin is the isotope record preserved in the Greenland ice cap. This record (see figures 16 and 17) had different character during the last period of glaciation than during the present period of interglaciation (Dansgaard et al., 1971, Hammer et al., 1985). During glacial time air temperatures over Greenland underwent excursions of the magnitude and abruptness expected if the conveyor were turning on and off on an millennial time scale. By contrast, during the 9 or so thousand years since the period of glaciation came to a close, Greenland's air temperature has remained nearly constant. The impression I get from this is that a frenetic glacial conveyor became firmly locked in the on-position at the beginning of post glacial time and has remained so ever since. A possible explanation for the behavior of the conveyor during glacial time is that when the northern end of the Atlantic Basin is surrounded by ice sheets, stable operation of Atlantic's circulation system is not possible (Broecker, et al., 1990b). Rather, because the ice sheets constitute a tremendous source of fresh water, circulation in the Atlantic tends to flip back and forth between the conveyor-on and conveyor-off modes.. When the conveyor is operative its heat output tends to melt back the ice releasing large amounts of fresh water to the Atlantic. The conveyor also efficiently exports excess salt from the Atlantic. The combination of meltwater dilution and salt export drives down the density of waters in the Atlantic until the point is reached where the conveyor can no longer function. 14 It goes off. With the conveyor inoperative, the export of salt and the dilution with meltwater are reduced to the point where water vapor export once again begins to enrich salt in the Atlantic. The salt content and hence also the density of Atlantic waters steadily rise until the conveyor turns on again. This cycle repeats over and over again. Although a rigorous demonstration that such an oscillator did operate is not possible, the facts we do have lend strong support to this scenario. First of all, we know that the 40 million cubic kilometers of excess ice present in the ice sheets of the northern hemisphere during peak glacial time began to melt about 13,000 years ago and was gone by about 8000 years ago. Thus the average flux of meltwater during this interval must have been about 0.25 Sverdrups. Assuming this to be the magnitude of the melting rate during the proposed conveyor-on episodes then the dilution of salt due to meltwater would have been comparable to today's rate of vapor export. If the assumption is made that then as now the conveyor was exporting salt at a rate comparable to the rate it was being enriched through vapor export, then during the conveyor-on episodes, salt would have been diluted at a rate corresponding to the input of meltwater. When the conveyor was off the dominant important term in the salt budget would be vapor export. As yet we do not have an adequate estimate of the rate of water vapor export in the presence of an ice sheet. However, since vapor export is dictated by the interaction of planetary winds with mountain ranges, the rate may not have been very different from today's. Additional support comes from the observation that the average duration of individual warm and cold episodes recorded in Greenland ice ranges from about one half to two millennia. This spacing is consistent with that expected from the salt oscillation hypothesis. A vapor loss of 0.35 Sverdrups removes a layer 15 cm thick from the Atlantic each year. If uncompensated by salt export, the salt content of Atlantic waters will rise 1.4 gm/liter per millennium! As a salt buildup or reduction of 1 to 2 gm/liter for the Atlantic is 15 about what is needed to tip the balance between conveyor-on and conveyor-off, a match exists between the observed timing and that predicted timing (Birchfield and Broecker, 1990). This is a strong point in favor of the oscillator hypothesis! For an oscillator to function requires a combination of a long time constant drift (in this case the buildup or drawdown of salt) and a short time constant stablization mechanism. In the previous section it was shown that a tendency exists for fresh water to pool at the surface of the northern Atlantic. When the conveyor comes on this pool is quickly destroyed raising the salinity in NADW source region. This stabilizes the conveyor in its on-position. Similarly when the conveyor stops the pool quickly reappears stabilizing the conveyor in its off position. As can be seen in figure 17, many of the Greenland temperature cycles are characterized by abrupt warmings followed by more gradual coolings. The salt oscillator hypothesis provides a natural explanation for this shape (see Broecker, in press b). The abrupt warmings are caused by turn-ons of the conveyor. Immediately following such a reinitiation, the conveyor runs with extra vigor. The reason is that in order to overcome the fresh water pool present in the northern Atlantic when the conveyor is inoperative, the salinity of the Atlantic would have to rise above the level required for steady state operation. Thus when the conveyor comes on the buoyancy contrast between deep water formed in the northern Atlantic and deep water present in the remainder of the ocean will be unusually large. This excess density will drive the conveyor at an unusually high rate. As a consequence, a greater amount of heat will be released to the atmosphere over the north Atlantic. However, once operative, the conveyor's strength will steadily wane. The reason is that the combination of dilution with meltwater and export of excess salt will lessen the buoyancy contrast between deep waters inside and outside the Atlantic. As the strength of the conveyor wanes, the amount of heat given off to the atmosphere over the 16 northern Atlantic will also decrease causing air temperatures to drop. Eventually the conveyor will shut down abruptly cutting off the supply of ocean heat. The atmospheric temperature cycle generated in this way resembles that seen in the ice core record (see figure 18). Only for the most recent of these cycles do we have sufficient auxiliary evidence to add muscle to this scenario. The abrupt warmings at about 12,700 and at about 10,000 radiocarbon years ago, provide smoking guns in this regard. Not only the oxygen isotope record in ice cores (Dansgaard, et al., 1989) but also that in lake sediments on the European continent (Lötter and Zbinden, 1989) demonstrate that both of these warmings were accomplished in only 50 years! Further, the geographic pattern of these temperature changes associated with Younger Dryas is as expected if they were caused by conveyor turn ons. Pronounced changes are confined to latitudes greater than 40°N and extend from the maritime provinces of Canada and the ice cap of Greenland on the west across the northern Atlantic, the British Isles and Scandinavia into Russia on the east (see Rind, et al., 1986 for summary). Finally, Boyle and Keigwin 1987, have shown based on carbon isotope and cadmium concentration measurements on benthic foraminifera shells from a deep sea core from the vicinity of Bermuda that water of Antarctic origin flooded the western basin of deep Atlantic during Younger Dryas time, confirming that a shutdown of deep water formation accompanied this cold event. ITS FUTURE When the Natural History article containing the great global conveyor belt diagram appeared, the editor put a sales "stimulator" on the cover which stated "Europe beware: the big chill may be coming." At the time I was much annoyed because no mention of the conveyor's future was made in the article. To make matters worse, even after reading the article itself, many people were left with the impression that I was warning of an imminent 17 conveyor shutdown. The fact is that at that time I thought that the coming greenhouse warming would, if anything, strengthen the conveyor by increasing the rate of vapor loss from the Atlantic Basin. I had not given serious thought to the question as to whether any changes associated with man's activities might threaten the conveyor. The first activity which comes to mind in this regard is the rerouting of water for agricultural use. Irrigation projects increase the recycling of water on the continents and thereby change the point at which a given water molecule reenters the ocean. Of particular interest in this regard is the Russian proposal to divert the great northward flowing Siberian Rivers to the south for agricultural use. The result of such a diversion would be to increase the vapor loss from the Atlantic Basin for instead of flowing out of river mouths into the Arctic, the water would move through the atmosphere across Asia into the Pacific Basin. The longterm result would be to strengthen the conveyor. In addition to increasing vapor export from the Atlantic Basin, the greenhouse warming will increase the transport of fresh water to the northern Atlantic. On the short term (i.e. decades), the salinity decrease created in northern surface waters would be more important than the Atlantic-wide salinity increase caused by increased vapor loss from the Atlantic Basin. The reason is that the replacement time for waters in the northern Atlantic is shorter than the replacement time for waters in the upper limb of the conveyor. So if a threat to the conveyor is in the making, it is most likely to come in this way. To be on guard we should pay close attention to the climate and oceanography of the northern Atlantic Basin. The finding by Brewer et al., 1983 that the salinity of Atlantic deep waters to the north of 50°N declined between 1972 and 1981 and the finding by Schlosser et al., 1991 that deep ventilation of the Greenland Sea was shutdown during the 1980's are indications that changes do occur. Unfortunately we have no way to tell whether these changes signal natural fluctuations or anthropogenically driven trends. 18 CONCLUSIONS The conveyor is only one of many elements which together constitute the Earth's climatic system. It stands out because of its dramatic impact on the climate of a single region on our planet. We must keep in mind however, that the abrupt global warmings which heralded the termination of the last major glaciation can certainly not be explained by the conveyor alone (Broecker and Denton, 1989, 1990). Rather, elements of the system, such as the Hadley cell, which influence cloudiness and atmospheric water vapor content must also have been involved. The challenge of the Global Change Research Initiative is to understand the complex web of interactions which tie together the operation of these diverse elements. ACKNOWLEDGEMENTS I would like to thank the Exxon Corporation and Livermore National Laboratory for their generous support of my research. Instead of requiring me to write long proposals and reports, they encourage me instead to put this effort into articles such as this. My research on ocean chemistry is supported by DOE's CO₂ program grant no. LLNL B130547; that on the climate history of the Atlantic by NSF Climate Dynamics grant no. ATM 89-21306 and by NOAA grant no. NA90-AA-D-AC520; and that on atmospheric vapor transport by EPRI grant RP 2333-6. 19 Table 1. Example of radiocarbon deficiency calculation GEOSECS STATION 113 (11°N, 20°W, 4741m) O₂ PO₄ 14C µm/kg µm/kg %o 239 0.91 -120 PO*4= * 12/175 175 O₂ + PO₄ - 195 239/175 175 + 0.91 - 1.95 239 = 0.91 µm/kg 1.67 - PO4 1.67 - 0.91 frac. no. comp. = 1.67 - 0.73 = 1.67 - 0.73 = 0.81 frac. so. comp. = 1 - - 0.81 = 0.19 14Cinitial = 0.81(-68) + 0.19(-158) = -84% ДД¹⁴C = 14Cinitial - 14Cmeasured = (-84) - (-120) = 36% apparent age = 36%o X 8.27 yΓ/% ≡ 300 years Sea-to-Air Transfer Warm Shallow Current Cold and Salty Deep Current Figure 1. The great ocean conveyor logo (Broecker, 1987). 0 , 4 60 50 -70 50 80 -60 80 90 -100 90 -110 80 I -120 110 -110 -100 -130 -90 -80 -140 -70 2 DEPTH (km) -130 -90 -150 -90 -100 -160 3 -100 WESTERN -110 ATLANTIC -80 14 C -90 -120 -70 4 %0 -100 -130 -140 -160 150 -110 -120 5 60 40 20 O 20 40 60 S LATITUDE N O IO 10 20 20 30 30 10 I 40 40 50 60 50 30 WESTERN NOO 40 20 110 90 ATLANTIC 120 80 H4SiO4 2 70 60 20 µmol/kg DEPTH (km) 30 50 30 40 30 3 90 40 20 100 110 30 50 4 60 70 80 40 120 90 100 50 IIO 60 120 5 60 40 20 O 20 40 60 S LATITUDE N Figure 2. Sections of radiocarbon and of dissolved silicate in the western Atlantic based on measurements made as part of the GEOSECS program. In both the North Atlantic Deep Water clearly stands out from the over and underlying waters of Antarctic origin. The intermediate and bottom waters which enter the Atlantic from the Antarctic have higher in silica concentrations and lower 14C/C ratios than the NADW which constitutes the conveyor's lower limb. 0.7 NADW 100 0.8 0.9 75 1.0 PO4* (µmol/kg) 1.1 1.2 50 1.3 DEEP 1.4 PAC. + IND. PERCENT NORTHERN COMP. 25 1.5 1.6 WSBW O 1.7 PO4* = PO4 + 175 O₂ - 1.95 Figure 3. Nomogram showing the relationship between PO₄ value (Broecker, et al., 1991a) and the percentage contribution of NADW to waters in the deep sea mix. Deep waters formed in the northern Atlantic have nearly uniform low PO₄ values (0.73 ± 0.03 µm/kg). Those which form around the Antarctic continent have high values (between 1.6 and 1.7 µm/kg). Bottom water formed in the Weddell Sea (WSBW) has a PO₄ value close to 1.67 µm/kg. The deep Pacific and Indian Oceans are flooded with a nearly uniform mix of these two end members (33% NADW + 67% WSBW). 120 160 160 120 80 40 0 40 80 120 = 0 60 60 NADW ME AN 0.73 1. 37 30 MEAN 40 1.36 60 40 0.8 NO. OF STATIONS NO. OF STATIONS 40 20 } 30 , 20 20 10 20 o E O 0.9 I O .30 1.35 1.40 20 10 20 PO4* (µmol/kg) 1.0 1.1 1.3 40 O 40 D .30 1.35 1.40 1.45 PO₄ * (µmoi/kg) 1.2 1.4 1.5 1.4 60 1.5 1.6 60 1.6 WSBW 1.67 120 160 160 120 80 40 o 40 80 120 Figure 4. Map of PO₄ values at 3 kilometers depth. The deep water source in the northern Atlantic has a PO₄ value of 0.73 µm/kg. and that in the Antarctic a value of about 1.67µm/kg. As shown by the histograms, waters on this depth horizon in the Indian and Pacific Oceans have nearly constant PO₄ values. Although the Pacific GEOSECS stations shows a range of PO₄ values, the lack of & geographic coherence suggests that this spread is the result of station to station shifts in the calibration of the nutrient analyses system. 63 : 528 3121 1 : 120 126 122 137 339 G-7 642 346 149 61 353 354 155 556 657 2 G58 059 560 ,61 364 102 693 G92 191 190 $ 8 5 2 PO4 1.0 1.0 1.1 0.8 1.2 3 3 1.3 1.2 DE DEPTH (km) WESTERN ATLANTIC DE TH DEPTH (km) 1.4 0.9 1.4 1.5 4 1.1 4 1.6 1.3 0.9 1.4 1.0 1.5 1.6 EASTERN ATLANTIC 5 5 40 20 O 20 40 40 60 °N EQU S °S 2 G428 G429 G430 2 WESTERN INDIAN 23 210 229 20 1.3 220 212 3. 3 233 343 230 413 " 1 use 331 2019 1.4 126 4480 426d DE PTH DEPTH (km) - ... 300 4 1.5 : 7 193 Ef "yzz 4 2 m. 3 2 1.6 2030 5 à 5 40 60 °S 3 G76 G77 G305 G303 G301 G299 6298 G293 G292 G290 G436 G435 G434 G433 G432 2 2 2 1.4 1.3 3 3 3 1.4 DEP TH (km) 5 DE (km) 1.4 4 DE DEPTH (km) 1.5 1.6 4 4 4 WESTERN EASTERN PACIFIC 1.5 INDIAN DRAKE PASS. 5 5 5 55 60 40 60 40 60 °S S °S Figure 5. Sections of PO₄ down the western basin of the Atlantic and around the Southern Ocean. The map shows the location of the GEOSECS stations and of the sections. Note how the <1.3 µm/kg and the >1.6 µm/kg waters blend as they move around the Antarctic continent eventually producing a nearly uniform mix with a PO₄* value close to 1.4 µm/kg. 3 3 Figure 6. The dotted arrows show the location where intense upwelling of deep water occurs. The clear arrows show the upper ocean routes of return flow which balance the outflow from the Atlantic of lower limb water. The numbers in the arrowheads represent the rough estimates of the magnitude of the fluxes (in Sverdrups). In addition to the features portrayed by the logo, this diagram shows that much of the upwelling occurs in the Antarctic and that much of the return flow occurs through Drake Passage. 80° 60° 40° 20° 0° 20° 60 60 + + 50° + .-25 50° -29 40° -6+ +-11 -9 + 40° -2 -5, -9 -31 3+ IO -13 -12 30 30° + -16 #20 -25 30° -6 F11 -15 -19+ 20 + + -28 + -30 +-7 -23 -20 -37 20° -27 + -27 20° C -17. -13 -20+ + *-11 -I6 -24 -31 10° -23 $23 -13 10° -11- -29 -9 -8 0° -6+ +18 -19+ *-23 .-21 0° -19 +-21 -6 10° -26 14 -25 -35 10 *-22 20° -17 + 20° -19 -17 -23 + -27 + 30° -26 -29. 30" RADIOCARBON 40° DECAY 40 14 C 50° %. 50" >2000 meters +++++ + + 60° + + 60° 80° 60° 40° 20° 0° 20° Figure 7. Water column averages for the radiodecay induced 14C deficiency (Broecker, et al., 1991a) at stations occupied during the GEOSECS, TTO, TAS and SAVE expeditions. The major gradient is from low deficiencies along the western margin to high deficiencies along the eastern margin. The residence times corresponding to the 10, 20 and 30%o deficiency contours are respectively 83, 166 and 249 years. The mean deficiency for this entire region of the Atlantic is estimated to be 22%o which corresponds to an isolation time of about 180 vears. 90°N 33.0 32.0 31.5 29.5 90°N 29.0 34.5 34.07 28.0 29 30.0 30.0 32.5 28.5 330 35.0 34.5 22.0 27.5 29 32.5 335-340 30.5 31 310 290 33.01 345 60° 32.0 340 350 DD 35.0 32.5 3525 I 50° 1 32.5 32.75 32.0 34.5 33.0 32.75 33.5 35.5. 33.5 340 36.0 345 30° 305 3525> 36.5 35.25 35.25 37 C 30' " 36.25 31.5 35.0 23725. 33.5 36.0 33.0- 3475. 3425. 36.0 32.0 34.5 3425 35.5 . 32.5% 34.25 340 33.75 33.5 35.0 36.0 33.0 , 345 33.0 F.O 35.25 33.5 34.0 350 35.5 35.0 asc 35.01 340 E : 34.0 3425 355 360 36.0 35.51 34.5 35.0 36.0 35.25 35.25 3575 ₹36.5 370 355 30 & 30° 35.5 3575 35.5 360 35.0 35.0 355 30 34.5 34.5 34.0 35.0 340 33.5 33.75. 345 33.5 340 60° 340 33.75 3425 33.75 .33.75 340 50 33.75 340- 33.75 34.25 3375- 90°S 0° 90'S 30°E 60° 90° 120° 150°E 180° 150°W 120° 90° 60° 30°W 0° Figure 8. Map of surface ocean salinity (Levitus, 1982). Regions of low salinity (i.e. S<34%) are cross hatched. Regions of high salinity (i.e. S > 36%o) are dotted. Note that on the average surface waters in the Atlantic are about 1%o higher in salinity that their about 2%o. Pacific counterparts. For the high northern latitudes the difference is even larger averaging 180 150 120 90 60 30 O 30 60 90 120 150 180 75 75 60 60 45 45 30 30 15 d 15 0 F O 15 15 DRAINAGE BASINS 30 ATLANTIC 30 45 PACIFIC 4! INDIAN 60 60 INTERNAL 180 150 120 90 60 30 O 30 60 90 120 150 180 Figure 9. Map showing continental areas whose drainage reaches the Atlantic, Pacific and Indian Oceans. The black areas are deserts from which no drainage to the sea occurs. About 0.35 Sverdrups of fresh water are transported from the Atlantic to the Pacific Ocean moving as vapor in part across Central America and in part across Africa. go VAPOR LOSS RATE 60 Sv 0.07 0.25 +0.18 30 0.36 0.19 +0.17 + 0 0.23 0.04 -30 9 0.03 -60 NET VAPOR LOSS IN SVERDRUPS 0.32 -90 F180 -150 -120 -90 -60 -30 0 30 60 go 120 150 180 Figure 10. Map showing vertically integrated and annually average water vapor flux vectors compiled by Oort (1983). Also shown is the boundary of the Atlantic's drainage basin and net fluxes across segments of this boundary. The tropical easterlies carry out more water vapor of the Atlantic basin across Central America than they bring in via Africa. So also do the northern westerlies; more water escapes across Asia than enters across the American cordilera. For the entire basin the rate of water vapor loss is 0.32 Sverdrups (see Zaucker and Broecker, in press). 0.35 - I9 15.35 UPPER LIMB I 4 20 20 LOWER LIMB CONVEYOR 4 4 AABW 30°S 60°N ATLANTIC'S SALT BUDGET SALINITY FLUX OUT %. Sv NADW 34.9 20.0 VAPOR 0.0 0.35 TOTAL 34.3 20.35 IN BERING STS 33.0 1.0 AABW 34.7 4.0 REMAINDER 34.3 15.35 TOTAL 34.3 20.35 Figure 11. Approximate water and salt budget for the Atlantic Ocean. The fluxes of Bering Sea surface water and of AABW are independently determined. The remainder water must then have a flux of 19.35 minus the amount of NADW recirculated within the Atlantic. With the guess of 4 Sverdrups recirculation adopted here the remainder water flux becomes 15.35 Sverdrups. The aggregate return water must have a salinity of 34.3% to balance the Atlantic's salt budget. Since the combined Bering Sea (Coachman and Aagaard, 1988)and AABW inputs have a salinity of 34.3%o, the remainder water must also have this salinity. 6 5 4 2 3 AT(°C) 60W O 1.5 3.0 3.5 2.0 1.0 Д S (%₀) .5 Figure 12. Maps showing the difference in North Atlantic surface water temperature (upper) and surface water salinity (lower) between the Manabe and Stauffer (1988) model circulation schemes with and without thermohaline circulation. As can be seen, a turnoff of thermohaline circulation leads to a pronounced cooling and freshening of the waters in the northern Atlantic. -60 O 60 120 4 -2 6 2 8 4 12 10 6 6 4 2 O 2 4 10 8 6 2 -60 O 60 120 Figure 13. Results of an experiment carried out using the GISS atmospheric GCM (Rind et. al, 1986). Two runs were made which differed only in the surface ocean temperature assigned to the northern Atlantic (see upper panel). The resulting winter air temperature change produced by this ocean temperature change is shown in the lower panel. PO4 (µmol/kg) SALINITY (%)) 0.0 0.5 1.0 1.5 35.0 35.5 36.0 36.5 O 500 DEPTH IN METERS 1000 1500 MEAN FOR SUPPLY H2O MEAN FOR SUPPLY H2O 2000 2500 SiO 2 (µmol/kg) POTENTIAL TEMP (°C) O 5 10 15 5 10 15 20 O 500 DEPTH IN METERS 1000 1500 2000 MEAN FOR SUPPLY H2O TTO STN 6 34.7°N 67.3° W IN GULF STREAM 2500 Figure 14. Phosphate, silica, salinity and potential temperature profiles at TTO station 6 in the Gulf Stream. Shown for comparison are the average values for these constituents required for the water constituting the upper limb source for NADW production (vertical dashed line). 0.3 UPPER 18.7 LIMB 1.0 20 20.0 LOWER LIMB 40°N NORTHERN ATLANTIC'S SALT BUDGET SALINITY FLUX OUT %. Sv NADW 34.95 20.0 TOTAL 34.95 20.0 IN BERING STS 33.0 1.0 VAPOR 0.0 0.3 UPPER LIMB 35.6 18.7 TOTAL 34.95 20.0 Figure 15. Water and salt budgets for the region in the northern Atlantic where NADW forms. The fluxes and salinities of new NADW, Bering Straits input and fresh water input are known. The water flux and salinity for the upper limb are obtained by difference. TODAY ~ ~10,000 YOUNGER DRYAS 14c YRS BP ALLEROD - 40 -35 -30 -25 S 18 O (%o) Figure 16. Oxygen isotope record for the Camp Century Greenland ice core (Dansgaard et al., 1971) covering the period from about 13,000 radiocarbon years B.P. to the present. 8¹⁸O (%o) DUST CONTENT (mg/kg) - 38 -36 -34-32 -30 O 2 3 250 YOUNGER DRYAS 1800 ALLEROD DEPTH (m) 200 1850 DISTANCE FROM BEDROCK (m) 150 1900 100 Figure 17. Oxygen isotope and dust records for that part of the Dye 3 Greenland ice core covering the time period ~45,000 to ~8000 years B.P. (Hammer et al., 1985). Note that many of the oxygen isotope events are characterized by rapid warmings followed by more gradual coolings. WARM CONVEYOR CONVEYOR 'ON' WEAKENING CONVEYOR 6°C 'ON' CONVEYOR 'OFF' COLD I I I 10 12 13 RADIOCARBON AGE THOUSANDS OF LIBBY YEARS Figure 18. Diagrammatic representation of the temperature record for the northern Atlantic basin from just before 13,000 to just after 10,000 radiocarbon years before present. The features shown in this diagram appear in records of 180 in ice, of ¹⁸O in lake sediment, of ocean planktonic foraminifera (Ruddiman and McIntyre, 1981), of pollen and of beetles (Atkinson et al., 1987). Of particular interest is rapidity of the warmings at 12,700 years BP and at 10,000 years BP. Detailed measurements in the Dye 3 ice core and in varved lake sediments reveal that these warmings took place in about 50 years! REFERENCES Atkinson, T.C., K.R. Briffa and G.R. Coope, Seasonal temperatures in Britain during the past 22,000 years, reconstructed using beetle remains, Nature, vol. 325, February, 1987. Baumgartner, A. and Reichel, E., Die Weltwasserbilanz, Oldenbourg, München, 1975. Birchfield, G.E., W.S. Broecker, A salt oscillator in the glacial northern Atlantic? Part II: A 'Scale Analysis' Model, Paleoceanography, vol. 5, no. 6, 835-843, Dec. 1990. Boyle, E.A. and L. Keigwin, North Atlantic thermohaline circulation during the past 20,000 years linked to high-latitude surface temperature, Nature, 330, 35-40, 1987. Brewer, P.G., W.S. Broecker, W.J. Jenkins, P.B. Rhines, C.G. Rooth, J.H. Swift, T. Takahashi, and R.T. Williams, A Climatic freshening of the deep Atlantic north of 50°N over the past 20 years, Science, 222:1237-1239, 1983. Broecker, W.S., D. Peteet and D. Rind, Does the ocean-atmosphere have more than one stable mode of operation? Nature, 315:21-25, 1985. Broecker, W.S., The Biggest Chill, Natural History Magazine, p. 74-82, Oct. 1987. Broecker, W.S., and G.H. Denton, The role of ocean-atmosphere reorganizations in glacial cycles, Geochimica et Cosmochimica Acta, 53:2465-2501, October, 1989. Broecker, W.S. and G.H. Denton, What Drives Glacial Cycles? Scientific American, Jan. 1990. Broecker, W.S., T.-H. Peng, J. Jouzel, and G. Russell, The magnitude of global fresh water transports of importance to ocean circulation, Climate Dynamics, 4:73-79, 1990a. Broecker, W.S., G. Bond, M. Klas, G. Bonani, and W. Wolfli, A salt oscillator in the glacial North Atlantic? 1. The Concept, Paleoceanography, vol. 5, no. 4:469-477, August 1990b. Broecker, W.S., S. Blanton, and T. Takahashi, W. Smethie and G. Ostlund, Radiocarbon decay and oxygen utilization in the dep Atlantic Ocean, submitted to Global Biogeochemical Cycles, vol. 5, 87-117, March, 1991a. Broecker, W.S., A. Virgilio, and T.-H. Peng, Radiocarbon Age of Water in the Deep Atlantic Revisited, Geophys. Res. Lett., vol. 18, no. 1, 1-3, Jan. 1991b. Broecker, W.S., The Conveyor's Upper Limb: The Source for North Atlantic Deep Water, in press a, Jour. Geophysical Research. Broecker, W.S., The Strength of the Nordic Heat Pump, 13,500 to 9500 B.P., in press, Erice volume. Coachman, L.K., and K. Aargaard, Transports through Bering Strait: Annual and Internannual Variability, Jour. Geophys. Research, vol. 93, no. C12, 15,535-15,539, 1988. Dansgaard, W., S.J. Johnsen, H.B. Clausen, and C.C. Langway, Jr., in The Late Cenozoic Glacial Ages, (ed. by Karl K. Turekian), New Haven and London, Yale Univ. Press, pp. 37-56, 1971. Dansgaard, W., J.W.C. White, and S.J. Johnson, The abrupt termination of the Younger Dryas climate event, Nature, 339:532-533, 1989. Gordon, A.L., and A.R. Piola, Atlantic Ocean upper layer salinity budget, Jour. Phys. Oceanogr., 13, 1293-1300, 1983. Gordon, A.L., Indian-Atlantic Transfer of Thermocline water at the Agulhas retroflection, Science, 227, 1030-1033, 1985. Gordon, A. Interocean Exchange of Thermocline Water, Jour. Geophys. Research, vol. 91, no. C4, 5037-5046, 1986. Hammer, C.U., H.B. Clausen, W. Dansgaard, A. Neftel, P. Kristinsdottir, and E. Johnson, Continuous impurity analysis along the Dye 3 deep core. In Greenland Ice Core: Geophysics, Geochemistry, and the Environment (eds. C.C. Langway, H. Oeschger, and W. Dansgaard); Amer. Geophys. Union Mon. 33, pp. 90-94. Levitus, S., Climatological Atlas of the World Ocean, NOAA Professional Paper No. 13, (U.S. GPO, Washington, DC)., 1982. Lotter, A.F. and H. Zbinden, Late-Glacial pollen analysis, oxygen-isotope record, and radiocarbon stratigraphy from Rotsee (Lucerne), Central Swiss Plateau, Eclogae geol. Helv. 82/1: 191-202, 1989. Maier-Reimer, E. and U. Mikolajewicz, Experiments with and OGCM on the cause of the Younger Dryas, in Oceanography, (A. Ayala-Castanares, W. Wooster, and A. Yanez- Arancibia, ed.), 87-100, UNAM Press, Mexico, 1989. Manabe S., and R.J. Stauffer, Two stable equilibria of a coupled ocean-coupled ocean- atmosphere model, J. Climate, 1:841-866, 1988. Oort, A.H., Global atmospheric circulation statistics, 1958-1973, NOAA Professional Paper 14, 1983. Rind, D. D. Peteet, W.S. Broecker, A. McIntyre and W. Ruddiman, The impact of cold North Atlantic sea surface temperatures on climate: Implications for the Younger Dryas cooling (11-10k), Climate Dynamics, 1:3-33, 1986. Ruddiman, W.F. and A. McIntyre, The North Atlantic Ocean during the last deglaciation, Paleogeogr., Palaeoclim., Palaeoecol, 35:145-214, 1981. Schlosser, P, G. Bönisch, M. Rhein and R. Bayer, Reduction of Deepwater Formation in the Greenland Sea during the 1980s: Evidence from Tracer Data, Science, 251:1054-1056, 1991. Suzuki, Y. and Y. Sugimura and T. Itoh, A Catalytic Oxidation Method for the Determination of Total Nitrogen Dissolved in Seawater, Marine Chemistry, 16, 83-97, 1985. Zaucker, F. and W.S. Broecker, Atmospheric water vapor transport from a general circulation model, in press, Jour. Geophysical Research. What Drives Glacial Cycles? Massive reorganizations of the ocean-atmosphere system, the authors argue, are the key events that link cyclic changes in the earth's orbit to the advance and retreat of ice sheets by Wallace S. Broecker and George H. Denton E ight times within the past mil- seasonality changes act directly on ling seasonality is the shape of the lion years, something in the the ice sheets of the Northern Hemi- earth's orbit. Over a period of 100,000 earth's climatic equation has sphere. A reduction in summer sun- years, the orbit stretches into a more changed, allowing snow in the moun- shine allows ice to build up, and an eccentric ellipse and then grows more tains and the northern latitudes to re- increase melts it away; the ice in turn nearly circular again. As the orbital ec- main where it had previously melted alters the earth's climate. In contrast, centricity increases, the difference in away. The snow compacted into ice, we think the ice sheets were a con- the earth's distance from the sun at and the ice built up into glaciers and sequence of broader climatic events. the orbit's nearest and farthest points ice sheets. Over tens of thousands of By altering patterns of evaporation grows, intensifying the seasons in one years, the ice sheets reached thick- and rainfall, the changes in seasonal hemisphere and moderating them in nesses of several kilometers; they intensity appear to have caused the the other. (At present the earth reach- planed, scoured and scarred the land- ocean and atmosphere (a single, cou- es its farthest point during the South- scape as far south as central Europe pled system) to flip from one mode ern Hemisphere winter; as a result, and the midwestern U.S. And then of operation to another, very different southern winters are a little colder- each glacial cycle came to an abrupt mode. With each flip, ocean circula- and summers a little warmer-than end. Within a few thousand years, the tion changed and heat was carried their northern counterparts.) ce sheets shrank back to their pres- around the globe differently, the prop- A third astronomical fluctuation ent-day configurations. erties of the atmosphere were altered, governs the interplay between the tilt Over the past 30 years, evidence has climate changed-and the ice sheets and eccentricity effects. It is the pre- mounted that these glacial cycles are grew or shrank. cession, or wobble, of the earth's spin ultimately driven by astronomical fac- axis, which traces out a complete cir- tors: slow, cyclic changes in the eccen- O ur proposal is not a rejection cle on the background of stars about tricity of the earth's orbit and in the of the astronomical theory of every 23,000 years. The precession tilt and orientation of its spin axis. By the ice ages but an extension determines whether summer in a giv- altering the intensity of the seasons, of it. The hypothesis was first pro- en hemisphere falls at a near or a far the astronomical cycles somehow tip posed in 1842, just a few years af- point in the orbit-in other words, the balance between glacial buildup ter the Swiss-American naturalist Lou- whether tilt seasonality is enhanced and glacial retreat. But what is the link is Agassiz argued that polished and or weakened by distance seasonality. between astronomy and the ice ages? scarred rocks and heaps of detritus in When these two controllers of season- How are the seasonality changes lever- the Alps recorded some past age of ality reinforce each other in one hemi- aged into global changes in climate? glaciers. In that year the French math- sphere, they oppose each other in the Any answer must contend with the ematician Joseph A. Adhémar suggest- opposite hemisphere. vast array of evidence that has ac- ed that astronomically driven changes cumulated about the nature, timing in the intensity of the seasons might WALLACE S. BROECKER and GEORGE and extent of the climatic shifts that periodically trigger glaciation. H. DENTON bring diverse interests to accompanied ice buildup and retreat. The Yugoslav astronomer Milutin Mi- their study of ice ages. Broecker got his Many workers have proposed that the lankovitch refined and formalized the Ph.D. at Columbia University in 1958 hypothesis in the 1920's and 1930's. and has pursued his career there. He The astronomical pacemaker he advo- is now professor of geochemistry at the cated has three components, two that Lamont-Doherty Geological Observatory ICE FIELD IN PATAGONIA ends in a deep change the intensity of the seasons of Columbia University. In addition to glacial lake. Such Southern Hemisphere ancient climates, he follows research in- and a third that affects the interaction glaciers have grown and shrunk in con- terests in ocean chemistry, isotope dat- between the two driving factors. The cert with the great northern ice sheets, ing and environmental science. Denton first is the tilt of the earth's spin axis. according to radiocarbon dating of veg- is professor of geology at the University etation (such as the trees in the fore- Currently about 23.5 degrees from the of Maine. After earning a Ph.D. at Yale vertical, it fluctuates from 21.5 de- University in 1965, he did postdoctoral round) that was overwhelmed by ad- ancing glaciers or that took root after grees to 24.5 degrees and back every work at the University of Stockholm and 41,000 years. The greater the tilt is, the then moved to Maine. He has spent 36 their retreat. The timing is a puzzle be- cause the intensity of summer sunshine, more intense seasons in both hemi- seasons in the field studying the tim- ing and extent of glacial advances, 22 of which is thought to influence ice growth, spheres become: summers get hotter them in Antarctica and elsewhere in the changes on quite different schedules at and winters colder. Southern Hemisphere. middle latitudes in the two hemispheres. The second, weaker factor control- SCIENTIFIC AMERICAN January 1990 49 Milankovitch calculated that these many years, however, the lack of an ations. It came from a seemingly odd three factors work together to vary the independent record of ice-age timing place, the sea floor. Single-cell marine amount of sunshine reaching the high made the hypothesis untestable. organisms called foraminifera house northern latitudes in summer over a themselves in shells made of calcium range of some 20 percent-enough, he I n the early 1950's Cesare Emiliani, carbonate. When the foraminifera die, argued, to allow the great ice sheets working in Harold C. Urey's labora- sink to the bottom and contribute to that advanced across the northern tory at the University of Chicago, the sea-floor sediments, the carbonate continents to grow during intervals of produced the first complete record of of their shells preserves certain char- cool summers and mild winters. For the waxings and wanings of past glaci- acteristics of the seawater they inhab- ited. In particular, the ratio of a heavy isotope of oxygen (oxygen 18) to or- dinary oxygen (oxygen 16) in the car- 21.5 DEGREES bonate preserves the ratio of the two oxygens in the water molecules. It is now understood that the ratio 24.5 DEGREES of oxygen isotopes in seawater closely SUN EARTH tracks the proportion of the world's water that is locked up in glaciers and ice sheets. A kind of meteorological distillation accounts for the link. Wa- ter molecules containing the heavier isotope tend to condense and fall as precipitation a tiny bit more readily 0 than molecules containing the lighter I isotope. Hence, as water vapor evapo- rated from warm oceans moves away from the source, its oxygen 18 pref- erentially returns to the oceans in 100 precipitation. What ultimately falls as snow on ice sheets and mountain gla- II ciers is relatively depleted of oxygen 18. As the oxygen 18-poor ice builds up, the oceans become relatively en- 200 riched in the isotope. The larger the ice sheets grow, the higher the propor- III tion of oxygen 18 becomes in seawa- ter-and hence in the sediments. THOUSANDS OF YEARS AGO Analyzing cores drilled from sea- 300 TERMINATIONS floor sediments, Emiliani found that the isotopic ratio rose and fell in IV rough accord with the cycles Milan- kovitch had predicted. Since that pi- oneering observation, oxygen-isotope 400 measurements have been made on V hundreds of cores. A chronology for the combined record enabled James D. Hays of Columbia University, John Im- brie of Brown University and Nicholas 500 VI Shackleton of the University of Cam- bridge to show in 1976 that the record contains the very same periodicities as the orbital processes. Over the past 800,000 years, the 600 global ice volume has peaked every VII 100,000 years, matching the period .8 .9 1.0 ICE VOLUME SUMMER SUNSHINE of the eccentricity variation. In addi- (THOUSANDS OF CALORIES PER tion, "wrinkles" superposed on each SQUARE CENTIMETER PER DAY) cycle-small decreases or surges in ice volume-have come at intervals of ASTRONOMICAL CYCLES (top) are the pacemaker of glaciation. The cycles-23,000 to roughly 23,000 and 41,000 years, in 100,000 years in length-affect the eccentricity of the earth's orbit, the orientation keeping with the precession and tilt of its spin axis (which slowly traces out a cone in space) and the tilt of the axis (which affects the width of the cone). The effect of the changes on the intensity of summer frequencies. Imbrie, working with a sunshine at high northern latitudes is shown at the left. The curve at the right indi- group called SPECMAP, later strength- cates the volume of the earth's ice sheets, determined from isotopic studies of sea- ened the case for the astronomical floor sediments. Ice volume climbs gradually for about 100,000 years and then falls theory even more when he showed abruptly in ice-age terminations that correspond to episodes of increasing summer that the amplitude of the shorter-pe- sunshine at northern latitudes. (Seasonality varies differently in the Southern Hemi- riod signals has varied exactly as one sphere, which suggests that northern seasonality must be what drives ice ages.) would expect if the signals were be- 50 SCIENTIFIC AMERICAN January 1990 90 ARCTIC OCEAN as BROOKS RANGE ALASKA 0 RANGE 60 Oo CASCADE RANGE SIERRA NEVADA 30 90 + MEXICAN VOLCANOES CENTRAL AMERICA 60 LATITUDE (DEGREES) 0 30 ANDES 0 30 30 PATAGONIA 60 60 90+ ANTARCTICA 90 0 2,000 4,000 6,000 8,000 ALTITUDE (METERS) .CE SHEETS AND MOUNTAIN GLACIERS expanded in both hemi- because the sea level was lower.) The graph traces the aver- spheres during the last ice age. The map (an unusual equal- age elevation of mountain snow lines on the American cordil- area projection) shows the extent of land ice (red) and sea ice lera, plotted along the north-south transect indicated on the (yellow) on all the continents at peak glaciation some 19,500 map. Ice-age snow lines (blue line) were about 1,000 meters years ago. (Land ice extended beyond some present coastlines lower than snow lines are today (red), regardless of latitude. SCIENTIFIC AMERICAN January 1990 51 ST. ELIAS MOUNTAINS, YUKON SOUTHERN ALPS, NEW ZEALAND 300 150 0 40 30 20 10 0 5 5 5 50 DEGREES SOUTH THOUSANDS OF YEARS AGO 10 15 20 THOUSANDS OF YEARS AGO 10 20 THOUSANDS OF YEARS AGO 10 15 15 50 DEGREES NORTH 20 25 25 25 2,000 1,500 1,000 500 0 100 80 60 40 20 0 .9 1.0 500 250 0 SOUTHERN ANDES SUMMER SUNSHINE ICE SHEET, GREAT LAKES (THOUSANDS OF CALORIES PER ICE SHEET. WASHINGTON STATE SQUARE CENTIMETER PER DAY) TIMING of glacial retreat was identical in the Northern Hemi- ery case dramatic retreat began 14,000 years ago. Changes in sphere (left) and in the Southern Hemisphere (center). The seasonal intensity could not have driven the retreat directly, be- graphs give the extent of mountain glaciers and ice sheets cause even though northern summers were getting stronger, from their source regions (in kilometers) and show that in ev- summers in the Southern Hemisphere were weakening (right). ing modulated by distance seasonality. retreated in the Southern Hemisphere des? If orbital cycles do indeed drive To be sure, there were loose ends. as well. Studies by geologists, includ- glacial cycles by acting directly on The 100,000-year variation has a much ing the late John H. Mercer of Ohio northern ice sheets, the response to weaker effect on seasonal sunshine State University and Stephen C. Porter seasonality changes in the high north- than the shorter cycles do, and yet it of the University of Washington, show ern latitudes must be strong enough apparently sets the fundamental fre- that during the last ice age, climate to override the effects of the very dif- quency of glaciation. The shorter cy- changed at the same times and by ferent changes in the Southern Hemi- cles emerge only in the wrinkles in the comparable amounts in the middle lat- sphere. One possibility is that the isotopic record. What is more, the cal- itudes of the Southern Hemisphere- northern ice sheets themselves trans- culated seasonality cycles rise and fall even though seasonality there varies late Northern Hemisphere seasonality smoothly, but the ice curve is saw- on a quite different schedule. into climatic change around the world. toothed: the ice grows episodically for They and others have found, for Two links between the northern ice nearly 100,000 years and then crash- example, that during the last ice age sheets and ice growth worldwide have es in a few thousand, in a period of the earth's mountain glaciers also ex- been proposed, but neither one bears strengthening northern summers. panded. The evidence-from the heaps up well under scrutiny. One invokes Workers have sought answers to of debris plowed up by the glaciers, sea level, which would have dropped both puzzles in the physics of the ice known as moraines-is as clear in the as the growth of the northern ice sheets and the underlying rock, which tropics (New Guinea, Hawaii, Colom- locked up much of the world's water. sinks under the weight of the ice. For bia and East Africa) and the southern Since glaciers can grow only on land, example, William R. Peltier and Wil- temperate latitudes (Chile, Tasmania the drop in sea level might have al- liam T. Hyde of the University of To- and New Zealand) as it is in north- lowed southern glaciers to expand ronto have built a theoretical model ern temperate latitudes (the Cascades, onto the exposed continental shelves that incorporates assumptions about the Alps and the Himalayas). On all even without a global change in tem- how the bedrock sinks and that close- the mountains studied so far, regard- perature. Later, when the northern ly reproduces both the dominance of less of geographic setting or precipi- ice sheets melted, the rise in sea lev- the 100,000-year cycle and the rapid tation rate, the snow line descended el might have broken up the margins retreat of the ice. In the model, it takes by about one kilometer, correspond- of the Southern Hemisphere glaciers, nearly 100,000 years for an ice sheet ing to a drop in temperature of about forcing them to retreat. The explana- to reach a critical size, at which point five degrees Celsius. tion is plausible only for Antarctica, the ductile rock below the earth's crust Where organic material was trapped however, because most mountain gla- begins to flow rapidly and allows the in the moraines, radiocarbon dating ciers do not approach the sea. burdened crust to sink. The surface shows that the glaciers advanced and The second proposal relies on the of the ice sheet drops; warmed by retreated on the same schedule. They high albedo, or reflectivity, of the the lower elevation, the ice can melt fluctuated near their maximum ex- vast northern ice sheets. By reduc- rapidly when the shorter-period cy- tent between about 19,500 and 14,- ing the absorption of sunlight by the cles bring the next episode of strong 000 years ago, about the same time as planet as a whole, the ice might have northern summers. the glaciation of northern continents led to global cooling and allowed gla- peaked. Then, just as the northern ice ciers to grow at southern latitudes. P eltier and Hyde's model, like sheets began to shrink, the mountain Yet computer climate models show many other models, assumes glaciers underwent a dramatic retreat that the albedo effects of Northern that Northern Hemisphere sea- that sharply reduced their size by Hemisphere ice sheets should be con- sonality changes drive glacial advance about 12,500 years ago. fined to northern latitudes. Also, if and retreat directly, with bedrock re- How could changes in summer sun- ice albedo does drive global climat- sponse shaping each cycle and setting shine at the latitude of Iceland have ic change, one would expect to find its length. Yet the assumption suffers caused glaciers to grow and retreat a pronounced north-to-south gradi- a crucial problem: glaciers grew and in New Zealand and the southern An- ent in the mountain-glacier record, 52 SCIENTIFIC AMERICAN January 1990 with mountains adjacent to the north- in surface waters regulates the atmos- organisms that inhabit water masses ^rn ice sheets recording the greatest pheric concentration. of specific temperature and salinity, now-line lowering and the Andes, say, Living things in turn control the sur- studied by William F. Ruddiman and showing very little change. No such face-water concentration, by acting as Andrew McIntyre of Columbia Univer- gradient is seen. a biological pump that transfers car- sity and by Detmar F. Schnitker of the Any causal link between the ice bon dioxide from the surface to the University of Maine. More recently a sheets and global climatic change ocean depths. In the course of photo- geochemical technique pioneered by also must contend with the timing synthesis, the tiny green plants of the Edward A. Boyle of the Massachusetts of the mountain-glacier retreat. Both ocean's sunlit upper layers capture Institute of Technology provided dra- the northern ice sheets and the moun- dissolved carbon dioxide to form or- matic and direct confirmation that the tain glaciers began their retreat from ganic tissue. Some of the plant matter, ocean circulated differently during the the last glacial maximum at the same as well as animal tissue nourished last glaciation. time, about 14,000 years ago. The by the plants, eventually sinks into Boyle discovered that, for unknown continental glaciers took about 7,000 the deep sea, where bacteria oxidize reasons, the distribution of cadmium years to melt away, whereas the moun- it back to carbon dioxide. Thus, the in today's oceans closely matches that tain glaciers shrank much more quick- gas is continuously pumped into the of phosphate and nitrate nutrients. Be- ly. The disparity suggests that the abyss, together with nutrients such as cause the cadmium ion has the same northern ice sheets cannot be calling phosphate and nitrate. charge and size as calcium, Boyle the tune for climate over the rest of The efficiency of this pump depends guessed that cadmium might substi- the earth. not only on the surface communi- tute for calcium in the calcium car- ty's population and species but also bonate of foraminiferal shells. If it I the ice sheets themselves can- on vertical mixing patterns. The exact does, measurements of cadmium in not link the astronomical cycles to link between pumping efficiency and shells from sediment cores might re- the climatic shifts, what can? Clues ocean circulation is controversial, but veal the distribution of nitrate and come from core samples drilled from one can imagine, for example, that if phosphate in the glacial ocean. depths of as much as two kilometers the mixing of deep waters with the Boyle's intuition proved correct in the ice that still blankets Greenland surface is slowed, surface plant life when he found that foraminifera and Antarctica. The first thing the ice will have more time to deplete the in the present-day ocean do incorpo- cores offer is confirmation of the glob- shallow water of carbon dioxide be- rate cadmium in a constant propor- al and synchronous character of ice- fore more of the gas is stirred up from tion to its abundance in seawater. He age climatic changes. the depths. During glacial time, some then measured cadmium in sediment The oxygen 18 content of glacial ice combination of altered mixing and cores. The result was exciting: a key S depleted in general, but the exact changes in ecology must have made signature of the Atlantic's present-day content records the local temperature the biological pump more efficient. circulation was missing during glacial at the time the ice was laid down. (The time, until about 14,000 years ago. colder a parcel of air becomes, the more of its water vapor is likely to T he first indications that the ice- Currently the Atlantic's deep water age ocean did operate different- contains only about half as much phos- have fallen already in precipitation, ly came from fossil evidence: phate and nitrate as the deep waters reducing the oxygen 18 content of changes in the populations of micro- of the Pacific and Indian oceans. The the remaining vapor.) Isotopic studies of the Greenland and Antarctic cores show that during the last glaciation both poles cooled-to as much as 10 degrees C below today's tempera- tures-and warmed in step. The ice also revealed something much more intriguing. Groups led by Hans Oeschger of the University of Bern and Claude Lorius of the Lab- oratory of Glaciology and Geophys- ics of the Environment, near Greno- ble, measured the carbon dioxide con- tent of the tiny bubbles of ancient air trapped in the ice. They found that during the last glaciation the carbon DEEP SALTY CURRENT dioxide concentration of the atmos- phere was about two thirds of its interglacial level. The carbon dioxide curve pointed to a missing ingredient in the climatic recipe: the ocean. Only a major shift in the ocean's operation could account for such a DEEP SALTY CURRENT threads the world's oceans, compensating for the transport of water vapor by the atmosphere. (Light blue arrows indicate shallow return flow.) The ramatic change in atmospheric com- current originates in the North Atlantic, where northward-flowing warm water that is osition. After all, the oceans hold unusually saline (and therefore dense) because of excess evaporation is chilled, 60 times as much carbon dioxide as which increases its density further. It sinks into the abyss and flows southward, out the atmosphere; because the gas read- of the Atlantic. Most of the salty water that is supplied by this Atlantic "conveyor" ily diffuses between the ocean surface mixes upward in the Pacific, making up for excess precipitation there. The Atlan- and the atmosphere, its concentration tic conveyor-and probably the entire system-was disrupted during glacial time. SCIENTIFIC AMERICAN January 1990 53 low nutrient content reflects the wa- the deep water in the world's oceans tion must have looked very different. ter's recent sojourn near the surface ultimately originates here. From its The sea and land evidence togeth- (where biological activity depletes source the water floods the deep At- er points to a simultaneous change the nutrients). Every winter at about lantic, curves around the southern tip in the operation of the ocean and the latitude of Iceland, water of rel- of Africa and joins the deep current the atmosphere 14,000 years ago. The atively high salinity, flowing north- that circles Antarctica and distributes pattern of ocean circulation shifted ward at intermediate depths (perhaps deep water to the other oceans. dramatically; glaciers in both hemi- 800 meters), rises as winds sweep As the deep water ages and travels spheres began retreating, signaling the surface waters aside. Exposed to away from the site of its formation, global warming; and the carbon diox- the chill air, the water releases heat, it collects sinking phosphate and ni- ide content of the atmosphere start- cooling from 10 degrees C to two trate, which results in a gradient of ed to rise to interglacial levels. We degrees. The water's high salinity to- increasing nutrient levels. By measur- think these events indicate a major re- gether with the drop in temperature ing the cadmium content of foraminif- organization of the joint ocean-atmos- makes it unusually dense, and it sinks era that lived near the bottom, Boyle phere system-a jump from a glacial again, this time all the way to the found that during glacial time the nu- mode of operation to an interglacial ocean bottom. trients were more uniformly distribut- mode. Indeed, we believe that abrupt The formation of the North Atlantic ed through the depths of the world's jumps among several ocean-atmos- deep water, as it is called, gives off a oceans. In addition, the concentration phere modes may underlie glacial cy- staggering amount of heat. Equal to in the glacial Atlantic peaked in the cles in general. about 30 percent of the yearly direct deepest parts rather than at interme- input of solar energy to the surface diate depths, as it does today. of the northern Atlantic, this bonus tion of the earlier microfossil studies. W e propose that changes in These results bore out the implica- seasonality are the ultimate of heat accounts for the surprisingly causes of these mode shifts. mild winters of Western Europe. (The The Atlantic "conveyor," which releas- Although we can suggest no simple warming is often mistakenly ascribed es vast quantities of heat to the North mechanisms linking seasonality, the to the Gulf Stream, which ends well to Atlantic and sends immense volumes ocean-atmosphere system and global the south.) The magnitude of the verti- of water into the abyss, was shut down climate, we can offer some insights. cal circulation is also immense, aver- until the last ice age ended 14,000 The atmosphere, which would cer- aging 20 times the combined flow of years ago. In the absence of this key tainly feel the effects of seasonal- all the world's rivers. Indeed, much of component, worldwide ocean circula- ity changes, strongly influences the circulation of the ocean. The link in- volves the distribution of salt. Prevail- ing winds transfer water evaporated from one part of the ocean to another region, where it falls as precipitation. The transport of vapor leaves a heri- tage of salt in the first region and dilutes the salinity of the second. Now, the tendency of surface waters to sink into the depths and initiate a vertical conveyor belt like that of the North Atlantic depends on their density. Density reflects both temper- ature and salinity, but salinity is the decisive factor. (Surface water cools almost to the freezing point through- out the high latitudes in winter, but only where it is unusually saline does it sink into the abyss.) The system has a built-in nonlinearity: a gradual shift in atmospheric circulation, by chang- ing salinity in regions such as the North Atlantic, could dramatically al- ter the global circulation pattern. In- deed, the Atlantic conveyor appears to be the most vulnerable part of the system, which may explain why it is Northern Hemisphere seasonality that drives global climatic changes. SEDIMENT CORE (left) from the North Atlantic testifies to an abrupt change in cir- A climatic event called the Younger culation at the end of the penultimate glaciation about 128,000 years ago. The tran- sition (identified by Gerard C. Bond of Columbia University) spans a few millime- Dryas, which took place several thou- ters and represents about 50 years. A scanning electron micrograph of çoarse ma- sand years after the glaciers started to terial from the dark sediments (bottom) reveals abundant rock fragments, rich in retreat, provides a smoking gun for silicon (blue in an X-ray Tap), presumably dropped by melting icebergs. The light- this part of our case. It vividly illus- colored sediments (top) include almost no rock and are made up mainly of shells, trates the link between the transport rich in calcium (red), from marine organisms that inhabit warm waters. (Shells in the of fresh water-in this case liquid wa- dark sediments came from cold-water species.) The sudden revival of the Atlantic ter and not vapor-and ocean circula- conveyor must have warmed the surface, eliminating icebergs and altering ecology. tion. About 11,000 years ago the re- 54 SCIENTIFIC AMERICAN January 1990 LAKE AGASSIZ ST. LAWRENCE RIVER MISSISSIPPI RIVER DIVERSION OF MELTWATER during the retreat of the North across the region of the Great Lakes to the St. Lawrence Riv- American ice sheet some 11,000 years ago may explain the er (arrow). The influx of fresh water to the North Atlantic di- 1,000-year cold spell known as the Younger Dryas. Lake Ag- luted the salinity of surface water, reducing its density and assiz, fed by meltwater, had been draining down the Mis- preventing it from sinking. The Atlantic conveyor was shut sissippi River to the Gulf of Mexico. When the retreat of the down: warm water could no longer flow northward, and a broad ice opened a channel to the east, however, the water flooded region around the North Atlantic was chilled (hatched area). treat of the glaciers was well under A massive influx of fresh water from the Gulf of Mexico record this diver- way, and temperatures had risen to the melting North American ice sheet sion. Their oxygen 18 content had their interglacial levels. Suddenly, in seems to have killed the conveyor been anomalously low, reflecting the as little as 100 years, northern Europe and precipitated the Younger Dryas. oxygen 16-rich meltwater discharging and northeastern North America re- The ice sheet started shrinking 14,000 from the Mississippi. About 11,000 verted to glacial conditions. Pollen rec- years ago; for the 7,000 years it took to years ago the isotopic ratio increased ords show that the forests that had melt away, it must have released fresh abruptly as the Lake Agassiz diversion colonized postglacial Europe gave way water at about the same rate as to- shut off the meltwater flow to the Gulf. to arctic grasses and shrubs (including day's Amazon River. At first nearly all The meltwater, meanwhile, poured the Dryas flower, for which the period the meltwater from the southern edge into the North Atlantic close to the site is named), and the Greenland ice core of the massive ice sheet flowed down of deep-water formation. There it re- records a local cooling of six degrees the Mississippi River to the Gulf of duced the salinity of surface waters C. About 1,000 years later, this cold Mexico. About 11,000 years ago, how- (and hence their density) by so much spell ended abruptly-in as little as ever, a major diversion sent meltwater that, in spite of severe winter cooling, 20 years, recent work by Willi Dans- in torrents down the St. Lawrence Riv- they could not sink into the abyss. The gaard of the University of Copenha- er to the Atlantic. conveyor belt stayed off until 1,000 gen suggests. A vast clearinghouse for meltwater, years later, when a lobe of ice ad- Boyle's cadmium measurements, to- known as Lake Agassiz, had formed vanced across the western end of the gether with the record of surface-wa- in the bedrock depression at the edge Lake Superior basin and once again ter foraminifera in the North Atlantic, of the retreating ice sheet in what is blocked the exit to the east. Lake Agas- tell what happened. Both indicators now southern Manitoba. Until 11,000 siz rose again by 40 meters, diverting return to their glacial state at the on- years ago the lake, larger than any the meltwater back down the Missis- set of the Younger Dryas. The convey- of the existing Great Lakes, had over- sippi. The conveyor belt was reactivat- or belt had shut down once again. flowed a bedrock lip to the south and ed, and Europe warmed up again. Deep-water formation had stopped, drained down the Mississippi. Then and so the warm intermediate-depth the retreat of the ice opened a channel T he Younger Dryas links freshwa- water that supplies Europe's bonus of to the east. The water level in Lake Ag- ter flow, ocean circulation and heat could no longer flow northward. assiz dropped by 40 meters as water climate-but only regional cli- The chill over the region was dispelled flowed across the region of the Great mate. Only around the North Atlan- only when the conveyor began run- Lakes and down the St. Lawrence. tic did the episode bring a sharp cool- ning again 1,000 years later. Foraminifera from surface waters of ing; elsewhere its effects were slight SCIENTIFIC AMERICAN January 1990 55 Clearly, our account of how changes in ocean-atmosphere operation could have cooled the planet is incomplete. Moreover, since we appeal to Northern Hemisphere seasonality to pace these mode shifts, we encounter the same problem faced by other theorists: Why MOUNTAIN SNOW LINES is the 100,000-year astronomical cy- cle dominant when it is the weakest of the three? Perhaps ice-sheet growth NORTH ATLANTIC DEEP-WATER has a feedback effect on atmospher- PRODUCTION ic circulation. The ocean-atmosphere system might become most suscepti- ICE SHEETS ble to a mode shift once the ice sheets reached a critical size-which might take 100,000 years. Still, much recent evidence favors DUST our basic proposal: transitions be- tween glacial and interglacial condi- tions represent jumps between two stable but very different modes of ATMOSPHERIC CARBON ocean-atmosphere operation. If the YOUNGER DRYAS DIOXIDE earth's climate system does jump be- tween quantized states, like the elec- 0 5 10 15 20 trons around an atom, all climate in- THOUSANDS OF YEARS AGO dicators should register a transition END OF THE LAST ICE AGE brought global changes, summarized here, that began simultaneously. In this regard, the evidence from the end of the last ice at the same time about 14,000 years ago even though they proceeded at different rates. The circulation of the North Atlantic shifted abruptly from glacial to intergla- age is most impressive. The warming cial conditions (with a brief relapse during the Younger Dryas cold snap) as deep- of North Atlantic surface waters, the water production resumed. At the same time, the amount of dust in the atmosphere onset of melting in the northern ice dropped and the concentration of carbon dioxide started to increase. The shifts sheets and the mountain glaciers of may have been part of a larger reorganization of the ocean and atmosphere that the Andes, the reappearance of trees warmed the planet and caused mountain glaciers and ice sheets to start retreating. in Europe and changes in plankton ecology near Antarctica and in the South China Sea-all took place be- or absent. Unlike the glaciations, the ice-age atmosphere was exceedingly tween 14,000 and 13,000 years ago. Younger Dryas affected only the trans- dusty. Dust, too, could have contribut- If the global climate system does port of heat (from low latitudes to the ed to the cooling, by reflecting sun- prove to have quantized states, clima- North Atlantic) and not the global cli- light. Unfortunately, its effect is hard tologists will have gained new insight mate. How could a change in ocean- to quantify. into the way astronomical forcing, atmosphere operation during the ice The dustiness and low methane acting mainly in high northern lati- ages have cooled the world as a whole? content of the ice-age air do suggest tudes, could transform climate world- The Greenland and Antarctic ice that the glacial mode of ocean-atmos- wide. They will also have new cause for cores suggest part of an answer. The phere operation had imposed a dry cli- concern over the earth's climatic fu- lower level of atmospheric carbon di- mate. Dust, after all, blows from areas ture. Just as 14,000 years ago the earth oxide they record for the last glaci- where vegetation is sparse, whereas was feeling the gradual forcing effect ation would certainly have contrib- methane is produced in swamps. Dry of stronger northern summers, so now uted to the cooling: carbon dioxide conditions (which are also recorded it is subject to gradual forcing as hu- is a greenhouse gas that warms the in ice-age landforms, such as sand man activity releases carbon dioxide earth's surface by trapping solar ener- dunes, and in pollen deposits) would and other greenhouse gases into the gy. Computer climate simulations sug- have had their own effect on global atmosphere. Will the climate system gest, however, that the global cooling temperatures. Temperature falls more again respond abruptly, by flipping to caused by the observed drop in carbon rapidly with increasing altitude in a an entirely new mode? dioxide would be at most two degrees drier atmosphere; hence, the drying C-less than half of what is recorded could have contributed to the depres- in the mountain glaciers. sion of mountain snow lines. FURTHER READING THE OCEAN. Wallace S. Broecker in Scien- Two other changes recorded in the ice cores must also have contributed. tific American, Vol. 249, No. 3, pages glacial level of methane. Methane, too, E ven added together, the impacts 146-160; September, 1983. Ice-age air contains only half the post- of carbon dioxide, methane, dust ICE AGES: SOLVING THE MYSTERY. John and drying may come up short in Imbrie and Katherine O. Imbrie. Har- is a greenhouse gas, although the ice- accounting for the temperature differ- vard University Press, 1986. age cooling attributable to reduced ence between the glacial and the inter- THE ROLE OF OCEAN-ATMOSPHERE REOR- methane amounts to just a few tenths glacial planet. What else could have GANIZATIONS IN GLACIAL CYCLES. Wal- of a degree. In addition, dust is about contributed? One possibility is that lace S. Broecker and George H. Denton 30 times as abundant in glacial-age in Geochimica et Cosmochima Acta, the ocean-atmosphere reorganization Vol. 53, No. 10, pages 2465-2501; Octo- ice as in more recent layers, confirm- changed the characteristics of clouds ber, 1989. ing evidence from other sites that the and made them more reflective. 56 SCIENTIFIC AMERICAN January 1990 SCIENTIFIC an Article from AMERICAN . c) 1990 BY SCIENTIFIC AMERICAN, INC. ALL RIGHTS RESERVED JANUARY, 1990 VOL. 262 NO. 1 COP Video Science Course for Children Winner of the Association for the Advancement of Science Award "Only 8 Years Old And He's Teaching Me About Science!" First Time in the U.S. on Learn what makes an airplane fly. Home Video. Not Sold in Stores! Learn how machines work. ADUENTURES SMITH SCIENCE "Look, Dad. Static electricity can actually bend water!" Learn about the powers of magnetism. Acclaimed Video Science Course for Children Gives Your Child a Critical Early Advantage © 1989 MBI A Head Start in Science for Kids learn about the fantastic powers of Children Ages 7-12. light, gravity, magnetism, inertia, electric- Early Advantage Satisfaction It's no secret: American schools are failing ity - and much more. The London Times 47 Richards Avenue guaranteed science. The National Science Founda- calls it "the most imaginative educational Norwalk, Conn. 06857 tion and the U.S. Dept. of Education have video program yet!" reported this time and again. The average Dozens of dazzling experiments and Call Toll-Free: 1-800-367-4534 demonstrations excite kids' curiosity, and European or Japanese student gets Yes! Please send me the Adventures Into 3 times more science instruction than the teach basic scientific principles. Science Home Video Course for Children. best U.S. science student! Give Your Child What I understand only VHS format is available. the Schools Aren't Providing. World Ranking of Science Student Achievement* Name Adventures Into Science includes every- PLEASE PRINT CLEARLY 1. England 5. Japan 9. Sweden thing to master basic scientific principles. Address 2. Hong Kong 6. Poland 10. Finland 3. Singapore 7. Norway 11. United States Twelve colorful video lessons on six 50- 4. Hungary 8. Australia 12. Canada minute videos that kids will watch over City/State/Zip and over. Emphasizes the "hands-on" *Source - National Science Foundation learning too absent in U.S. schools. A (Performance in physical sciences, 4th-12th grades.) special Parent's Guide includes safe ex- Signature (All orders subject to acceptance.) In tomorrow's super-tech world, science periments for you and your children to will be essential for success. U.S. kids are do at home. You'll be thrilled when your Charge each of six monthly installments of already being left behind. But now there's $27.74* to my credit card: child starts explaining science to you! Be VISA MasterCard Diners Club American Express something you can do about it. delighted with his excited progress. Turn Your TV into a Teacher. Satisfaction Guaranteed. Credit Card No. Exp. Date the first time ever, your child can start Adventures Into Science is priced afford- I prefer not to use my credit card and will learning science from the same learn-at- ably - just $26.95 per video. And there's pay by check. Enclosed is my deposit of $55.48* home video program proven in Europe. no risk: you may return the set within 15 for each set. I will pay the balance as billed in Adventures Into Science is entertaining days for a full refund. Order today! Give four monthly installments of $27.74* each. and effective. Winner of the prestigious your child a critical early advantage in a *Includes one-time $4.75 shipping/handling charge. CT residents add 8%; TN residents add 73/4% sales tax. Advancement of Science Award! competitive world. Allow 2 to 4 weeks for shipment. 490-4819 PALEOCLANOGRAPHY, VOL. 5. NO. 4. PAGES 469-477. AUGUST 1990 A SALT OSCILLATOR IN THE GLACIAL ATLANTIC? 1. THE CONCEPT Wallace S. Broecker, Gerard Bond, and Millie Klas Lamont-Doherty Geological Observatory of Columbia University, Palisades, New York Georges Bonani and Willy Wolfli Institut fur Mittelenergiephysik, ETH Honggerberg, Zurich, Switzerland Abstract. As shown by the work of Dansgaard and 50 ppm changes in the CO₂ content of the air trapped his colleagues, climate oscillations of one or so in the ice [Stauffer et al., 1984], changes in the SO₄ millennia duration punctuate much of glacial section of concentration in the ice [Finkle and Langway, 1985] the Greenland ice cores. These oscillations are and changes in 10Be content of the ice [Beer et al., characterized by 5°C air temperature changes, sev- 1984]. As the 180 signature for these events has now eralfold dust content changes and 50 ppm CO₂ been found in three Greenland ice cores (Camp changes. Both the temperature and CO₂ change are Century [Dansgaard et al., 1971], Dye 3 [Dansgaard et best explained by changes in the mode of operation of al., 1982], and Renland (W. Dansgaard personal the ocean. In this paper we provide evidence which communication, 1990)), they must be real climatic suggests that oscillations in surface water conditions of events. However, the validity of the CO2 signal as a similar duration are present in the record from a deep record of atmospheric composition remains in limbo. sea core at 50°N. Based on this finding, we suggest The reason is that similar events were not found in a that the Greenland climate changes are driven by study of an Antarctic ice core [Neftel et al., 1988]. As oscillations in the salinity of the Atlantic Ocean which discussed by Dansgaard and Oeschger [1989], this ab- modulate the strength of the Atlantic's conveyor sence could reflect either a smoothing of the record in circulation. low deposition Antarctic record or meltwater layers in the Dye 3 ice. Oeschger et al. [1984], first suggested INTRODUCTION that the Greenland events constitute jumps between two modes of operation of the climate system. One of the most puzzling features of the ice core Broecker et al. [1985] went a step further and records from Greenland is the series of large and rather postulated that Oeschger's modes involved a turning abrupt changes in oxygen isotope composition during on and off of the Atlantic's conveyor circulation sys- the last period of glaciation [Dansgaard et al., 1971, tem. 1982, 1989]. Based on the correlation between the Rather than pursuing the cause of the entire series δ¹⁸O in present-day precipitation and mean annual of oscillations, recent attention has been focused on a temperature for high-latitude stations, these changes single event, the Younger Dryas, which followed upon translate into 5°C jumps in air temperature. Correlating the heels of the last glacial termination. This event has with the oxygen isotope changes are severalfold roughly the same shape, amplitude, and duration as changes in the dust content of the ice (see Figure 1), those during the glacial period. The lure of the Younger Dryas is that in contrast to the others it is well documented in pollen records from the lands surround- Copyright 1990 ing the northern Atlantic [Watts, 1980; Rind et al., by the American Geophysical Union. 1986] and in foraminifera records from the northern Atlantic itself [Ruddiman et al., 1977; Ruddiman and Paper number 90PA01194. McIntyre, 1981]. A case has been made that the 0883-8305/90/90PA-01194 $10.00 Younger Dryas event was triggered by a diversion of 470 Broecker et al.: A Salt Oscillator in th Glacial Atlantic? level rise immediately before and immediately after the Depth m Younger Dryas was several times more rapid than that during the Younger Dryas requires a reexamination of Distance m from 250 the diversion theory for the cause of the Younger YOUNGER DRYAS bedrock Dryas. 1800 ALLEROD In this paper we provide evidence that the climatic oscillations seen in the Greenland ice core record have equivalents in the marine sediment record for the northern Atlantic and propose that these events are the 200 result of an oscillation in the salt content of the Atlantic Ocean. Further, we propose that while the Younger 1850 Dryas was likely just another of these salt oscillations, its onset and demise were paced by meltwater diver- sions. 150 EVIDENCE FROM THE SEAFLOOR 1900 While abundant evidence for the Younger Dryas event has been found in sediments from the northern Atlantic, equivalents of the series of events recorded in the glacial section of the Greenland ice cores have yet 100 to be documented in the marine record [Broecker et al., 1988b]. Working with M. Kominz on the color record 1950 in deep-sea sediments, one of us (G. Bond), demon- strated that a fine structure exists in the gray tone record extracted from photos of cores from Deep Sea Drilling Project (DSDP) site 609 at 50°N -24°W in the ALKALINE 50 eastern Atlantic (see Figure 2). As shown in Figure 3, DUST CONTENT PART OF the fine structure in this gray tone record resembles CONDUCTIVITY 2000 that in the 180 record in the Camp Century core. cm⁻¹ m o 80 SILTY ICE, >20 8(¹⁸o) 70 o -38 -36 -34 -32 % o 1 2 mg/kg Fig. 1. Oxygen isotope composition, dust content, and alkalinity versus depth for 60 the bottom 300 m of the Dye 3 Greenland ice core [Hammer et al., 1985] covering the time interval from about 8000 to at least 80,000 years ago. Each cold event (i.e., more negative δ¹⁸O values) is + accompanied by a severalfold increase in + dust content of the ice and an increase in the alkalinity of the ice. 40 meltwater from the Mississippi drainage to the St. Lawrence drainage which led to the drowning of deep- 60 40 20 O water production in the northern Atlantic [Broecker et Fig. 2. The site of DSDP core 609 al., 1988a]. Radiocarbon dating of the paleoshorelines (square) is placed on a map showing the of Lake Agassiz and of the oxygen isotope record in location of sites at which an the Gulf of Mexico sediments supports a synchroneity Neogloboquadrina pachyderma (left between the timing of the diversion of meltwater dis- coiling) abundance peak corresponding to charge and the onset of the Younger Dryas [Broecker the Younger Dryas is present (pluses) and et al., 1989]. The problem with this focus is that its is absent (open circle) in deep-sea logical extension is to postulate that each of the series cores. The sites (solid circles) at of Greenland climate jumps during the last glacial which the sedimentation rate is too slow period was caused by a diversion of meltwater. Such for the recording of the Younger Dryas an origin is quite unlikely. In fact, the recent are also shown [from Broecker et documentation by Fairbanks [1989] that the rate of sea al., 1988a]. Broecker et al.: A Salt Oscillator in the Glacial Atlantic? 471 35 18 30 m Above Base DYE 3 ICE CORE 310 m Above Base -30 CAMP CENTURY ICE CORE 40 100 -35 45 m STAGE 3 STAGE 4 STAGE 2 280 m Above Base Above Base GRAY SCALE LIGHTER -30 STAGE 5 A 50 DSDP 609 splice 30720 + 1557234080 5020 CARBON 14 DATES B o 800 750 700 650 600 550 500 450 400 350 300 250 200 150 100 50 o COMPOSITE DEPTH IN CENTIMETERS Fig. 3. Comparison of the gray scale record for DSDP site 609 from 50°N in the eastern Atlantic with the oxygen isotope records from the two Greenland ice cores [Dansgaard et al., 1971, 1982]. The radiocarbon ages were determined on hand- picked foraminifera shells by the AMS technique described by Suter et al. 1984]. The ice core record has been "stretched" to compensate for flow induced foreshortening and scaled to best match the marine record. B designates the location of termina- tion II and A the couplet on which foraminiferal and lithic fragments counts were made. The depths in the ice core are mea- sured above bed rock. Note that the upper ~1700 m (i.e., the last 8000 years) of each ice core record is not included. In order to test whether these gray tone variations shift in foram ecology suggest that the waters at this represent changes in surface water conditions, we ex- site underwent an abrupt warming which marked the amined the greater than 150 um size fraction from six termination of a major cold period (i.e., marine stage site 609 samples, using color as a guide to their posi- 6). If sedimentation was continuous, then at the tioning. The results are listed in Table 1. One pair of ambient sedimentation rate of about 6 cm/1000 years samples spans termination II which occurs at about (see Table 2), this transition from cold conditions to 780 cm depth in the composite section (see B in Figure warm conditions must have occurred in less than one 3). Samples B1 and B2 are only a few millimeters century! That this boundary is sharp was apart, one from just below and the other from just demonstrated long ago by Ruddiman et al. [1977] who above the color change which marks the termination. studied six cores from the 50°N to 54°N latitude range As can be seen in Table 1, the sample from just below in the northern Atlantic. They found that the faunal the boundary consists of roughly equal number of change associated with termination II occurred over a lithic fragments and foraminifera shells. The lithic depth range of as little as a few centimeters which fragments are dominated by quartz grains with lesser translates to a time interval as short as 1000 years. It amounts of feldspar and a few dark minerals. As these remains to be determined whether the termination II in grains are too bulky to have been carried by the wind, the 609 record has been sharpened by an erosional they were presumably dropped from melting icebergs. event or whether the record in the other cores has been The foraminifera shells in this sample are 90% blurred by bioturbation. Neogloboquadrina pachyderma (left coiling) signifying Broecker and Denton [1989] have proposed that cold water conditions. By contrast, the sample from deglaciations are triggered by reorganizations of the just above the color boundary is nearly free of lithic ocean-atmosphere system. In particular, they point to fragments and only 2% of the forams are N. pachy- a sudden resumption of conveyor circulation in the derma (left coiling). The remainder are warmer water Atlantic as a key ingredient of such reorganizations. species akin to those dwelling at this locale today. As the conveyor brings vast amounts of warm water to Both the disappearance of ice rafted debris and the the surface of the northern Atlantic its resumption 472 Broecker et al.: A Salt Oscillator in the Glacial Atlantic? TABLE 1. Analyses of Coarse Fraction (>150 µm) Sieved From Samples From ODP Site 609 (50°N, 24°W, 3.9 km) Sample Depth in Coarse* Total N pach(L) Other** Lithic N pach(L) Lithic Frags Gray Scale+ Core, Fraction, Forams, Forams, Forams, Frags, Total Forams Total Forams cm % g¹ g¹ g¹ g¹ Couplet (~40,000 years B.P.) A-1 201.0 14.4 3300 1680 1620 1430 0.51 0.42 darker A-2 202.5 14.6 3740 1350 2390 720 0.36 0.19 intermediate A-3 204.2 15.5 3170 1190 1980 850 0.38 0.27 intermediate A-4 206.5 10.8 3990 550 3400 440 0.14 0.11 lighter Termination II (~128,000 years B.P.) B-1 just above 780 9.1 3600 85 3515 80 0.02 0.02 lightest B-2 just below 780 20.6 2890 2600 290 2590 0.90 0.90 darkest *>63µm; percent by weight **Inflata, Bulloides, Glutinata, Pachyderma (R), etc. +See Figure 2. TABLE 2. AMS Radiocarbon Ages on Hand-Picked Foraminifera Shells From DSDP Core 609 (49° 53'N, 24° 14'W, 3884 m) Depth, Species Radiocarbon Corrected Accumulation Age, Age, Rate cm years years cm/10³ years 21-23 G. bulloides 5420± 80 5020 63-65 G. inflata 12380±120 11880 79-81 G. inflata 12740±140 12340 105-107 N. pachyderma 19340±220 18940 Mean 112-113 N. pachyderma 21510±220 21110 115-115 N. pachyderma 21770±220 21370 118-120 N. pachyderma 22780±340 22380 5.95 139-141 G. bulloides 25660±440 25260 153-155 N. pachyderma 29570±660 25170 174-176 G. bulloides 31120±730 30720 780 -- 128,000* 6.09 *Age of Termination II would explain not only the rapidity of the change as- suggesting a duration of 1000 years for the half cycle. sociated with termination II but also its magnitude. As listed in Table 1 and shown graphically in Figure 4, Since the meltback of the northern hemisphere ice these samples have N. pachyderma (left coiling) and sheets triggered by such a reorganization required lithic fragment to total foraminifera ratios lying be- about 7000 years for completion, icebergs must have tween those for the two samples spanning termination been discharged into the Atlantic long after the termina- II. Further, the light portion of the couplet and the tion event recorded by the site 609 color record. The dark portion of the couplet differ in the same sense as absence of lithic fragments above the color change then do the samples on either side of termination II; i.e., the suggests that the southward drifting icebergs melted darker sample is richer in N. pachyderma (left coiling) away before drifting this far to the south (i.e., 50°N). and in lithic fragments than the lighter sample. How- The remaining four samples come from the gray ever, the range of the difference is only about 40% of scale couplet located at about 2 meters depth in the core that seen across termination II (see Figure 4). Hence (age ~35,000 years; see A in Figure 3). One (A-1) is while the fine scale color changes do indeed record from the lightest portion of the couplet. Another (A-4) water temperature changes at the site of core 609 (i.e,, is from the darkest portion of the couplet. The other 50°N), these changes are not as extreme as those two (A-2 and A-3) are from the zone of gradation be- across the transition from full glacial to full interglacial tween these extremes. The four samples span 6 cm conditions. Broecker et al.: A Salt Oscillator in the Glacial Atlantic? 473 1.0 cal cores on which radiocarbon measurements were STAGE 6 conducted. In core 609, no evidence for bioturbation can be seen at termination II. While in other portions of this core burrow like structures are present, the mixing may be stochastic rather than pervasive. At LITHIC FRAGMENTS this point we cannot assess to what extent the changes TOTAL FORAMS associated with the couplets have been reduced by mixing. 0.5 Isotopic measurements (see Table 3) on foraminifera COUPLET reveal no significant change across this couplet. The 1% higher δ¹⁸O value for N. pachyderma than for Globorotalia inflata and G. bulloides suggests that the N. pachyderma grew in colder water than did the other two species. The important message from these results is that the isotopic evidence does not require 0.0 STAGE 5e any change in the temperature for either the water in 0.0 0.5 1.0 which N. pachyderma grew or that in which the subpolar species grew. Rather, the evidence suggests N. PACHYDERMA (L.C.) that only the blend changed (i.e., during the dark TOTAL FORAMS portion of the couplet more material formed in polar Fig. 4. Plot of the lithic fragment to water was incorporated and during the light portion) total foraminifera shell ratio versus the less (see Table 1). This change in blend could have N. pachyderma (left coiling) to total been accomplished by mixing (through bioturbation) foraminiferal shell ratio for the six of sediment consisting of the pure polar end-member DSDP site 609 samples listed in Table 1. deposited during a conveyor "off" portion of the cycle The higher either ratio, the colder the and sediment consisting of pure subpolar endmember surface waters. deposited during the conveyor "on" portion of the cycle. By contrast, Ruddiman and McIntyre [1981} and Duplessy et al. [1986] viewed the situation in In situations such as this, blurring created by bio- terms of migration of a "front" separating polar and turbation must be considered. Radiocarbon measure- subpolar waters. Following their hypothesis the ments demonstrate that for open ocean sites in the cycle could be more nearly sinisoidal and need not tropics, the sediment is currently being homogenized show the full glacial to interglacial amplitude. to a depth of 6-12 cm [Nozaki et al., 1977; Peng et al., Evidence presented here could also be interpreted in 1977, 1979; Berger and Killingley, 1982]. However, terms front migration on time scales shorter than the studies by Ruddiman and Glover [1972, 1982] and by duration of the couplet. During each half of the Ruddiman et al. [1980] clearly demonstrate that the couplet, the site was alternately covered by each water depth of bioturbation must be considerably less in mass. In this scenario, the dark portion of the couplet sediments from the northern Atlantic than in the tropi- represents a period when the site was covered more TABLE 3. Oxygen Isotope and Carbon-Isotope Measurements on Samples From Couplet A. Depth G. inflata G. bulloides N. pachyderma cm Oxygen Isotopic Ratio (%o) 201.0 1.69 1.94 2.93 darker 202.5 1.84 1.93 3.03 204.3 -- 1.94 3.06 206.5 2.12 1.94 2.97 lighter Carbon Isotopic Ratio (%o) 201.0 1.29 -0.28 -0.15 darker 202.5 1.04 -0.17 +0.05 204.3 -- -0.36 -0.12 206.5 0.90 -0.17 -0.01 lighter Made by C. Charles in the Lamont-Doherty Geological Observatory laboratory of R. Fairbanks. 474 Broecker et al.: A Salt Oscillator in the Glacial Atlantic? frequently by polar waters and the light portion of the ON couplet a period when the site was less frequently covered by polar waters. Until the record of these ATLANTIC CONVEYOR events has been studied in more detail in cores from OFF several latitudes, it is premature to conclude which of these hypotheses is the correct one. CONVEYOR ICE SHEET Boyle and Rosener [1990] consider this same ques- tion in connection with a study of the Cd/Ca record kept by benthic foraminifera from a deep sea core in NORTHERN SURF. WATER SSURF!SMEAN INDUCES MELTING RISE IN SALINITY the Bermuda area. They show that bioturbation could easily alter a rectangular signal to a low-amplitude sine wave. THE ATLANTIC SALT OSCILLATOR SALINITY MEAN ATLANTIC We propose here the framework of a mechanism which could create the events observed in the Greenland ice cores. It involves oscillations in the oceans thermohaline circulation. The theoretical basis for such oscillations is given in papers by Stommel, 1961; Welander, 1982, and Ruddick and Zhang, SALINITY 1989. Schnitker [1982] purports to have found NORTHERN ATL. SURF. WATER evidence for such changes based on benthic foraminifera records for cores from the northwestern TIME Atlantic. Our hypothesis concerns changes in the operation of the Atlantic's conveyor circulation. For 1000yrs ease of presentation we speak in terms of the conveyor Fig. 5. Diagrammatic presentation of the going from "on" to "off". By "on" we mean a mode salt oscillator proposed here. The of operation akin to today's. By "off" we mean a conveyor is turned off by the decrease mode in which the heat release to the atmosphere over in Atlantic salinity caused by ice sheet the northern Atlantic and the export of North Atlantic melting and by conveyor export of salt. Deep Water (NADW) to the southern ocean were This shutdown causes the salinity of high greatly reduced. It is possible that the conveyor is still latitude surface waters in the Atlantic operative in the "off" mode but runs much slower and to take a sudden drop thus stabilizing forms a more nearly closed circulation loop within the the shutdown (see second panel from top). Atlantic itself. The scale S SURF MEAN (second panel from We propose that during glacial time the Atlantic's the top) is the ratio for the salinity of conveyor circulation was modulated by periods of salt surface waters in the northern Atlantic buildup and periods of salt drain down. For such an to the mean salinity for the Atlantic as oscillator to function a mechanism must be identified to a whole. The assumption is made that change the salt content of the Atlantic on the time scale reduction in meltwater delivery which of a millenium. Such a mechanism exists. Currently, would follow closely on the heels of the the Atlantic is losing fresh water (via vapor transport) shutdown would only partially compensate at the rate of 0.35±0.10 Sv = 10⁶ m³/s [Broecker et for the salinity drop associated with the al., 1990]. If uncompensated by the exchange of more shutdown (see second panel from top). salty Atlantic waters with less salty water from outside During the next 1000 or so years the the Atlantic, the salt content of the Atlantic would rise salinity of the Atlantic as a whole (see at the rate of about 1%o per 1000 years. The density third panel from top) rises until the change created by an increase of one or so salinity conveyor reinitiates. This reinitiation units is enough to tip the balance between the con- causes the salinity of high-latitude veyor "off" and conveyor "on" situation. surface waters to undergo a sudden rise. With this in mind let us consider the hypothetical This rise would be quickly countered by cycle shown in Figure 5. We start with the conveyor the introduction of meltwater. It is off. The assumption is made that in the conveyor-off necessary to the success of this scenario mode the rate of salt export from the Atlantic is less that the compensation not be complete. than the rate of salt buildup due to vapor loss. Hence In other words, the salinity reduction in the salt content and also the density of the Atlantic wa- high-latitude surface waters due to the ter increases. Further, in the absence of the heat re- inflow of meltwater cannot be larger than leased by the conveyor to the atmosphere over the the salinity increase associated with the northern Atlantic, the ice sheets would undergo net initiation of the conveyor. Broecker et al.: A Salt Oscillator in the Glacial Atlantic? 475 growth (i.e., little meltwater would be added to the the diversion of meltwater back to the Mississippi al- Atlantic). When a critical density threshold is reached, lowed the salinity of northern Atlantic surface waters the conveyor springs into action. To reach this to achieve the value necessary for conveyor initiation. threshold, the deep water formed in the northern Hence the Younger Dryas may well have been yet an- Atlantic must have a density great enough to push its other salt oscillation with the diversions of meltwater way around the tip of Africa into the southern ocean. pacing the conveyor switches. Once in action the conveyor raises the salinity of northern Atlantic surface waters to a value much closer CONCLUSIONS to the oceanic mean than was the case in the conveyor off mode [see Manabe and Stauffer, 1988]. Because The claim by Fairbanks [1989] that the rate of of this the conveyor is self-stabilizing. However, this meltwater generation during the Younger Dryas was tendency toward stabilization would be countered by much smaller than during the warm episodes immedi- renewed meltwater input (driven by the heat released ately before and after this cold event forced us to re- by the conveyor) which would dilute the salt in the think the cause of not only the Younger Dryas but also surface waters in the region where deep waters form. of the other rapid climate changes recorded in the An assumption necessary for the success of this sce- Greenland ice cores. We propose that a salt oscillator nario is that the salinity reduction in northern surface operates in the Atlantic Ocean during times of major ice waters generated by the introduction of meltwater be cover of the adjacent lands. When the Atlantic's con- smaller than the salinity enhancement in these waters veyor system is operating, a combination of salt export created by the conveyor itself (see Broecker [1990] for and meltwater input cause the salinity and hence the discussion of this problem). In the conveyor "on" density of Atlantic waters to decrease. When the den- mode, the combination of salt export via the conveyor sity becomes sufficiently low that the deep water pro- and meltwater input would gradually reduce the duced in the northern Atlantic no longer can force its salinity of the entire Atlantic. Eventually the density way into the southern ocean, the conveyor stops. This would be reduced to the point where the conveyor stoppage greatly reduces the rate of salt export and re- would shut down. The shutdown would be self- duces the rate of meltwater input causing the salt con- stabilizing since it would cause a large drop in the tent of the Atlantic to increase until the conveyor is salinity of the surface waters of the northern Atlantic. reinitiated. The Younger Dryas event was the last in a Of course, the cessation of meltwater input which long series of these events. Its onset and demise were would accompany the shutdown would counter this likely paced by diversions in the route of meltwater stabilization. Again, it is necessary to postulate that discharge. the salinity reduction associated with the shutdown was larger than the salinity increase associated with the APPENDIX: COMPARISON OF THE CONDI- stoppage of meltwater flow. This sequence of events TIONS ASSOCIATED WITH THE CONVEYOR is summarized in the appendix. In a separate paper "ON" AND "OFF" MODES OF THE ATLANTIC (G. E. Birchfield and W.S. Broecker, A salt oscillator in the glacial Atlantic?, 2, A 'scale analysis' model, Conveyor On submitted to Paleoceanography, 1990) a mathematical representation of this oscillator will be presented. 1. Warm water rising from the thermocline to the surface of the northern Atlantic releases heat to the overlying air. This extra heat causes the ice sheets sur- IMPLICATIONS TO THE YOUNGER DRYAS rounding the Atlantic to retreat. 2. The large throughput of conveyor water main- If the climatic events recorded in the glacial sections tains the salinity of the surface waters in the northern of the Greenland ice cores are the result of a salt oscil- Atlantic at values close to the mean for the Atlantic. lator operating in the Atlantic Ocean, then the question 3. Deep water formed in the northern Atlantic car- arises whether the Younger Dryas is yet another such ries excess salt the length of this ocean into the oscillation. We suspect that this may be the case. southern ocean. During the warm interval preceding the Younger Dryas, the combination of meltwater input and salt ex- Conveyor Off port (via the conveyor) steadily reduced the salt content of the Atlantic culminating with a conveyor shutdown 1. In the absence of a strong inflow of warm water which brought on the Younger Dryas cold epoch. from the Atlantic's thermocline, surface waters of the However, since this shutdown appears to have been northern Atlantic are much cooler. This permits a net triggered by the diversion of meltwater from the growth of the ice sheets around the northern Atlantic. Mississippi to the St. Lawrence, we would have to 2. In the absence of the strong throughput of con- postulate that this sudden input of this excess fresh veyor water, the salinity of northern Atlantic surface water to the region where deep water formed crippled waters drops well below the mean for Atlantic waters. an already weakened conveyor. During the Younger 3. In the absence of the conveyor the export of salt Dryas the salt content of the Atlantic steadily increased from the Atlantic into the southern ocean is greatly re- until the conveyor reinitiated. Again, it appears that duced. 470 Broecker et al.: A Salt Oscillator in the Glacial Atlantic? Acknowledgments. Funds provided by a gift from The abrupt termination of the Younger Dryas climate Exxon Research division and by the National Science event, Nature, 339, 532-533, 1989. Foundation Climate Dynamics Program made this Dansgaard, W., and H. Oeschger, Past environmental work possible. Part of this work was funded by the long-term records from the Arctic, in The Swiss National Science Foundation. Lamont-Doherty Environmental Record in Glaciers and Ice Sheets, Geological Observatory of Columbia University edited by H. Oeschger and C. Langway, Jr., pp. contribution 4642. 287-318, John Wiley, New York, 1989. Duplessy, J.-C., M. Arnold, P. Maurice, E. Bard, J. REFERENCES Duprat, and J. Moyes, Direct dating of the oxygen- isotope record of the last deglaciation by 14C Beer, J., H. Oeschger, M. Andree, G. Bonani, M. accelerator mass spectrometry, Nature, 320, 350- Suter, W. Wolfli, and C.C. Langway, Jr., Temporal 352, 1986. variations in the ¹⁰Be concentration levels found in Fairbanks, R.G., A 17,000-year glacio-eustatic sea the Dye 3 ice core, Greenland, Ann. Glaciol,, 5, 16- level record: influence of glacial melting rates on the 17, 1984. Younger Dryas event and deep-ocean circulation, Berger, W. H., and J. S. Killingley, Box cores from Nature, 342, 637-642, 1989. the equatorial Pacific: 14C sedimentation rates and Finkle, R.C., and C.C. Langway, Jr., Global and local influences on the chemical composition of benthic mixing, Mar. Geol., 45, 93-125, 1982. Boyle, E.A., and P. Rosener, Further evidence for a snowfall at Dye 3, Greenland: The record between 10 ka B.P. and 40 ka B.P., Earth Planet. Sci. Lett., link between late Pleistocene North Atlantic surface 73, 196-206, 1985. temperatures and North Atlantic deep water produc- Hammer, C.U., H.B. Clausen, W. Dansgaard, A. tion, Global Planet. Change, in press, 1990. Broecker, W.S., and G. H. Denton, The role of Neftel, P. Kristinsdottir, and E. Johnson, ocean-atmosphere reorganizations in glacial cycles, Continuous impurity analysis along the Dye 3 deep core. in Greenland Ice Core: Geophysics, Geochim. Cosmochim. Acta, 53, 2465-2501, 1989. Broecker, W. S., D. Peteet, and D. Rind, Does the Geochemistry, and the Environment, Geophys. ocean-atmosphere have more than one stable mode Monogr. Ser., vol. 33, edited by C.C. Langway, of operation, Nature, 315, 21-25, 1985. H. Oeschger and W. Dansgaard, AGU, 90-94, Broecker, W. S., M. Andree, W. Wolfli, H. Washington, 1985. Manabe, S., and R.J. Stauffer, Two stable equilibria Oeschger, G. Bonani, J. Kennett, and D. Peteet, The chronology of the last deglaciation: Implications of a coupled ocean-coupled ocean-atmosphere model, J. Clim., 1, 841-866, 1988. to the cause of the Younger Dryas event, Paleoceanography, 3, 1-19, 1988a. Neftel, A., H. Oeschger, T. Staffelbach, and B. Broecker, W.S., M. Andree, G. Bonani, W. Wolfli, Stauffer, CO₂ record in the Byrd ice core 50,000- H. Oeschger, and M. Klas, Can the Greenland 5,000 years B.P., Nature, 331, 609-611, 1988. Nozaki, Y., J. Cochran, K. Turekian, and G. Keller, climatic jumps be identified in records from ocean and land?, Quat. Res., 30, 1-6, 1988b. Radiocarbon and 210pb distribution in submersible- Broecker, W. S., J. P. Kennett, B. P. Flower, J. taken deep-sea cores from Project FAMOUS, Earth Teller, S. Trumbore, G. Bonani, and W. Wolfli, Planet. Sci. Lett., 34, 167-173, 1977. The routing of Laurentide ice-sheet meltwater during Oeschger, H., J. Beer, U. Siegenthaler, B. Stauffer, the Younger Dryas cold event, Nature, 341, 318- W. Dansgaard, and C.C. Langway, Jr., Late-Glacial 321, 1989. climate history from ice cores, A. G. U., Geophys. Broecker, W. S., T.-H. Peng, J. Jouzel, and Gary Monogr. 29, Climate Processes and Climate Russell, The magnitude of global fresh water Sensitivity (M. Ewing Ser.), 299-306, 1984. transports of importance to ocean circulation, Clim. Peng, T.-H., W.S. Broecker, G. Kipphut, and N. Dyn., 4, 73-79, 1990. Shackleton, Benthic mixing in deep sea cores as Broecker, W.S., Salinity history of the northern determined by ¹⁴C dating and its implications Atlantic during the last deglaciation, regarding climate stratigraphy and the fate of Paleoceanography, in press, 1990. fossil fuel CO₂, in The Fate of Fossil Fuel CO2 in Dansgaard, W., S. J. Johnsen, H. B. Clausen, and C. the Ocean, edited by N. Anderson and A. Malahoff, C. Langway, Climatic record revealed by the Camp pp. 355-374, Plenum, New York, 1977. Century ice core. Late Cenozoic Glacial Ages, Peng, T.-H., W. S. Broecker, Rates of benthic mixing edited by K.K. Turekian, pp. 37-56, Hartford in deep-sea sediment as determined by radioactive Conn., Yale University Press, 1971. tracers, Quat. Res., 11, 141-149, 1979. Dansgaard, W., H.B. Clausen, N. Gundestrup, C.U. Rind, D., D. Peteet, W. Broecker, A. McIntyre, and Hammer, S.J. Johnsen, P. Kristinsdottir, and N. W. Ruddiman, The impact of cold North Atlantic sea Reeh, A new Greenland deep ice core, Science, 218, surface temperatures on climate: Implications for the 1273-1277, 1982. Younger Dryas cooling (11-10 Clim. Dyn., 1, 3- Dansgaard, W., J.W.C. White, and S.J. Johnson, 33, 1986. Broecker et al.: A Salt Oscillator in the Glacial Atlantic? 477 Ruddick, B., and L. Zhang, The mythical thermoha- during the last glaciation, Ann. Glaciol., 5, 160-164, line oscillator?, J. Mar. Res., 47, 717-746, 1989. 1984. Ruddiman, W. F., and L.K. Glover, Vertical mixing Stommel, H., Thermohaline convection witth two of ice-rafted volcanic ash in North Atlantic sedi- stable regimes of flow, Tellus, 13, 224-228, 1961. ments, Geol. Soc. Am. Bull., 83. 2817-2836, Suter, M., R. Balger, G. Bonani, H.J. Hofmann, E. 1972. Morenzoni, M. Nessi, W. Wolfli, M. Andree, J. Ruddiman, W.F., and A. McIntyre, The North Beer, and H. Oeschger. Precision measurements of Atlantic Ocean during the last deglaciation, 14C in AMS--Some results and prospects, Nucl. Palaeogeogr. Palaeoclimatol. Palaoecol., 35, 145- Instrum. Methods Physics Res., 5, 117-122, 1984. 214, 1981. Watts, W. A., Regional variation in the response of Ruddiman, W.F., and L.K. Glover, Mixing of vegetation of Lateglacial climatic events in Europe. volcanic ash zones in subpolar North Atlantic in Studies in the Late Glacial of North-West Europe, sediments, in The Ocean Floor, edited by R.A. edited by J.J. Lowe, J.M.Gray, J.E. Robinson, Scrutton and M. Talwani, pp. 37-60, John Wiley, pp. 1-22, Pergamon, New York, 1980. New York, 1982. Welander, P., A simple heat-salt oscillator, Dyn. Ruddiman, W.F., C.D. Sancetta, and M. McIntyre, Atmos. Oceans, 6, 233-242, 1982. Glacial/interglacial response rate of subpolar North Atlantic waters to climatic change: The record in ocean sediments, Philos. Trans. R. Soc. London, Ser. B, 280, 119-142, 1977. G. Bonani and W. Wolfli, Institut fur Mittelenergie- Ruddiman, W.F., G.A. Jones, T.-H. Peng, L.K. physik, ETH Honggerberg, 8093 Zurich, Switzerland. Glover, B.P. Glass and P.J. Liebertz, Tests for size G. Bond, W.S. Broecker, and M. Klas, Lamont- and shape dependency in deep-sea mixing, Sediment Doherty Geological Observatory of Columbia Univer- Geol., 25, 257-276, 1980. sity, Palisades, NY 10964. Schnitker, D., Climate variability and deep ocean circulation: Evidence from the north Atlantic, Palaeogeogr. Palaeoclimatol. Palaeoecol., 40, 213- 234, 1982. (Received February 27, 1990; Stauffer, B., H. Hofer, H. Oeschger, J. Schwander, revised May 22, 1990; and U. Siegenthaler, Atmospheric CO₂ concentration accepted May 24, 1990.) 1 for each hamburger or bowl of rice we eat. The Freons are manufactured by industry as foaming agents, refrigerants, and propel- lants. Except for the freons, the green- house gases are products of activities es- Biggest sential to human survival. If five or so billion people are to be maintained on our planet, we must continue the greenhouse Chill experiment. We are hooked. Scientists struggle to increase our un- derstanding of how the earth's environ- mental system operates in the hope that we will be able to predict at least some of by Wallace S. Broecker the coming consequences. If so, we can develop strategies to cope with the "bad" We, the inhabitants of planet Earth, are and take advantage of the "good" results performing a gigantic climate experi- of this experiment. ment. Begun by our grandparents, its re- These inquiries have recently revealed sults will be recorded by our grandchil- a piece of disquieting information. Geo- dren. The experiment involves the pro- logical studies suggest the earth's climate duction and release into the atmosphere of system resists change until pushed beyond gaseous molecules made up of three or some threshold; then it leaps into a new more atoms; the most important of these mode of operation. The situation is akin to are carbon dioxide (CO₂), methane that of a radio with automatic frequency (CH₄), and the freons (CF₃Cl and control. When the dial on such a radio is CF₂CI₂). Unlike the two-atom molecules, turned, instead of one station fading out oxygen (O₂) and nitrogen (N₂), which and the next one fading in, the radio re- make up 99 percent of our atmosphere, mains locked on one station until a thresh- these multiatom molecules have the ca- old is crossed, at which point it suddenly pacity to capture packets of outgoing radi- jumps to another station. The implication ation from the earth. Just as a blanket of this finding for future climates is clear: helps retain our body heat, these gases the effects of the greenhouse gas buildup retain the earth's heat. Hence, the result may come in sudden jumps, rather than of our experiment will be to make the gradually. Such jumps would pose great surface of our planet warmer. threats to humans and wildlife. Unfortunately, our knowledge of the Our suspicion that the earth's climate earth's climate system is still not good changes in leaps comes from the evidence enough to reliably predict the effects of recorded in deep-sea sediments and in ice. this heating on wildlife, agriculture, and a The most studied of these records is the host of other matters important to amount of heavy oxygen found in the pre- humans. We will only know the results of served shells of microscopic animals on the buildup of these "greenhouse" gases if the ocean floor. The heavy form of oxygen our learning rate greatly accelerates. in water vapor tends to be lost as atmos- In the face of such uncertainty, one pheric moisture is transported to the might ask why the experiment is not de- icecaps. The larger the icecap, the more clared dangerous to the well-being of the heavy oxygen remains behind in the sea- planet and abandoned. The reason is that water. Thus, in eras when the icecaps were the generation of greenhouse gases is not large, shelled organisms contained more an enterprise designed by scientists. heavy oxygen than they did when the Rather, it is an inescapable byproduct of icecaps were small; the shells therefore our civilization. Carbon dioxide is pro- contain a history of the ice ages. duced when coal, oil, and natural gas are The oxygen isotope record tells us that burned. When carbon atoms, which make over the last million years the polar up the bulk of these fuels, combine with icecaps have changed in a cyclic fashion, oxygen molecules from the atmosphere, going from the rather small size of the an amount of CO2 weighing roughly three current warm period to the very large size times more than the fuel burned is gener- ated. There is no feasible way to prevent this CO2 from escaping into the atmos- phere. Methane is produced by living or- ganisms. The metabolic systems of steers Published in Natural History, and the bacteria in the mud of rice pad- Vol. 96, No. 10, October, 1987, dies are methane producers. Hence. some pp. 74-82. methane will be added to the atmosphere Great Ocean Conveyor Belt ATLANTI OCEAN PACIFIC OCEAN EAN Guidand, Salty Deep Current Cold SultyDeep Current illustrations by Joe LeMonner 2 3 at the maximum of the last glaciation. The awakening came in the early 1980s More important, these fluctuations in ice when Hans Oeschger and his group at the volume have been shown to be in tune with university in Bern, Switzerland, carried periodic changes in the earth's orbit out detailed measurements of the CO₂ around the sun, generated by gravitational content of air trapped in the ice from a interactions among the objects making up deep boring made at a site in southern our solar system. Because the timing of Greenland. These measurements concen- the oxygen isotope changes (as deter- trated on a section of the core on which mined by age measurements on deep-sea earlier studies made by the Danish group sediment cores) matches what would be of Willi Dansgaard had shown repeated expected if the changes were driven by the leaps in Greenland's air temperature. To earth's changing orbit, scientists are rea- everyone's surprise, each of Dansgaard's sonably certain of the cause-and-effect jumps was accompanied by a 20 percent relationship. change in the CO2 content of the air Why do changes in the characteristics trapped in bubbles in the ice (and hence in of the earth's orbit have anything to do the CO2 content of air above Greenland at with climate? The answer is that these the time the ice formed). changes alter the earth's seasons. The rela- Eyebrows were raised by Oeschger's tive amounts of each year's sunlight re- CO2 jumps because while the tempera- ceived during the winter months, as op- ture jumps could be written off as a curios- posed to the summer months, changes in ity of Greenland, the CO2 changes could accordance with the changing orbit. Ex- not. The atmosphere's CO2 is well mixed actly how changes in the strength of the with its other gases, hence a measurement seasons drive the expansion and contrac- in Greenland typifies the entire globe. tion of the earth's polar icecaps remains a Furthermore, the changes in CO2 content matter of debate. found by the Oeschger group occurred in While the oxygen isotope record in the times as short as a few hundred years. To deep-sea sediments provided evidence bring about these changes in CO₂ requires pointing to the earth's orbital cycles as the some extraordinary change in the earth's pacemaker of glaciation, it also tended to chemical cycles, particularly those operat- lull scientists into concluding that the ing in the ocean. Scientists were therefore earth's climate responds gradually when forced to the realization that the leaps in pushed. This conclusion was drawn de- Greenland's climate were far-reaching, in- spite the realization that the response of volving the workings of the ocean as well polar icecaps to changing climate would as those of the atmosphere. have to be so sluggish that a smooth oxy- The new look at the ocean triggered by gen isotope record would be expected no the finding of the Oeschger group brought matter how abrupt the changes in environ- to the fore the potential importance of a mental conditions might be. So lulled curious tie that exists between the func- were we that other clues in paleoclimatic tioning of today's ocean and today's records that pointed to abrupt response atmosphere. This tie results in a globe- were largely disregarded. straddling ocean current that keeps north- Hear to Atmosphere Arctic Air Canada. V Shallow Warni Southward Deep. Cord CHICK Sea-to-Air Heat Transfer At the northern Atlantic limit of the ocean conveyor belt, surface waters release heat into the atmosphere. greatly moderating Europe's climate. 4 ern Europe unusually warm. Paris lies al- orates (mainly from plants) and some runs earth's orbits and our climate. When the most a full ten degrees farther north than down the rivers and back to the sea. This belt is in operation, the warmth it delivers New York, yet its mean annual tempera- cycle must exactly balance: for each mole- prevents ice from accumulating on the ture is similar to that of New York. cule that evaporates from the sea, one lands surrounding the northern Atlantic: The extra heat received by northern molecule must return to it either by pre- when the conveyor is not in operation. Europe is carried by a conveyor-belt-like cipitation on its surface or from the mouth these lands are sufficiently cold to permit ocean current. The part of the conveyor of a river. While this is true for the ocean their glaciation. If this is indeed the case, closer to the surface moves to the north; as a whole, it need not be true for each part then the orbitally induced changes in the conveyor's deeper part moves to the of the ocean. In fact, in today's world, an seasonality must somehow alter the extent south. The important point is that the wa- imbalance exists between the Atlantic and to which the water evaporating from the ter of the upper part is warm, while that of the Pacific. The Atlantic loses more water Atlantic Ocean escapes removal by the the lower part is cold. The temperature by means of evaporation than it gains by precipitation that falls on continental ar- change occurs at the northern limit of the precipitation and continental runoff. The eas whose drainage is back into the Atlan- belt (in the region around Iceland). Here, situation is reversed in the Pacific, which tic. Salt buildup is caused only by that during the winter months, water warmed receives more water as rain and runoff fraction of the water evaporating from the during its passage through the tropical than it loses by evaporation. While this Atlantic that escapes these basins and and temperate Atlantic meets air cooled imbalance is compensated for by a net falls as rain in the Pacific or on continental during its passage over frigid Canada. The flow of seawater from the Pacific to the drainage basins feeding the Pacific. meeting results in the transfer of heat Atlantic, it leaves a mark on the ocean's As we do not yet understand enough from the sea to the air. The amount of heat salt budget. Salt does not evaporate. Thus, about the rules controlling the transport of is staggering, measuring about 30 percent the transport of water vapor from the At- water vapor through the atmosphere, we of that received by the surface of the lantic to the Pacific enriches the waters of cannot say why changes in seasonality North Atlantic from the sun. The result of the North Atlantic in salt content. The cause changes in the transport of water this transfer is twofold. First, the sting of enrichment in salt must be compensated vapor from one ocean basin to another. the cold Canadian air masses is removed for by a flow of more salty water from the We can only say that compelling evidence before the air hits northern Europe. Sec- Atlantic to the Pacific. This is accom- exists in the marine-sediment record for a ond, the waters are cooled and conse- plished by the great conveyor belt: the flipping on and off of the ocean conveyor quently made more dense. The extra den- water sinking to the abyss in the northern belt. Since the most vulnerable attribute sity allows the water to sink to the abyss Atlantic carries excess salt. of the conveyor is water-vapor transport and feeds the lower part of the conveyor. The ocean conveyor system maintains from the Atlantic to the Pacific, some link Thus the ocean current acts as a pump, higher surface water temperatures in the between this transport and seasonality extracting heat from low-latitude air and northern Atlantic than in the northern Pa- seems logical. transferring it to high-latitude air. cific. Warmer waters have a higher vapor Evidence for rapid jumps in climate on The water that sinks to the bottom of pressure and lose more water to the air by the land surrounding the northern Atlan- the northern Atlantic flows down the full evaporation. Thus the rate of evaporation tic was discovered many decades ago by length of the Atlantic, around Africa, from the Atlantic is higher than that from scientists studying pollen grains preserved through the southern Indian Ocean, and the Pacific. This creates a global "still": in sediments. The record from bogs cre- finally up the Pacific Ocean. This deep water is extracted from the warm Atlantic ated during the early phases of the retreat current carries twenty times more water and transferred through the atmosphere of the icecap that covered Scandinavia than the combined world rivers. to the cool Pacific. and the British Isles during the last glacia- There is also an ocean conveyor belt in The phenomenon that maintains this tion (20,000 to 14,000 years ago) shows a the North Pacific but it runs the opposite situation is a devilish one; the circulation transition from the herbaceous shrubs of way around. Deep waters move toward pattern is self-reinforcing and hence self- the cold period back to the forests of a the north and upwell to the surface. From stabilizing. The deep current is driven by warmer period. Those early postglacial there they move toward the equator in the the extra density supplied to the waters of forests persisted for about 2,000 years and upper ocean. So in today's world, the At- the northern Atlantic through the enrich- then were suddenly replaced by shrubs lantic Ocean conveyor belt carries tropical ment of salt. The enrichment of salt is akin to those of glacial time. This intense heat for delivery to the atmosphere at high driven by the heat carried by the water cold snap lasted about 700 years and then northern latitudes, while the Pacific con- that flows northward in the upper Atlan- just as suddenly came to an end, permit- veyor belt carries cold surface waters tic. Thus we have a classic chicken and ting the forests to return. This brief rever- southward, pushing the invading warm egg situation; excess evaporation causes sion to cold conditions, which punctuated waters back toward the equator. Today's the deep current and the deep current the period of deglaciation, was named the major ocean current system thus heats the causes excess evaporation. Younger Dryas (dryas is one of the herba- lands adjacent to the northern Atlantic. The self-stabilization of this great con- ceous plants that clothed the landscape While we don't have the complete an- veyor belt is like the radio automatic fre- during the glacial time). swer to why our ocean operates in this quency control already mentioned. And Like other signs pointing to rapid cli- fashion, we do have the first principles. like that control, the mode of operation of mate change, this rather extraordinary The pattern of circulation is governed by the joint ocean-atmosphere operation will and relatively short-lived return to cold the sea's salt. To understand this we must jump if pushed too far. The evidence con- conditions was not given very high billing consider the transport of water through tained in paleoclimatic records seems to until Oeschger's group found the rapid the atmosphere. The water that evapo- be telling us that the conveyor of today's CO2 changes. It then became the focus of rates from the ocean falls eventually as attention because detailed records for ocean did not function during the glacial rain or snow. Some of this precipitation time. Hence it is tempting to conjecture many localities on the earth's surface were reaches the land and some reaches the sea. that the turning on and off of the conveyor available for the time interval when the Some precipitation that falls on land evap- constitutes an important link between the earth emerged from its last episode of 5 glaciation. Furthermore, the records from must have averaged about half the current basins of what are now Lakes Erie and those sites showed that the Younger Dryas discharge of the Northern Hemisphere's Ontario was diverted to the sea through had a distinct geographic pattern. It is a combined rivers. Where it went would the Hudson River valley. Then the water prominent feature in records from the therefore have an important impact on the entering the basins of what are present- floor of the northern Atlantic and the sur- ocean's salt cycle. day Lakes Huron and Michigan was di- rounding continental areas (Maritime Geologists studying the deposits left be- verted to the Saint Lawrence River valley Canada, Greenland, and northern Eu- hind during the retreat of the largest of the through a channel in the glacially de- rope). But it is not found in records from glacial icecaps, the one that covered Can- pressed landscape north of Lake Erie. Fi- the continental United States. ada from the Rocky Mountains to the nally, the biggest and most important of This geographic pattern has been Atlantic Ocean, have reconstructed the these diversions occurred when a lobe of shown by computer simulations of the routes taken by the meltwater released ice blocking off the eastern end of what is global climate system to be what would be from its southern margin. Their studies now Lake Superior melted, allowing wa- expected if the ocean conveyor belt were reveal that during the initial phases of ter to flow into Lake Huron and from to be turned off (and with it the enormous melting, all routes converged on the Mis- there to the Saint Lawrence. At this point amount of heat delivered to the atmos- sissippi River and hence all the meltwater all the water melting from the southern phere above the northern Atlantic). Air flowed into the Gulf of Mexico. However, margin of the Laurentide ice sheet was temperatures over the eastern fringe of starting about 11,800 years ago, those flowing eastward into the Atlantic. Nearly Canada, over Greenland, and over much routes were captured one after another by all of this water reached the Atlantic via of western Europe would plunge about eastward-leading channels that progres- the Saint Lawrence River. While the di- 15°F. By contrast, no significant change in sively opened as the ice front retreated to version of the meltwater from the Missis- temperature would be seen in the United the north. First, the water entering the sippi to the Saint Lawrence occurred in States. Hence, the geographic pattern of the Younger Dryas event points to the ocean conveyor as the causal villain. Stronger Summer Radiation Evidence from ocean-sediment cores taken in the northern Atlantic strengthens this case. Not only do the shells of the surface-dwelling plankton show a dra- matic shift back to cold-water species dur- ing the Younger Dryas interval but also, as shown by Edward Boyle of MIT, the content of the trace metal cadmium in the shells of bottom-dwelling foraminifera in- Weaker dicates that cold deep waters, akin to Cold North Atlantic Basin Temperatures those present during glacial times, re- turned during the Younger Dryas. So we get the following picture. A full- Younger Dryas fledged glaciation dominated our planet (regionally) Abrupt end of from about 20,000 to about 14,000 years glacial period ago. This corresponded to a time when the (giobally) earth's orbital cycles gave rise to reduced seasonality at high latitudes in the North- ern Hemisphere. During this time the Warm ocean was "on another station," meaning Large its conveyor belt was not in operation. Volume of Icecaps Then, as the orbits led to gradual strength- ening of seasonality, the conveyor was Sluggish melting of glacial icecaps switched back on. The heat released by the conveyor to the atmosphere over the northern Atlantic caused the great ice sheets to melt. But about 2,000 years later the conveyor came to an abrupt halt, bringing back cold conditions to the lands Small around the northern Atlantic. Then after a 30,000 20,000 10,000 Present 700-year hiatus, the conveyor belt sprang years ago years ago years ago back into action and has run steadily up to the present. Why the brief stoppage? Fluctuations in the earth's orbit, affecting how much summer sunlight the earth To understand what appears to be the receives (top graph), have expanded and contracted the polar icecap, driving the answer to this question, we must first ap- Northern Hemisphere in and out of glacial (Ice Age) episodes. Although the volume of preciate that a staggering 12-million cubic the ice changes sluggishly and gradually (bottom graph), an abrupt global warming miles of water was bound up in the conti- has marked the end of each glacial period, indicating that ocean-atmosphere nental ice sheets of the Northern Hemi- operation can change suddenly (middle). The Younger Dryas event, a 700-year return sphere at the peak of the last glaciation. to glacial conditions that interrupted the present warm episode, is thought to have been During the 5,000-year period when this caused by a sudden diversion of melting Canadian ice sheet waters from the ice melted away, the flow of meltwater Mississippi to the Saint Lawrence River. 6 North Icecap Pole TH ICA 120° ATLANTIC OCEAN EUROPE 90° 30° Climate During Younger Dryas Period Warm Region Cold Region 60% 30° The sudden, localized cooling of Europe and Maritime Canada during the Younger Dryas event indicates that the great ocean conveyor belt suddenly stopped operating. several discrete steps over a period of nately, the paleoclimatic record provides about 800 years, the largest of these diver- no clues as to how the earth's climate sions was the last, involving the meltwater system responds when warmed beyond its released from about 60 percent of the prevailing state. Over the period for which southern perimeter of the Laurentide ice our paleoenvironmental records are suffi- sheet. As dated by the radiocarbon ciently detailed to permit such reconstruc- method, it occurred about 11,000 years tions, climate has not been significantly ago. Within the margin of error of such warmer than today's. So we are pushing age determinations, this corresponds to the earth into an unknown realm. We have the time the Younger Dryas began. no way to predict how the great ocean Scientists understand why the sudden conveyor will respond, nor can we be sure diversion of a large meltwater flow from of other important elements of the system, the Mississippi River to the Saint Law- which are subject to dramatic change. rence would affect the ocean conveyor. As The computer simulations that have stated above, an essential feature of this greatly improved our ability to predict system is the densification through cool- weather have also told us some important ing of high salt content water in the north- things about the possible response to the ern Atlantic. Today, the sinking of this greenhouse buildup. But because of their water feeds a globe-encircling deep cur- basic design, these computer models can- rent. The sudden influx directly into the not tell us anything about the ocean- northern Atlantic of an amount of water atmosphere system and leaping climates. equivalent to that carried by the Amazon At present no one knows how to incorpo- River would almost certainly diminish the rate the oceans into these simulations. De- salinity of surface waters in the northern cades may pass before this can be done. Atlantic enough to disrupt the deep-water Even then I doubt if computer simulations formation process. One might say that the will offer much insight into the changes in Younger Dryas was not only a warning climate that might be triggered by the about the manner in which climate reacts greenhouse buildup. when pushed but also clearly showed that The upshot is that we must take our the coupling between the transport of greenhouse experiment more seriously. fresh water across the earth's surface and Rather than treating it as a cocktail hour the transport of salt within the sea is a curiosity, we must view it as a threat to critical element in the earth's response to human beings and wildlife that can be climate change. resolved only by serious study over many If this reading of the paleoclimatic decades. We must expand our efforts to record is correct, then we must face up to understand the operation of each of the the reality that our climatic system does units of the earth's climate system and not operate in an orderly manner. As the how they interact. Only in this way will greenhouse gases we produce build up, the our grandchildren be able to prepare ocean-atmosphere system may leap to yet wisely for the changes that are bound to be another mode of operation. Unfortu- wrought by our great experiment. 1 for each hamburger or bowl of rice we eat. The Freons are manufactured by industry as foaming agents, refrigerants, and propel- lants. Except for the freons, the green- house gases are products of activities es- Biggest sential to human survival. If five or so billion people are to be maintained on our planet, we must continue the greenhouse Chill experiment. We are hooked. Scientists struggie to increase our un- derstanding of how the earth's environ- mental system operates in the hope that we will be able to predict at least some of by Wallace S. Broecker the coming consequences. If so, we can develop strategies to cope with the "bad" We, the inhabitants of planet Earth, are and take advantage of the "good" results performing a gigantic climate experi- of this experiment. ment. Begun by our grandparents, its re- These inquiries have recently revealed sults will be recorded by our grandchil- a piece of disquieting information. Geo- dren. The experiment involves the pro- logical studies suggest the earth's climate duction and release into the atmosphere of system resists change until pushed beyond gaseous molecules made up of three or some threshoid; then it leaps into a new more atoms; the most important of these mode of operation. The situation is akin to are carbon dioxide (CO₂), methane that of a radio with automatic frequency (CH₄), and the freons (CF₃Cl and control. When the dial on such a radio is CF₂CI₂). Unlike the two-atom molecules, turned, instead of one station fading out oxygen (O₂) and nitrogen (N₂), which and the next one fading in, the radio re- make up 99 percent of our atmosphere, mains locked on one station until a thresh- these multiatom molecules have the ca- old is crossed, at which point it suddenly pacity to capture packets of outgoing radi- jumps to another station. The implication ation from the earth. Just as a blanket of this finding for future climates is clear: helps retain our body heat, these gases the effects of the greenhouse gas buildup retain the earth's heat. Hence, the result may come in sudden jumps, rather than of our experiment will be to make the gradually. Such jumps would pose great surface of our planet warmer. threats to humans and wildlife. Unfortunately, our knowledge of the Our suspicion that the earth's climate earth's climate system is still not good changes in leaps comes from the evidence enough to reliably predict the effects of recorded in deep-sea sediments and in ice. this heating on wildlife, agriculture, and a The most studied of these records is the host of other matters important to amount of heavy oxygen found in the pre- humans. We will only know the results of served shells of microscopic animals on the buildup of these "greenhouse" gases if the ocean floor. The heavy form of oxygen our learning rate greatly accelerates. in water vapor tends to be lost as atmos- In the face of such uncertainty, one pheric moisture is transported to the might ask why the experiment is not de- icecaps. The larger the icecap, the more clared dangerous to the well-being of the heavy oxygen remains behind in the sea- planet and abandoned. The reason is that water. Thus, in eras when the icecaps were the generation of greenhouse gases is not large, shelled organisms contained more an enterprise designed by scientists. heavy oxygen than they did when the Rather, it is an inescapable byproduct of icecaps were small; the shells therefore our civilization. Carbon dioxide is pro- contain a history of the ice ages. duced when coal, oil, and natural gas are The oxygen isotope record tells us that burned. When carbon atoms, which make over the last million years the polar up the bulk of these fuels, combine with icecaps have changed in a cyclic fashion, oxygen molecules from the atmosphere, going from the rather small size of the an amount of CO2 weighing roughly three current warm period to the very large size times more than the fuel burned is gener- ated. There is no feasible way to prevent this CO2 from escaping into the atmos- phere. Methane is produced by living or- ganisms. The metabolic systems of steers Published in Natural History, and the bacteria in the mud of rice pad- Vol. 96, No. 10, October, 1987, dies are methane producers. Hence. some PP. 74-82. methane will be added to the atmosphere Great Ocean Conveyor Belt ATLANTI OCEAN PACIFICOCEAN EAN Coldand Salty Deep Current Cold-und-Sulty-Deep Cold illustrations by Joe LeMonner 2 3 at the maximum of the last glaciation. The awakening came in the early 1980s More important, these fluctuations in ice when Hans Oeschger and his group at the volume have been shown to be in tune with university in Bern, Switzerland, carried periodic changes in the earth's orbit out detailed measurements of the CO2 around the sun, generated by gravitational content of air trapped in the ice from a interactions among the objects making up deep boring made at a site in southern our solar system. Because the timing of Greenland. These measurements concen- the oxygen isotope changes (as deter- trated on a section of the core on which mined by age measurements on deep-sea earlier studies made by the Danish group sediment cores) matches what would be of Willi Dansgaard had shown repeated expected if the changes were driven by the leaps in Greenland's air temperature. To earth's changing orbit, scientists are rea- everyone's surprise, each of Dansgaard's sonably certain of the cause-and-effect jumps was accompanied by a 20 percent relationship. change in the CO2 content of the air Why do changes in the characteristics trapped in bubbles in the ice (and hence in of the earth's orbit have anything to do the CO₂ content of air above Greenland at with climate? The answer is that these the time the ice formed). changes alter the earth's seasons. The rela- Eyebrows were raised by Oeschger's tive amounts of each year's sunlight re- CO2 jumps because while the tempera- ceived during the winter months, as op- ture jumps could be written off as a curios- posed to the summer months, changes in ity of Greenland, the CO2 changes could accordance with the changing orbit. Ex- not. The atmosphere's CO2 is well mixed actly how changes in the strength of the with its other gases, hence a measurement seasons drive the expansion and contrac- in Greenland typifies the entire globe. tion of the earth's polar icecaps remains a Furthermore, the changes in CO2 content matter of debate. found by the Oeschger group occurred in While the oxygen isotope record in the times as short as a few hundred years. To deep-sea sediments provided evidence bring about these changes in CO2 requires pointing to the earth's orbital cycles as the some extraordinary change in the earth's pacemaker of glaciation, it also tended to chemical cycles, particularly those operat- lull scientists into concluding that the ing in the ocean. Scientists were therefore earth's climate responds gradually when forced to the realization that the leaps in pushed. This conclusion was drawn de- Greenland's climate were far-reaching, in- spite the realization that the response of volving the workings of the ocean as well polar icecaps to changing climate would as those of the atmosphere. have to be so sluggish that a smooth oxy- The new look at the ocean triggered by gen isotope record would be expected no the finding of the Oeschger group brought matter how abrupt the changes in environ- to the fore the potential importance of a mental conditions might be. So lulled curious tie that exists between the func- were we that other clues in paleoclimatic tioning of today's ocean and today's records that pointed to abrupt response atmosphere. This tie results in a globe- were largely disregarded. straddling ocean current that keeps north- Heat to Atmosphere Arctic Air Canada. V THE Southward Deep. Cord Curren Sea-to-Air Heat Transfer At the northern Atlantic limit of the ocean conveyor belt, surface waters release heat into the atmosphere, greatly moderating Europe's climate. 4 ern Europe unusually warm. Paris lies al- orates (mainly from plants) and some runs earth's orbits and our climate. When the most a full ten degrees farther north than down the rivers and back to the sea. This belt is in operation, the warmth it delivers New York, yet its mean annual tempera- cycle must exactly balance: for each mole- prevents ice from accumulating on the ture is similar to that of New York. cule that evaporates from the sea, one lands surrounding the northern Atlantic: The extra heat received by northern molecule must return to it either by pre- when the conveyor is not in operation. Europe is carried by a conveyor-belt-like cipitation on its surface or from the mouth these lands are sufficiently cold to permit ocean current. The part of the conveyor of a river. While this is true for the ocean their glaciation. If this is indeed the case, closer to the surface moves to the north; as a whole, it need not be true for each part then the orbitally induced changes in the conveyor's deeper part moves to the of the ocean. In fact, in today's world, an seasonality must somehow alter the extent south. The important point is that the wa- imbalance exists between the Atlantic and to which the water evaporating from the ter of the upper part is warm, while that of the Pacific. The Atlantic loses more water Atlantic Ocean escapes removal by the the lower part is cold. The temperature by means of evaporation than it gains by precipitation that falls on continental ar- change occurs at the northern limit of the precipitation and continental runoff. The eas whose drainage is back into the Atlan- belt (in the region around Iceland). Here, situation is reversed in the Pacific, which tic. Salt buildup is caused only by that during the winter months, water warmed receives more water as rain and runoff fraction of the water evaporating from the during its passage through the tropical than it loses by evaporation. While this Atlantic that escapes these basins and and temperate Atlantic meets air cooled imbalance is compensated for by a net falls as rain in the Pacific or on continental during its passage over frigid Canada. The flow of seawater from the Pacific to the drainage basins feeding the Pacific. meeting results in the transfer of heat Atlantic, it leaves a mark on the ocean's As we do not yet understand enough from the sea to the air. The amount of heat salt budget. Salt does not evaporate. Thus, about the rules controlling the transport of is staggering, measuring about 30 percent the transport of water vapor from the At- water vapor through the atmosphere, we of that received by the surface of the lantic to the Pacific enriches the waters of cannot say why changes in seasonality North Atlantic from the sun. The result of the North Atlantic in salt content. The cause changes in the transport of water this transfer is twofold. First, the sting of enrichment in salt must be compensated vapor from one ocean basin to another. the cold Canadian air masses is removed for by a flow of more salty water from the We can only say that compelling evidence before the air hits northern Europe. Sec- Atlantic to the Pacific. This is accom- exists in the marine-sediment record for a ond, the waters are cooled and conse- plished by the great conveyor belt: the flipping on and off of the ocean conveyor quently made more dense. The extra den- water sinking to the abyss in the northern belt. Since the most vulnerable attribute sity allows the water to sink to the abyss Atlantic carries excess salt. of the conveyor is water-vapor transport and feeds the lower part of the conveyor. The ocean conveyor system maintains from the Atlantic to the Pacific, some link Thus the ocean current acts as a pump, higher surface water temperatures in the between this transport and seasonality extracting heat from low-latitude air and northern Atlantic than in the northern Pa- seems logical. transferring it to high-latitude air. cific. Warmer waters have a higher vapor Evidence for rapid jumps in climate on The water that sinks to the bottom of pressure and lose more water to the air by the land surrounding the northern Atlan- the northern Atlantic flows down the full evaporation. Thus the rate of evaporation tic was discovered many decades ago by length of the Atlantic, around Africa, from the Atlantic is higher than that from scientists studying pollen grains preserved through the southern Indian Ocean, and the Pacific. This creates a global "still": in sediments. The record from bogs cre- finally up the Pacific Ocean. This deep water is extracted from the warm Atlantic ated during the early phases of the retreat current carries twenty times more water and transferred through the atmosphere of the icecap that covered Scandinavia than the combined world rivers. to the cool Pacific. and the British Isles during the last glacia- There is also an ocean conveyor belt in The phenomenon that maintains this tion (20,000 to 14,000 years ago) shows a the North Pacific but it runs the opposite situation is a devilish one; the circulation transition from the herbaceous shrubs of way around. Deep waters move toward pattern is self-reinforcing and hence self- the cold period back to the forests of a the north and upwell to the surface. From stabilizing. The deep current is driven by warmer period. Those early postglacial there they move toward the equator in the the extra density supplied to the waters of forests persisted for about 2,000 years and upper ocean. So in today's world, the At- the northern Atlantic through the enrich- then were suddenly replaced by shrubs lantic Ocean conveyor belt carries tropical ment of salt. The enrichment of salt is akin to those of glacial time. This intense heat for delivery to the atmosphere at high driven by the heat carried by the water cold snap lasted about 700 years and then northern latitudes, while the Pacific con- that flows northward in the upper Atlan- just as suddenly came to an end, permit- veyor belt carries cold surface waters tic. Thus we have a classic chicken and ting the forests to return. This brief rever- southward, pushing the invading warm egg situation; excess evaporation causes sion to cold conditions, which punctuated waters back toward the equator. Today's the deep current and the deep current the period of deglaciation, was named the major ocean current system thus heats the causes excess evaporation. Younger Dryas (dryas is one of the herba- lands adjacent to the northern Atlantic. The self-stabilization of this great con- ceous plants that clothed the landscape While we don't have the complete an- veyor belt is like the radio automatic fre- during the glacial time). swer to why our ocean operates in this quency control already mentioned. And Like other signs pointing to rapid cli- fashion, we do have the first principles. like that control, the mode of operation of mate change, this rather extraordinary The pattern of circulation is governed by the joint ocean-atmosphere operation will and relatively short-lived return to cold the sea's salt. To understand this we must jump if pushed too far. The evidence con- conditions was not given very high billing consider the transport of water through tained in paleoclimatic records seems to until Oeschger's group found the rapid the atmosphere. The water that evapo- be telling us that the conveyor of today's CO2 changes. It then became the focus of rates from the ocean falls eventually as attention because detailed records for ocean did not function during the glacial rain or snow. Some of this precipitation time. Hence it is tempting to conjecture many localities on the earth's surface were reaches the land and some reaches the sea. available for the time interval when the that the turning on and off of the conveyor Some precipitation that falls on land evap- constitutes an important link between the earth emerged from its last episode of 5 glaciation. Furthermore, the records from must have averaged about half the current basins of what are now Lakes Erie and those sites showed that the Younger Dryas discharge of the Northern Hemisphere's Ontario was diverted to the sea through had a distinct geographic pattern. It is a combined rivers. Where it went would the Hudson River valley. Then the water prominent feature in records from the therefore have an important impact on the entering the basins of what are present- floor of the northern Atlantic and the sur- ocean's salt cycle. day Lakes Huron and Michigan was di- rounding continental areas (Maritime Geologists studying the deposits left be- verted to the Saint Lawrence River valley Canada, Greenland, and northern Eu- hind during the retreat of the largest of the through a channel in the glacially de- rope). But it is not found in records from glacial icecaps, the one that covered Can- pressed landscape north of Lake Erie. Fi- the continental United States. ada from the Rocky Mountains to the naily, the biggest and most important of This geographic pattern has been Atlantic Ocean, have reconstructed the these diversions occurred when a lobe of shown by computer simulations of the routes taken by the meltwater released ice blocking off the eastern end of what is global climate system to be what would be from its southern margin. Their studies now Lake Superior melted, allowing wa- expected if the ocean conveyor belt were reveal that during the initial phases of ter to flow into Lake Huron and from to be turned off (and with it the enormous melting, all routes converged on the Mis- there to the Saint Lawrence. At this point amount of heat delivered to the atmos- sissippi River and hence all the meltwater all the water melting from the southern phere above the northern Atlantic). Air flowed into the Gulf of Mexico. However, margin of the Laurentide ice sheet was temperatures over the eastern fringe of starting about 11,800 years ago, those flowing eastward into the Atlantic. Nearly Canada, over Greenland, and over much routes were captured one after another by all of this water reached the Atlantic via of western Europe would plunge about eastward-leading channels that progres- the Saint Lawrence River. While the di- 15°F. By contrast, no significant change in sively opened as the ice front retreated to version of the meltwater from the Missis- temperature would be seen in the United the north. First, the water entering the sippi to the Saint Lawrence occurred in States. Hence, the geographic pattern of the Younger Dryas event points to the ocean conveyor as the causal villain. Stronger Summer Radiation Evidence from ocean-sediment cores taken in the northern Atlantic strengthens this case. Not only do the shells of the surface-dwelling plankton show a dra- matic shift back to cold-water species dur- ing the Younger Dryas interval but also, as shown by Edward Boyle of MIT, the content of the trace metal cadmium in the shells of bottom-dwelling foraminifera in- Weaker dicates that cold deep waters, akin to Cold North Atlantic Basin Temperatures those present during glacial times, re- turned during the Younger Dryas. So we get the following picture. A full- Younger Dryas fledged glaciation dominated our planet (regionally) Abrupt end of from about 20,000 to about 14,000 years glacial period ago. This corresponded to a time when the (globally) earth's orbital cycles gave rise to reduced seasonality at high latitudes in the North- ern Hemisphere. During this time the Warm ocean was "on another station," meaning Large its conveyor belt was not in operation. Volume of Icecaps Then, as the orbits led to gradual strength- ening of seasonality, the conveyor was Sluggish melting of glacial icecaps switched back on. The heat released by the conveyor to the atmosphere over the northern Atlantic caused the great ice sheets to melt. But about 2,000 years later the conveyor came to an abrupt halt, bringing back cold conditions to the lands Small around the northern Atlantic. Then after a 30,000 20,000 10,000 Present 700-year hiatus, the conveyor belt sprang years ago years ago years ago back into action and has run steadily up to the present. Why the brief stoppage? Fluctuations in the earth's orbit, affecting how much summer sunlight the earth To understand what appears to be the receives (top graph), have expanded and contracted the polar icecap, driving the answer to this question, we must first ap- Northern Hemisphere in and out of glacial (Ice Age) episodes. Although the volume of preciate that a staggering 12-million cubic the ice changes sluggishly and gradually (bottom graph), an abrupt global warming miles of water was bound up in the conti- has marked the end of each glacial period, indicating that ocean-atmosphere nental ice sheets of the Northern Hemi- operation can change suddenly (middle). The Younger Dryas event, a 700-year return sphere at the peak of the last glaciation. to glacial conditions that interrupted the present warm episode, is thought to have been During the 5,000-year period when this caused by a sudden diversion of melting Canadian ice sheet waters from the ice melted away, the flow of meltwater Mississippi to the Saint Lawrence River. 6 OF North Icecap Pole TH ICA 120° ATLANTIC OCEAN EUROPE 90° 30° Climate During Younger Dryas Period Warm Region V. Cold Region 30 60% 30° The sudden, localized cooling of Europe and Maritime Canada during the Younger Dryas event indicates that the great ocean conveyor belt suddenly stopped operating. several discrete steps over a period of nately, the paleoclimatic record provides about 800 years, the largest of these diver- no clues as to how the earth's climate sions was the last, involving the meitwater system responds when warmed beyond its released from about 60 percent of the prevailing state. Over the period for which southern perimeter of the Laurentide ice our paleoenvironmental records are suffi- sheet. As dated by the radiocarbon ciently detailed to permit such reconstruc- method, it occurred about 11,000 years tions, climate has not been significantly ago. Within the margin of error of such warmer than today's. So we are pushing age determinations, this corresponds to the earth into an unknown realm. We have the time the Younger Dryas began. no way to predict how the great ocean Scientists understand why the sudden conveyor will respond, nor can we be sure diversion of a large meltwater flow from of other important elements of the system, the Mississippi River to the Saint Law- which are subject to dramatic change. rence would affect the ocean conveyor. As The computer simulations that have stated above, an essential feature of this greatly improved our ability to predict system is the densification through cool- weather have also told us some important ing of high salt content water in the north- things about the possible response to the ern Atlantic. Today, the sinking of this greenhouse buildup. But because of their water feeds a globe-encircling deep cur- basic design, these computer models can- rent. The sudden influx directly into the not tell us anything about the ocean- northern Atlantic of an amount of water atmosphere system and leaping climates. equivalent to that carried by the Amazon At present no one knows how to incorpo- River would almost certainly diminish the rate the oceans into these simulations. De- salinity of surface waters in the northern cades may pass before this can be done. Atlantic enough to disrupt the deep-water Even then I doubt if computer simulations formation process. One might say that the will offer much insight into the changes in Younger Dryas was not only a warning climate that might be triggered by the about the manner in which climate reacts greenhouse buildup. when pushed but also clearly showed that The upshot is that we must take our the coupling between the transport of greenhouse experiment more seriously. fresh water across the earth's surface and Rather than treating it as a cocktail hour the transport of salt within the sea is a curiosity, we must view it as a threat to critical element in the earth's response to human beings and wildlife that can be climate change. resolved only by serious study over many If this reading of the paleoclimatic decades. We must expand our efforts to record is correct, then we must face up to understand the operation of each of the the reality that our climatic system does units of the earth's climate system and not operate in an orderly manner. As the how they interact. Only in this way will greenhouse gases we produce build up, the our grandchildren be able to prepare ocean-atmosphere system may leap to yet wisely for the changes that are bound to be another mode of operation. Unfortu- wrought by our great experiment. National Aeronautics and Space Administration NASA George C. Marshall Space Flight Center Marshall Space Flight Center, Alabama 35812 AC(205)544-2121 DA01 Reply to Attn of: The Honorable Albert Gore, Jr. United States Senate Washington, DC 20510 Dear Senator Gore: This letter is a continuation of our response to your October 1990, request to Dr. Roy Spencer, a scientist in the Remote Sensing Branch of the Marshall Space Flight Center's Space Science Laboratory, to keep you informed of his monitoring of global satellite temperature data. Our initial update was provided in January 1991, reflecting temperatures through October 1990. As you can see from the results through February 1991, global temperatures are remaining above average, continuing the most recent warm period which began in late 1989. The net global trend for the last 12 years is now +0.03 C/decade, made up of a Northern Hemisphere trend of +0.09°C/decade and a Southern Hemisphere trend of -0.03°C/decade. The margin of error for the global temperature trend is +0.02°C/decade, and +0.03 C/decade for each Hemisphere. These measurements have been corrected for stratospheric influences. I appreciate the opportunity to provide you with this information, and for your interest in Dr. Spencer's work and your support of the Marshall Center and NASA. Sincerely, T. J.9.10 J. Lee Director Enclosure CC: HDQS/S/Dr. Fisk Tilford 19 = MIOIM 65016 OEO'O- = ON381 DECADAL 10064 0001 6861 8861 L861 9861 5861 +861 E861 2861 1861 086 I 676 0'1 - MMM www.m. 0'1 so. HEMISPHERE 65016 DECADAL TREND = 0.094 1006/ 0'1 - 0661 6861 886 11 2861 9861 5861 1994 E861 1992 1861 086 I 016 01 NO. HEMISPHERE 65016 DECADAL TREND = 0.032 10064 0661 6 6861 8861 /861 9861 586 1 +86 1 E86 1 2861 I 1861 080110261 0'1- Mummur 0'1 GLOBAL LETTERS TO NATURE is used instead of our atmosphere-ocean general circulation model⁴ because of the latter's large computational requirement. The simple climate/ocean model (Fig. 1) determines the sur- face temperature of the atmosphere and the temperature of the ocean as a function of depth from the ocean surface to the ocean floor. In the model, the ocean is subdivided vertically into 40 layers, with the uppermost being the mixed layer and the deeper layers each being 100 m thick. The ocean is also subdivided horizontally into a polar region where bottom water is formed, and a non-polar region where there is vertical upwelling. In the non-polar region, heat is transported upwards toward the surface by the upwelling, and downwards by physical processes whose effects are treated as equivalent to diffusion. Heat is also removed from the mixed layer in the non-polar region by transport to the polar region and downwelling towards the ocean bottom- this heat is ultimately transported upwards from the ocean floor in the non-polar region. Five quantities must be prescribed in this simple climate/ocean model: the temperature sensitivity of the climate system, characterized by the equilibrium warming induced by a CO₂ doubling. 1T2r; the vertically uniform upwel- ling velocity for the global ocean, W; the vertically uniform thermal diffusivity, K, by which all non-advective vertical heat transport in the ocean is parameterized; the depth of the oceanic mixed layer, h; and the warming of the polar ocean relative to the warming of the non-polar ocean, П. For the IPCC report we chose three values of T2 (4.5. 2.5 and 1.5 °C) and h = 70 m, = m yr⁻¹, K = 0.63 S. and П = 1.0. We chose the latter value as a result of simulations by atmospheric GCM/mixed- layer ocean models of the equilibrium climate change induced by a doubling of the atmosphere CO₂ concentration⁵. These simulations show a poleward amplification of the surface tem- perature change in the winter hemisphere, thereby suggesting Revised projection of future that П should equal 1.0. Here we prescribe П based on our recent analysis⁴ of the greenhouse warming transient time evolutions of CO₂ and CO₂ simulations with an atmosphere-ocean GCM. Examining the changes in Michael E. Schlesinger & Xingjian Jiang temperature in the polar (downwelling) and non-polar (upwel- ling) regions gives values of П that range from 0.569 for the Department of Atmospheric Sciences. University of Illinois at uppermost layer (a depth of 0-50 m) to 0.004 for the lowermost Urbana-Champaign, 105 South Gregory Avenue. Urbana. layer (a depth of 2,750-4,350 m), with a depth-averaged value Illinois 61801. USA of 0.161. The smallness of the depth-averaged value and the value of П for the deep ocean are probably due to the brevity FOR the Intergovernmental Panel on Climate Change (IPCC) of the 1 CO₂ and 2x CO₂ simulations, each being only 20 years report¹. using a simple climate/ocean model. we made projections of the greenhouse warming to 2100. Projections were made for four greenhouse-gas scenarios, whose radiative effects in 2100, expressed in terms of an equivalent amount of CO2, ranged from 2 to 5.5 times the pre-industrial CO2 concentration. The projected h global warming in 2100 for these scenarios. relative to 1990, ranged from 0.62-2.31 °C for the minimum assumed CO2-doubling tem- perature sensitivity, T2x = 1.5 °C, to 1.61-5.15 °C for the maximum sensitivity T2x = 4.5 °C. Here we broaden these projec- K tions to include a recently suggested lower sensitivity, T2x = 0.5 °C. We also revise all projections by prescribing. using the results of our analysis of simulations by a coupled atmosphere- ocean general circulation model, a lower value for a key parameter W of the simple ocean model, П. which indicates the warming of the Polar region polar ocean relative to the warming of the non-polar ocean. We find that, for any value of T₂ₓ, the atmospheric temperature increases more rapidly with time as a consequence of the reduction in П. We also find that a delay of ten years in initiating a 20-year Non-poiar region transition from the IPCC 'business-as-usual' scenario to any other IPCC scenario has only a small effect on the projected warming in 2100, regardless of the value of T2x. This indicates that the penalty for a 10-year delay is small. Our earlier analysis for the IPCC report. and this revision thereof, both use an energy-balance climate/upwelling- diffusion ocean model (Fig. 1) similar to that introduced by Hoffert et al.² and used by Hoffert and Flannery to predict FIG. 1 Schematic representation of the energy-balance climate/upwelling- CO2-induced climate change. This simple climate/ocean model diffusion ocean model. NATURE VOL 350 21 MARCH 1991 219 LETTERS TO NATURE TABLE 1 Potential warming reduction in 2100 obtained by a linear transition from IPCC scenario A to IPCC scenario B, C or D during either 1990-2010 or 2000-2020 Scenario A to B Scenario A to C Scenario A to D T2x 1990-2010 2000-2020 1990-2010 2000-2020 1990-2010 2000-2020 (°C) (°C) (°C) (°C) (°C) (°C) (°C) 0.5 0.39 0.37 (96%) 0.55 0.52 (96%) 0.64 0.61 (96%) 1.5 1.04 0.99 (95%) 1.45 1.38 (96%) 1.71 1.63 (95%) 4.5 2.31 2.18 (95%) 3.17 3.00 (95%) 3.78 3.57 (95%) long. It is therefore likely that 0.161 <0.569, with a value ity. Any estimate of T2x is further confounded by external closer to the mean of 0.365, and we accordingly choose П = 0.4. forcing mechanisms, both natural (for example, volcanic and We have calculated the temperature change from 1765 to 1990 solar-irradiance variations) and anthropogenic (for example, induced by the historical evolution of greenhouse gases (T. sulphate emissions⁸), the magnitudes of which are also Wigley, personal communication), and from 1990 to 2100 for uncertain. the four IPCC scenarios. These scenarios can be characterized Our temperature projections to 2100 for the IPCC scenarios by the year in which the radiative effect of greenhouse gases is using П = 0.4 give larger temperature increases than those using equivalent to a doubling of the pre-industrial CO2 concentration: П = 1.0, with the difference being larger for larger T2x. This A ('business-as-usual'), 2015; B, 2031; C, 2038 and D, 2100. We occurs because less heat is transferred to the deep ocean as П have performed these calculations for presumed temperature decreases, hence the warming of the upper ocean and atmos- sensitivities of T2x = 4.5, 1.5 and 0.5 °C, the first two values phere increases. For the business-as-usual scenario (A) and with being the extremes adopted by IPCC, and the latter value T2x = T(2100) T(1880) = for П = 1.0, but 7.2 °C proposed by Lindzen⁶ (and personal communication, 1990). for = 0.4. The corresponding values for T2x = are Comparing the calculated greenhouse-induced temperature 2.9 °C and 3.1 °C, whereas those for T₂ₓ = 0.5 °C are both changes for these different sensitivities with the observed global 1.1 °C. These results clearly show that the magnitude of the mean temperature changes from 1880 to 1990 (ref. 1) shows that potential greenhouse-gas-induced climate change ranges from the patterns of temperature change are not identical in shape catastrophic to minor depending on the true value of T2x for or magnitude (Fig. 2). In particular, the calculated temperature the climate system. Consequently, it is imperative to narrow the changes increase monotonically in time and attain values of range of possible values of T₂ₓ, for example, by determining 1.30, 0.61 and 0.24 °C in 1990 relative to 1880 for the three the minimum possible value below which the Earth would not sensitivities, respectively, but the observed global mean tem- have undergone glacial-interglacial cycles. This can be done by perature does not increase monotonically in time, and is ~0.5 °C model simulations of these cycles and comparison with palocli- in 1990 relative to 1880. Thus, there are significant disparities mate data. between the observed and reconstructed temperature evolutions We have calculated the temperature changes from 1990 to for both T2x = 4.5 and 0.5 °C, with a best fit being obtained 2100 for the IPCC scenarios B, C and D, in which greenhouse for T2x = 1.24 °C, a value less than the minimum temperature gas emissions are reduced. These calculations show, as did our sensitivity used in the IPCC calculations. This sensitivity is also IPCC calculations, that reducing the future increase in green- below that for zero feedback (only radiative-convective adjust- house gases reduces the future increase in temperature, with the ment of temperature without changes in other quantities such size of the reduction being larger for larger T₂ₓ. For example, as water vapour, clouds and sea ice), and the 1990-2100 temperature rise for scenario B is less than that implies a net negative feedback of f = -0.06 [AT₂ₓ/(AT₂ₓ)₀⁻ = for scenario A by 2.45, 1.08 and 0.40 °C for T2x =4.5, 1.5 and 1/(1-f)]. But as discussed by Wigley and Raper⁷, it is difficult 0.5 °C, respectively. If a linear transition from 1990 to 2010 is to estimate T2x from the observed temperature record because of the unknown contribution from the climate's natural variabil- 2.5 1.4 A 2.0 1.2 1.0 T2x 4.5 °C 1.5 Temperature change (°C) A-B 0.8 A-C 1.0 0.6 1.5 °C Temperature change (°C) D 0.4 Obs 0.5 0.2 0.5 °C 0 0 1990 2000 2020 2040 2060 2080 2100 Year -0.2 1850 1875 1900 1925 1950 1975 2000 FIG. 3 Projected greenhouse-gas-induced temperature change to 2100 for a prescribed climate sensitivity of T2x =1.5 for: IPCC scenario A (solid Year line); a linear transition from 1990 to 2010 from IPCC scenario A to either FIG. 2 Historical greenhouse-gas-induced temperature change calculated IPCC scenario B, C or D (dotted lines); and a linear transition during 2000 from 1765 to 1990 for prescribed climate sensitivities of T2x =4.5, 1.5, to 2020 from IPCC scenario A to either IPCC scenario B, C or D (dash-dotted and 0.5 °C, together with the observed temperature change (Obs). All lines). All calculations begin in 1765 and all temperature changes are relative temperature changes have been referenced to 1880. to 1990. 220 NATURE VOL 350 21 MARCH 1991 made from the emission rate of scenario A to the emission rate of scenario B, the reduction in temperature rise in 2100 is 2.31, 1.04 and 0.39 °C for T2x = 4.5. 1.5 and 0.5 °C, respectively (Fig. 3, Table 1). If the initiation of this transition in emission rate is deferred 10 years until 2000, the reduction in temperature rise in 2100 is 2.18, 0.99 and 0.37 °C; that is, at least 95% of that possible by not deferring the transition. The equivalent atmos- phere CO₂ concentration in 2100 increases as a consequence of this deferral, thereby increasing the equilibrium temperature change (the 'committed warming, approximately equal to the increase in temperature change shown in Fig. 3) by 0.15, 0.05 and 0.02 °C for T2, = 4.5, 1.5 and 0.5 °C, respectively. Corre- sponding results are obtained for 20-year linear transitions from the emission rate of scenario A to the emission rate of either scenario C or scenario D (Fig. 3). with the reduction in tem- perature rise in 2100 equal to at least 95% of that possible by not deferring the transition, regardless of the temperature sensi- tivity (Table 1). The corresponding increases in the committed warming in 2100 for a 10-year deferral in transitions from scenario A to C(D) are 0.22(0.28). 0.07(0.09) and 0.02(0.03)°C for T2x = 4.5, 1.5 and 0.5 °C, °C, respectively. This indicates that the penalty is small for a 10-year delay in initiating the transition to a regime in which greenhouse-gas emissions are reduced. To us this small penalty does not indicate that we should 'wait and see' and do nothing during this decade-quite the contrary. The study of the greenhouse effect, both theoretically and observationally, should be accelerated into a 'crash pro- gramme' so that we do not squander the time that nature has given us to obtain a realistic understanding of the climate response to increasing concentrations of greenhouse gases. Received 12 November 1990: accepted 28 January 1991 1. Houghton. T.. Jenkins. G.J. & Ephraums. J. J. (eds) Climate Change: The IPCC Scientific Assessment (Cambridge University Press. 1990). 2. Hoffert. M. I., Callegan. A J. & Hsieh. C.-T. J. geophys Res. 85, 6667-6679 (1980). 3. Hoffert. M.I. & Flannery. B. P in Projecting the Climatic Effects of Increasing Carbon Dioxide (eds MacCracken, M.C. & Luther F 1.).).190(US Department of Energy. Washington, DC 1985). 4. Schiesinger. M. E. & Jiang. X. J. Clim. 3. 1297-1315 (1990). 5. Schiesinger. M. E. & Mitchell. J F B. Rev. Geophvs 25. 760-798 (1987). 6. Lindzen. R. S. Bull Am. meteorol. Soc. 71. 288-299 (1990). 7. Wigley. T.M.L. & Raper. S.C. B. in Greennouse-gas-inouced Climatic Change A Critical Appraisal of Simulations and Observations (ed. Schlesinger. M E 471-482 (Elsevier. Amsterdam. 1991) 8. Charlson. R. J.. Langner. J & Rodhe. H. Nature 348. 22 (1990). ACKNOWLEDGEMENTS This research was supported by the US NSF and the US Department of Energy Carbon Dioxide Research Program. Office of Health and Environmental Research