Evidence for Large Decadal Variability in the Tropical Mean Radiative Energy Budget

See allHide authors and affiliations

Science  01 Feb 2002:
Vol. 295, Issue 5556, pp. 841-844
DOI: 10.1126/science.1065837


It is widely assumed that variations in Earth's radiative energy budget at large time and space scales are small. We present new evidence from a compilation of over two decades of accurate satellite data that the top-of-atmosphere (TOA) tropical radiative energy budget is much more dynamic and variable than previously thought. Results indicate that the radiation budget changes are caused by changes in tropical mean cloudiness. The results of several current climate model simulations fail to predict this large observed variation in tropical energy budget. The missing variability in the models highlights the critical need to improve cloud modeling in the tropics so that prediction of tropical climate on interannual and decadal time scales can be improved.

Earth's climate system is driven by a radiative energy balance between the solar or shortwave (SW) radiation absorbed by Earth and the thermal infrared or longwave (LW) radiation emitted back to space. The balance both modifies and is modified by the components of the Earth-atmosphere system such as clouds, the surface, and the atmosphere (1). Therefore, the TOA radiation budget is crucial in determining climate variability and feedbacks, whereas its measurement provides a severe test of our ability to represent physical processes important for simulations of future climate.

A new set of Earth radiation balance data is now being provided by the NASA CERES (Clouds and the Earth's Radiant Energy System) instrument on the Tropical Rainfall Measuring Mission (TRMM) for 8 months in 1998 and by the Terra satellite mission that began in March 2000 and is expected to continue through 2007 (2, 3). In addition, with the 16-year record of the Earth Radiation Budget Satellite (ERBS), it is now possible to examine 22 years of accurate satellite observed broadband radiative fluxes (4–6). Anomalies in tropical mean broadband thermal LW flux emitted by the Earth have been determined from this 22-year record and are shown in Fig. 1.

Figure 1

Satellite record of tropical mean (20°S to 20°N latitude) anomalies in broadband thermal emitted LW flux. Anomalies are referenced to the ERBS scanner baseline period of 1985 through 1989, which most climate models use as a baseline for comparison (5). Results are shown from seven different broadband instruments on six spacecraft missions (6).

Calibration absolute accuracies for the seven broadband radiometers inFig. 1 are estimated to be roughly 1% for the pre-CERES instruments and 0.5% for the newer records. The anomaly results during data overlap periods from all instruments are consistent to within about 1 Wm−2 or about 0.5% of the 253 Wm−2 ERBE tropical mean LW flux. Only stability of calibration enters the anomaly record for the ERBS nonscanner (NS) data, whereas both stability and absolute calibration affect the scanner (SC) anomaly records from five different instruments.

The LW anomaly record in Fig. 1 shows much larger variations than expected, especially for the 1997/1998 El Niño event, which reaches a tropical mean anomaly of 8 Wm−2, the largest seen to date. Other notable short-term anomalies are the rapid drop in LW flux resulting from the Mount Pinatubo Eruption in 1991 (7), followed by the expected 2-year recovery period as the volcanic aerosols are removed from the stratosphere and upper troposphere. Although the 1983 El Niño event is thought to be comparable in magnitude to the 1997/1998 El Niño, there is no comparable tropical mean LW flux anomaly. A plausible hypothesis is that the earlier 1982 El Chichon eruption caused a reduction in LW flux similar to that from the Mount Pinatubo Eruption in 1991. The resulting LW fluxes in 1982 to 1984 would, in this case, be a partial cancellation of the El Chichon and El Niño signals. But the most surprising result in the figure is an apparent drop of about 2 Wm−2 in LW flux from the late 1970s to the mid 1980s, followed by a rise of about 4 Wm−2 from the late 1980s to the mid to late 1990s. Indeed, if the Pinatubo volcanic reduction from 1991 to 1993 is removed, the tropics appear to undergo an increase in LW flux from 1989 to 1994, followed by a relatively steady value from 1994 through 2001, with the exception of the large 1997/1998 El Niño event.

Because radiative forcings of 1 Wm−2 or less are important for climate change prediction, natural variability of 4 Wm−2 in the LW part of the tropical radiation budget is considered a major change. The reality of these large changes is supported by the consistency of the results from seven independent broadband radiation instruments, all supporting the same pattern of decadal variability. Though there are two narrow spectral band radiometer sources of LW estimates, these are not considered as accurate as the broadband LW fluxes. The latest versions of these narrowband data sets disagree with each other as well as with the broadband data when used to determine tropical decadal changes in LW flux (8). Because the ERBS NS record spans the entire period and includes regular solar constant measurements by both SW and Total channels, we re-examined this record extensively for any potential problems with calibration and found none at levels higher than 0.2 to 0.5 Wm−2 (9). Therefore, we conclude that it is unlikely that this decadal variability can be explained by instrument calibration changes.

Is this decadal increase of 4 Wm−2 a signal of global warming? Certainly, it is not a direct one. The flux changes are far too large to be explained by the small surface and atmosphere warming over this time period, which will tend to be offset by increased CO2 and water vapor greenhouse gas trapping. This can be demonstrated with the use of CERES scanner data in 1998 and the ERBE scanner data in 1985–1989 used in Fig. 1. These data have sufficiently small fields of view (10 and 40 km, respectively) to allow separation of the data into clear-sky and cloudy-sky regions of the tropics. Recent analysis of the clear-sky scanner LW flux data (10) showed that changes during and after the 1998 El Niño were consistent with the observed changes in sea surface temperature (SST) and water vapor. By the end of the El Niño in July 1998, the clear-sky LW fluxes measured by CERES agreed with the ERBE climatological July 1985–1989 average values to within 0.5 Wm−2. The largest anomaly of clear-sky LW fluxes reached only 2 Wm−2 during the El Niño peak of February and March 1998, whereas the total anomaly in Fig. 1 reaches about 8 Wm−2. This leaves the factor of 4 to 8 larger anomalies shown in Fig. 1 to be explained by changes in the cloudy-sky conditions in the tropics.

Though 8 Wm−2 is large for a clear-sky radiation anomaly, clouds can cause changes of up to 200 Wm−2 or larger in both LW emitted and diurnally averaged SW reflected radiation fields for deep, thick clouds in the tropics. The LW-emitted flux changes are caused by cloud increases in greenhouse trapping, whereas SW-reflected flux changes are caused by cloud increases in reflected solar energy. The net radiative effect of clouds is a balance between the LW greenhouse and SW cooling effects.

Figure 2 shows the observed broadband anomalies for LW-emitted, SW-reflected, and net radiation flux changes in the tropics. Aerosols from the Mount Pinatubo eruption cause the large increase in SW-reflected flux in 1991 to 1993. As tropical mean LW-emitted flux increases in the mid to late 1990s, SW-reflected flux decreases. Both effects are consistent with a decrease in tropical cloudiness. Figure 2 also shows evidence of increased seasonal variability. To more clearly isolate these variations in Fig. 2 from El Niño and Mount Pinatubo signals,Fig. 3 shows the average tropical mean LW flux, SW flux, and SW albedo anomaly averaged separately for each seasonal month over the 4-year period 1994 through 1997, which contains the large semi-annual anomalies in SW and Net fluxes. The results show a decrease of more than 0.01 in tropical mean albedo in the spring and fall seasons, but little change in the summer and winter seasons, indicating a changed phasing of seasonal cloudiness in the tropics (11). Because Fig. 3 shows that the LW flux increase in 1994 through 1997 is almost constant with season, we further conclude that the seasonal cycle in albedo is likely dominated by changes in low-level cloudiness that have little effect on the LW fluxes but a large impact on the SW fluxes.

Figure 2

Satellite record of tropical mean (20°S to 20°N latitude) anomalies in broadband thermal emitted LW flux, solar reflected SW flux, and net radiative flux. Net flux is defined as solar insolation – SW-reflected flux – LW-emitted flux. A smaller set of satellites is shown for SW- and net-radiative flux anomalies. Only those satellites whose orbits systematically sample the entire diurnal cycle in about a month, so the large diurnal cycle of solar reflected fluxes can be accurately determined across the entire tropics (25).

Figure 3

Satellite record of the tropical mean (20°S to 20°N latitude) seasonal cycle of LW thermal emitted flux anomaly (top panel), SW flux anomaly (middle panel), and SW albedo anomaly (lower panel). Albedo is the fraction of solar radiation incident on Earth that is reflected back to space. The anomalies are averaged for the 1994 to 1997 period, which shows a large semi-annual cycle in the SW and net fluxes in Fig. 2. Anomaly results for the same 1994 to 1997 period from the climate model runs are shown as a dashed line (average of all models) and gray shading (total range of model results). Model runs are the same ones used in Fig. 4.

The above analysis indicates strong evidence for decadal variations in the radiative balance components in the tropical atmosphere. These changes are sufficiently large that, in principle, they should be seen in climate model predictions. We have tested this hypothesis by analyzing decadal integrations of the atmospheric model components of four climate models and one weather assimilation model, forced by the observed SSTs (12) for the 1985 through 1998 period. The integrations ignore the effects of volcanic aerosols. The climate models include the Hadley Centre atmospheric climate model HadAM3 (13, 14), the National Center for Atmospheric Research (NCAR) model CCM3 (15), the Geophysical Fluid Dynamics Laboratory (GFDL) Climate Model (16), and the GFDL EP (Experimental Prediction) model (17). We also included the National Center for Environmental Prediction (NCEP)–NCAR 50-Year Reanalysis, which uses the NCEP 4-D Assimilation Model (18). For all model runs, the tropical mean anomalies were calculated as in the satellite data, using the 1985 through 1989 period as the baseline.

Figure 4 compares the atmospheric model results to the ERBS NS satellite observations presented in Fig. 2. There is remarkably little variation in the tropical mean fluxes from the models when compared to the data. Even near the peak of the 1997/1998 El Niño event, in early 1998, the tropical mean model response is only about one-third that of the observations.

Figure 4

Comparison of the observed broadband LW and SW flux anomalies for the tropics with climate model simulations using observed SST records. The models are not given volcanic aerosols, so they should not be expected to show the Mount Pinatubo eruption effects in mid-1991 through mid-1993. The dashed line shows the mean of all five models, and the gray band shows the total range of model anomalies (maximum to minimum).

No significant decadal variability is exhibited by the climate and reanalysis models. Correlation coefficients of the observed and modeled tropical mean LW flux anomalies are significant at the 95% level only for the Hadley Centre model at 0.6 and for the NCEP Reanalysis model at 0.3 (19). None of the models show significant correlations with the observed SW flux anomalies.

The seasonal model anomalies for 1993 through 1997 are shown for comparison in Fig. 3, averaged consistently with the observations. The models miss the semi-annual cycle in the SW flux anomalies after 1993, i.e., they are missing the observed seasonal cycle change in tropical albedo shown in Fig. 3.

We conclude that the large decadal variability of the LW and SW radiative fluxes shown in Figs. 1 through 3 appear to be caused by changes in both the annual average and seasonal tropical cloudiness. In general, these changes are not well predicted by current climate models, or by the NCEP Reanalysis. Indeed, current assessments (1) of global climate change have found clouds to be one of the weakest components in climate models. This leads to a threefold uncertainty in the predictions of the possible global warming over the next century. Though the models represent reasonably well the large regional shifts of convective cloudiness during an El Niño event, the current results indicate that the models are struggling to produce the more subtle, but still large, decadal changes seen in the radiation data. Three potential reasons for the disagreement are as follows: (i) The observed cloud and radiation changes are forced by SST changes, but the clouds in the models do not respond correctly to the forcing. Note that the SST forcing includes both global change as well as natural decadal variability such as the Arctic Oscillation and the Pacific Decadal Oscillation. (ii) The radiation budget and cloud fluctuations are forced by changes in the climate system other than SST. (iii) The radiation budget and cloud fluctuations are an unforced natural variability.

Independent evidence for decadal tropical change includes an observed increase in tropical mean temperature lapse rate, which is also not reproduced by climate models (20). The recent IPCC Climate Change report (1) indicates an increase in lapse rate beginning about 1991 [fig. 2.12 in (1)], coinciding with the rise in LW fluxes in Fig. 1. Our results are qualitatively consistent in that an increased LW flux will primarily increase atmospheric cooling, whereas a decrease in reflected solar will increase surface heating.

However, we caution against interpreting the decadal variability as evidence of greenhouse gas warming. Whether the changes seen in the radiative balance in the last two decades are the result of natural variability or are a response to global change remains to be determined. A major step in understanding these changes is given in a companion paper in this issue (21), which offers a hypothesis for the link between these radiative balance changes and corresponding changes in the dynamical climate system, a system that appears to be much more variable than previously thought.

  • * To whom correspondence should be addressed. E-mail: b.a.wielicki{at}


View Abstract

Navigate This Article