Climate models can’t get precipitation right

Published January 1, 2001

Apparently, the hotly awaited U.S. National Assessment of the Potential Consequences of Climate Variability and Change for the Nation is not worth the CO2 growth-enhanced paper it’s printed on. That’s the assessment of a study recently released by the George C. Marshall Institute.

The study’s author, climatologist David Legates of the University of Delaware, challenges the conclusions of the National Assessment, pointing out that General Circulation Models (GCMs) used as the basis for the predictions are incapable of accurately producing the information needed to assess regional climate impacts.

Because of the mathematical complexity of these models, he notes, an error in the estimation of one parameter can have repercussions throughout the modeled climate system. For instance, an error in accounting for clouds’ absorption of radiation can ultimately impact not only surface temperature, but also cloud cover, air pressure, winds, and any of a number of other weather parameters.

Using precipitation as an example, Legates states that because precipitation depends on so many factors, modeling it correctly makes it likely that the model also properly accounts for other important features. By the same token, though, errors in precipitation modeling affect things such as soil moisture, winds, and even air temperature.

The two models the National Assessment uses, one from the Canadian Climate Centre and the other from Britain’s Hadley Centre for Climate Prediction and Research, differ substantially in their forecasts of regional temperature changes across the United States. They also exhibit big differences in precipitation variability, both between each other and relative to the observations. The Marshall Institute report concludes that state-of-the-art climate models are flawed to such an extent they cannot be used as the basis for public policy.

Further proof of GCMs’ difficulty with precipitation was recently provided by NOAA’s Arnold Gruber and three coauthors in a paper published in the Bulletin of the American Meteorological Society. Although they focused on a comparison of two data sets in which precipitation measured by rain gauges was merged with precipitation estimated from satellites, they also did a comparison of one of these data sets with GCM estimates of precipitation over the tropics.

The results do not bode well for the GCMs. Gruber plotted the monthly variations of a statistic called the “anomaly correlation” (AC) from 20N to 20S. An anomaly map shows departures at each grid point from the long-term mean value of precipitation (for each month). The AC statistic relates the pattern of anomalies in the observations for each month to the pattern predicted by the model. A perfect comparison garners an AC of +1. If the two fields are statistically unrelated, the AC is zero.

The AC is less than 0.2 (for both raw and smoothed anomaly fields). In effect, that means this GCM cannot reproduce observed rainfall over the tropics.

Masters of understatement, the authors write simply that the “model simulation itself is still far from perfect.”


Gruber, A. et al., 2000. The comparison of two merged rain gauge-satellite precipitation datasets. Bulletin of the American Meteorological Society, 81, 2631-2644.

Legates, David R., 2000. A Layman’s Guide to the General Circulation Models Used in the National Assessment, George Marshall Institute.