Assimilation of SSM/I-Derived Surface Rainfall and Total Precipitable Water for Improving the GEOS Analysis for Climate Studies

Abstract

This article describes a variational framework for assimilating the SSM/I-derived surface rain rate and total precipitable water (TPW) and examines their impact on the analysis produced by the Goddard Earth Observing System (GEOS) Data Assimilation System (DAS). The SSM/I observations consist of tropical rain rates retrieved using the Goddard Profiling Algorithm and tropical TPW estimates produced by Wentz. In a series of assimilation experiments for December 1992, results show that the SSM/I-derived rain rate, despite current uncertainty in its intensity, is better than the model-generated precipitation. Assimilating rainfall data improves cloud distributions and the cloudy-sky radiation, while assimilating TPW data reduces a moisture bias in the lower troposphere to improve the clear-sky radiation. Together, the two data types reduce the monthly mean spatial bias by 46% and the error standard deviation by 26% in the outgoing longwave radiation (OLR) averaged over the Tropics, as compared with the NOAA OLR observation product. The improved cloud distribution, in turn, improves the solar radiation at the surface. There is also evidence that the latent heating change associated with the improved precipitation improves the large-scale circulation in the Tropics. This is inferred from a comparison of the clear-sky brightness temperatures for TIROS Operational Vertical Sounder channel 12 computed from the GEOS analyses with the observed values, suggesting that rainfall assimilation reduces a prevailing moist bias in the upper-tropospheric humidity in the GEOS system through enhanced subsidence between the major convective centers. This work shows that assimilation of satellite-derived precipitation and TPW can reduce state-dependent systematic errors in the OLR, clouds, surface radiation, and the large-scale circulation in the assimilated dataset. The improved analysis also leads to better short-range forecasts, but the impact is modest compared with improvements in the time-averaged signals in the analysis. The study shows that, in the presence of biases and other errors of the forecast model, it is possible to improve the time-averaged ?climate content? in the data without comparable improvements in forecast. The full impact of these data types on the analysis cannot be measured solely in terms of forecast skills.