Abstract
The performance of the Twentieth-Century Reanalysis (20CR) in reproducing observed monthly mean precipitation over the global domain is compared to that of comprehensive reanalyses that also assimilate upper-air and satellite observations [the Climate Forecast System Reanalysis (CFSR), ECMWF Interim Re-Analysis (ERA-Interim), and NCEP–U.S. Department of Energy reanalysis (NCEP2)] and to that of an atmospheric general circulation model (GCM) ensemble simulation [Global Ocean Global Atmosphere (GOGA)] that is forced with observed sea surface temperature (SST). Wintertime rainfall variability in the midlatitude continents and storm tracks is captured with great accuracy, similar to the comprehensive reanalyses, but summertime rainfall is not, probably in consequence of the greater importance of convection in the summer season. Over the tropics, the accuracy of all reanalyses is much less than over the midlatitudes. Over tropical land, the performance of 20CR is better than NCEP2 and similar to ERA-Interim and CFSR, but over the tropical oceans the most recent reanalyses perform significantly better. Across the twentieth century, the clearest gain from the assimilation of a denser observational dataset is the expansion of the area of good skill. A comparison of the accuracy and ensemble spread in the 20CR and GOGA ensembles highlights regions where SST forcing is a stronger source of skill than data assimilation for 20CR. In contrast, for some tropical regions such as the Sahel, the assimilation of sea level pressure is effective in constraining precipitation values—but model biases in the teleconnections with global SST limit the performance of 20CR.
