Abstract
Algorithms for deriving estimates of monthly mean rainfall from infrared satellite observations are described and a classification scheme based on the amount of spatial and temporal information used is discussed. Estimates from a simple IR‐only algorithm averaged over the period 1986–92 are compared with two long‐term averages (climatologies). While the intensity and location of mean annual features are quite similar in all of these, the estimates are found to be systematically higher over land. Over the oceans, the differences between the estimates and the other data sets are similar in magnitude to the differences between the climatologies themselves. Tropical Pacific precipitation exhibited large interannual changes during the period, with greater/lesser amounts near the dateline during the warm/cold ENSO episodes that occurred during the period. The estimated tropical rainfall during this period exhibited most of the features found in historical studies based on rain gauge observations. Comparisons of the monthly estimates to observations of rainfall over the U.S. indicate that the large positive bias of the GPI has a systematic dependence on latitude and season, with largest values found during winter and to the north. Removal of this bias leaves residual errors of < 50% in monthly estimated rainfall over the eastern half of the country.
