On the relationship between the GOES precipitation index and ISCCP data set variables

The GOES Precipitation Index (GPI) is used for global, monthly rainfall est imation in the Global Precipitation Climatology Project(GPCP). Previous wor k has identified the existence of locally and seasonally varying bias in th e GPI estimates. Most sources of bias involve cloud properties, as the GPI method uses the fraction of pixels with infrared cloud temperature below 23 5 degrees K to estimate monthly rainfall totals averaged over 2.5 degrees x 2.5 degrees latitude/longitude grid boxes. In this work, the bias in the G PI is compared to cloud variables derived by the International Satellite Cl oud Climatology Project (ISCCP). ISCCP data are used as predictor variables in regression models with the GPI estimation error as the dependent variab le. The GPI estimation error is calculated using the global min gage analys is produced by the GPCP for those locations where the rain gage network den sity is high. Fourteen ISCCP cloud variables were selected as the predictor s in linear and nonlinear regression models. The nonlinear model explains o ver 60% of the variance of the GPI bias, while the linear model explains ab out 45% of the variance. Comparison with another method of GPI bias estimat ion is discussed.