Scale-dependent precipitation forecast error in the GFS

Precipitation forecasts from the National Center for Environmental Prediction's (NCEP) Global Forecast System (GFS) are verified against satellite precipitation observations. The satellite observations are from UC-Irvine's PERSIANN satellite data products, which provide a 6-hourly precipitation total on a 0.25x0.25 degree grid for a global domain excluding high latitudes (> 50 degree south and north). The GFS model forecasts come from a standard archive of 1x1 degree with daily aggregated precipitation.

RMS errors between GFS and PERSIANN satellite observations are computed for 2005-2006. Then these errors are averaged for summer (April-September) and winter (October-March) seasons, respectively. A 2D spatial-wavelet decomposition is applied to these RMS errors, resulting in a scale-dependent distribution of GFS precipitation forecast errors. In particular, we have analyzed the GFS precipitation forecast errors at a large scale [~O(1000) km] and at a smaller near-convective scale [~O(100) km]. These analyses provide guidance for how and where GFS should improve its data assimilation and physical parameterization for precipitation forecasts at different scales.