Adaptive Use of TRMM Observations for Tropical Precipitation Forecasts.

Abstract

In order to assess and improve the short-range tropical precipitation forecasts, an adaptive use of TRMM (Tropical Rainfall Measuring Mission) rainfall via a physical initialization (PI) technique is explored. An EOF-based perturbation method is first employed to generate 25 initial ensemble members. From these initial ensemble analyses, twelve-hour forecasts are performed. Predicted 12-hour accumulated rainfall fields are then used to find sensitive (adaptive) regions, owing to initial condition uncertainties in terms of the rainfall forecast error variance. To shorten the computing time, the 14-level Florida State University Global Spectral Model at a resolution of T106 (FSUGSM T106L14) is employed in the above procedure. Next, based on a map of the rainfall forecast error variance, higher resolution T170 experiments are carried out from the adaptively initialized states. The rainfall forecast skill of the adaptive PI experiments is compared to that of the normal mode initialization (NMI), and the regular PI experiments. In the regular PI, it is performed over the whole tropics. Our results indicate that the adaptive PI of the TRMM rainfall over regions of large rainfall variances due to initial condition uncertainties, may potentially improve 3-day rainfall forecast skills over the global tropics. In addition, more notable improvements are evident over the adaptive regions, themselves.