Subseasonal prediction of tropical cyclone precipitation

Abstract

The accurate prediction of tropical cyclone precipitation (TCP) at an extended-range could be crucial to mitigate the impacts of TC-related flooding. This study examines probabilistic predictions of weekly-accumulated TCP and total precipitation using 11 subseasonal forecast systems. Raw, uncalibrated, categorical forecasts of basin-wide TCP are only skillful in the ECMWF model and only up to 15 days in advance and except in the northern Indian Ocean and the South Pacific. Calibration, through linear regression, improves forecasts and makes several forecast systems (GEOS, UKMO) skillful up to 15 days in advance but only in some basins. In most models and basins, such as the GEOS model in the Atlantic basin, the bias in the forecast probability of TC occurrence is the main factor driving biases in TCP and decreasing forecast skill. At the regional-scale, calibrated ECMWF forecasts are skillful beyond 15 day leads and globally. The poor prediction of TCP in raw forecasts is shown to affect total precipitation prediction skill. Therefore, biases in the TC occurrence probability forecast is the leading cause of low skill of TCP and may play a role in the skill of total precipitation.