Improved rainfall forecasts in daily use over East Africa

Abstract

Ensemble forecasting has proven to be a vital tool for predicting extreme, life-threatening or only partially predictable weather events. As well as providing probabilistic products, individual ensemble members provide context and realisations of possible extreme weather. However, many National Meteorological Services in East Africa lack the computing resources required to run their local area models in ensemble mode over the full period of the two-week medium range. In this paper we test the performance of a short and medium-range forecast system recently put into daily use, comprising the global ECMWF ensemble forecast, post-processed using cGAN, a neural network model. We find that cGAN achieves probabilistic forecast skill comparable to that obtained using isotonic distributional regression (IDR), a recent statistical post-processing method, applied to either the ECMWF ensemble or the machine-learning models FuXi and GraphCast. Unlike IDR, cGAN generates individual ensemble-member realisations (potentially many thousands), including exceptional weather events. This ability was deemed important when assessed in an operational context by the National Meteorological and Hydrological Services of Kenya and Ethiopia. cGAN is cheap to train/run and requires no additional post-processing. It is run on laptops making it practical as a high-resolution medium-range ensemble forecast system for Meteorological Services with limited computational resources.

arXiv comment32 pages, 24 figures