Statistical Forecasts for the Occurrence of Precipitation Outperform Global Models over Northern Tropical Africa


Precipitation forecasts are of high socioeconomic importance for agriculturally dominated societies in northern tropical Africa. However, recent work has shown that state-of-the-art operational global ensemble prediction systems (EPSs) are not able to outperform climatology-based forecasts for 1–5-day accumulated precipitation during the monsoon season, even after applying sophisticated statistical post-processing methods in the form of Bayesian model averaging (BMA) and ensemble model output statistics (EMOS). Raw ensemble forecasts are uncalibrated and unreliable, mostly due to poor prediction for low precipitation amounts. The parameterization of convection is a potential cause for the sobering lack of ensemble forecast skill in a region dominated by mesoscale convective systems (MCSs). One possible alternative to dynamical model forecasts and post-processing are statistical forecasts based on spatio-temporal correlation patterns in past observations. Here, we investigate this approach exemplarily for the prediction of the probability of precipitation (PoP) in northern tropical Africa. Such an approach is deemed promising, as rainfall in this region is predominantly related to long-lived MCSs, which in turn are modulated by synoptic-scale African easterly waves (AEW). The results presented here are based on Tropical Rainfall Measuring Mission (TRMM) satellite-based rainfall estimates during the period 1998–2014. The statistical model is developed for every grid-point using a logistic regression approach applied to the rainfall of the previous two days. Results do indeed show correlation patterns consistent with MCS and AEW propagation but also indications of latitudinal shifts in the monsoon system. The statistical PoP forecasts are reliable and have a higher resolution than climatology-based forecasts. Improvements in the predictive performance in terms of the Brier score reach up to 20%. In view of the promising initial results, we will expand the concept to precipitation amounts, which we expect to be modulated by preceding rainfall characteristics as well, and possibly to other regions. Furthermore, we will explore whether statistical forecasts benefit from the inclusion of NWP predictions of larger-scale features such as regional moisture patterns, the African easterly jet or tropical wave modes.

100th American Meteorological Society Annual Meeting