Methods of ensemble forecasting of rainfall variability in tropical West Africa

Generally speaking, there are several ways to produce ensemble forecasts. The standard approach is simply using the raw output of a numerical model-based forecast ensemble. This method has been shown to exhibit poor skill for 24h precipitation forecasts over tropical Africa, since forecasts are, amongst other things, highly miscalibrated. Miscalibration can largely be cured by statistical postprocessing, often at the prize of reduced discrimination. However, if applied to tropical Africa again, the skill of postprocessed ensemble forecasts is still often not significantly better than a simple reference forecast based on past observations (here termed Extended Probabilistic Climatology (EPC), third approach) (Vogel et al. 2018; 2020). A fourth approach is to ignore numerical model output altogether and to generate purely data-driven forecasts based on spatio-temporal dependencies inferred from gridded satellite rainfall estimates. As recently demonstrated by Vogel et al. (2021), this approach based simply on information of precipitation in the two days preceding the forecasted day shows some promise for the prediction of 24-hour precipitation occurrence probability in tropical Africa.

This contribution explores this potential further by advancing the statistical model and providing meteorological interpretations of the performance results. Advances include (a) the use of a recently developed correlation metric, the Coefficient of Predictive Ability (CPA), to identify predictors, (b) forecast evaluation with robust reliability diagrams and score decompositions, (c) a study domain over tropical Africa nested in a considerably enlarged spatio-temporal domain to identify coherent propagating features and, (d) the introduction of a novel coherent-linear propagation factor to quantify the coherence of propagating signals. The statistical forecast is compared with the EPC, the European Centre for Medium-Range Weather Forecasts (ECMWF) operational ensemble forecast, and a statistically post-processed ensemble forecast. All methods show poor skill within the main rainbelt over northern tropical Africa, where differences in Brier scores between the different approaches are hardly statistically significant. However, the data-driven forecast outperforms the other methods along the fringes of the rainbelt, where meridional rainfall gradients are large. The coherent-linear propagation factor, in concert with metrics of convective available potential energy and convective instability, reveals that high stochasticity in the rainbelt limits predictability. At the fringes of the rainbelt, the data-driven approach leverages coherent precipitation features associated with propagating tropical weather systems such as African easterly waves.