Assessing Climate-system Historical Forecast Project (CHFP) seasonal forecast skill over Central Africa

The present study investigates the predictive skill of the Climate-system Historical Forecast Project (CHFP) seasonal forecast in Central Africa (CA) using deterministic and categorical evaluation methods with focus on rainfall. The skill is evaluated for all the seasons December–February (DJF), March–May (MAM), June–August (JJA), September–November (SON) at 1- and 4-month lead-time (lead-1 and lead-4) that are consistent with many regional climate outlooks. It is found that for DJF and JJA at lead-1, the 8 models of the CHFP represent well the seasonal mean rainfall in CA with correlations greater than 0.7. For MAM and SON seasons, the scores are less good and the Japan Meteorological Research Institute version 1 (JMAMRI1) model presents the best scores. For the MAM season at lead-4, the JMAMRI1 model is better. The CHFP Multi-model ensemble (MME) mean captures the spatial differences in the seasonal mean climatology of precipitation and clearly resolves the bi-modal and uni-modal natures of observed precipitation. For the DJF season, at lead-1, the CHFP MME correctly captures the maximum rainfall observed in the Southern Democratic Republic of Congo (DRC) and northern Angola. The rainfall intensity is slightly overestimated. Results indicate that for DJF and MAM, the Probability Of Detection (POD), accuracy, Success Ratio (SR), and Equitable Threat Score (ETS) are higher for the less than precipitation climatology than for greater than precipitation climatology events. This indicates that CHFP forecasts may be more useful in forecasting less than precipitation climatology conditions than greater than precipitation climatology events conditions. That is, the CHFP forecast ensemble is better able to capture the dominant mechanisms responsible for years of decreased rainfall rather than increased rainfall. It follows that the CHFP models appear to be a valuable tool that can provide some key seasonal features up to 4 months in advance, which can thus help decision-makers of this region to take appropriate adaptation and mitigation measures.