Predictability of large-scale extreme droughts from global bias-corrected seasonal forecasts

Extreme droughts are increasing in severity and frequency due to climate change, highlighting the need for reliable early-warning information at seasonal time scales. Bias-corrected seasonal forecasts offer a promising basis for anticipating drought conditions several months in advance, yet their predictive skill and potential operational value for the most severe droughts at the global scale remains insufficiently quantified. Here, we assess the performance of seasonal drought forecasts derived from the SEAS5-BCSD dataset using the Standardized Precipitation Evapotranspiration Index (SPEI). Thirty-six of the most extreme drought events between 1981 and 2024 are identified from ERA5 reanalysis, selecting two events per continent and accumulation period (SPEI-1, SPEI-3, and SPEI-6), and assessing forecast performance using probabilistic, spatial, and impact-oriented verification metrics, including the Continuous Ranked Probability Skill Score (CRPSS), the Fractions Skill Score (FSS), and the Potential Economic Value (PEV). The results demonstrate that SEAS5-BCSD drought forecasts outperform climatology for nearly all analyzed events. Across all verification metrics, forecast skill is generally highest for SPEI-1 events and for the selected African droughts, whereas the selected North American events exhibit comparatively lower skill. ... mehrWhile the exact location and intensity of the most severe drought cores remain difficult to predict, useful probabilistic information is often preserved within the ensemble distribution. Maximum PEV values frequently exceed 0.7, indicating substantial potential value for decision-making under suitable cost-loss assumptions. These findings highlight the benefits of probabilistic, ensemble-based interpretation of seasonal drought forecasts. Rather than relying solely on the ensemble mean, considering the full forecast distribution enables more risk-informed drought preparedness by accounting for both the likelihood and the potential severity of extreme events.