Seasonal Forecasts as a Tool for Agricultural Drought Early Warning in Portugal

The increasing frequency and severity of agricultural drought pose a growing threat to Mediterranean farming systems, intensifying soil water deficits, crop water stress, productivity losses, and degradation of water quality. This study evaluates ECMWF SEAS5 seasonal forecasts (SFs) for drought monitoring and early warning in Portugal, considering the accumulation period, lead time, event type, seasonality, and spatial variability. Forecast performance was assessed by comparing the Standardized Precipitation Evapotranspiration Index (SPEI) from SEAS5 forecasts (2017–2024) with SPEI from bias-corrected ERA5-Land reanalysis, using Cohen’s Kappa and confusion-matrix metrics (True Positive Fraction—TPF; False Positive Fraction—FPF). SF generally reproduced the main spatial gradients and seasonal variability of climatic water availability and captured the major drought episodes. However, some limitations were identified, notably the SF’s coarse spatial resolution and a persistent warm-season wet bias, resulting in drought severity underestimation. Exact categorical agreement was limited (29–42%), though quadratically weighted Kappa indicated that most disagreements occurred between adjacent SPEI classes. Although forecast performance decreased with lead time, it improved with accumulation period, with SPEI-9 retaining substantial agreement (quadratically weighted Kappa ≥ 0.6) through the sixth lead month in the warm season. Dry conditions were better identified during the cool season (TPF up to 60%) and wet conditions in the warm season (TPF up to 89%), the latter linked to higher false alarm rates (FPF 6–11%). Extreme SF percentiles occasionally reduced the wet bias and improved detection, but generally showed larger errors, whereas the SF median provided the most balanced skill. Overall, these findings suggest that SFs have the potential to provide meaningful added value for drought early warning in Portugal, while bias correction, downscaling methodologies, and multi-system integration represent promising pathways to further enhance their usability for agricultural decision-making.