Geostatistical Predictive Model of Drought Severity: A Case Study of Southern Portugal

Abstract Droughts represent one of the most significant natural hazards, with widespread implications for agriculture, ecosystems, and human livelihoods, particularly in regions like southern Portugal. In recent decades, severe drought events have become increasingly frequent and intense. Considering the potential impacts of droughts, particularly concerning water and river basin management—a critical factor in southern Portugal—the main motivation for this study was the creation of a short-term forecast alert system for drought severity. This study aims to assess the effectiveness of the random forest (RF) model in forecasting drought severity, using consecutive dry days (CDD) as a drought indicator. Daily precipitation data from 17 meteorological stations across southern Portugal were used to classify drought severity into three categories (classes A, B, and C). The RF models were applied to predict the likelihood of each drought severity class, using historical data to forecast future drought conditions. The model's performance results indicate that the RF model performs well in predicting moderate to severe drought classes, with particularly strong accuracy during the peak dry season in July and August. However, the model exhibited challenges in forecasting extreme drought classes, especially during the transitional months of June and September, likely due to the variability in precipitation patterns during these periods. The findings demonstrate the utility of the RF model as a reliable tool for early drought warning and water resource management in southern Portugal. Furthermore, the methodology presented in this study can be adapted to other regions, making it a versatile approach to addressing drought-related challenges worldwide.