LAUSR

The ongoing drought constitutes a pivotal environmental challenge for the Mediterranean Region of Türkiye, where elevated climatic variability and erratic precipitation patterns result in considerable agricultural and hydrological stress. This study applied two artificial intelligence models—artificial neural network (ANN) and Random Forest (RF)—to forecast meteorological drought using the Standardized Precipitation Index (SPI) derived from nearly a century of monthly precipitation data (1929–2024) across eight provinces: Adana, Antalya, Burdur, Hatay, Isparta, Kahramanmaraş, Mersin, and Osmaniye. The models were evaluated at four accumulation periods (SPI-3, SPI-6, SPI-12, and SPI-24) using multiple statistical indicators. The findings indicated that artificial neural networks (ANNs) attained the highest predictive accuracy at extended timescales (SPI-12 and SPI-24), with R2 values reaching up to 0.94. This outcome signifies the capacity of ANNs to discern nonlinear and persistent drought patterns. The RF model exhibited enhanced stability and responsiveness in short-term forecasts (SPI-3, R2 = 0.89), effectively reproducing rapid fluctuations in rainfall. The comparative findings underscore the complementary strengths of the two models: ANN is better suited for the analysis of long-term drought trends and the study of climate adaptation, while RF offers reliable, low-complexity forecasting for the operational monitoring of drought. Utilizing solely precipitation data, the approach furnishes a cost-effective and transferable framework for data-limited regions. The study proposes a reproducible AI-based methodology that enhances the precision of drought prediction, supports early-warning applications, and strengthens regional water resource management under increasing climatic uncertainty.