Machine learning based prediction of long-term drought persistence over the Arabian Peninsula

The Arabian Peninsula is among the most water-stressed regions globally, where limited precipitation, high evapotranspiration and rapid socio-economic development exacerbate vulnerability to drought. Emerging evidence indicates a significant intensification of drought conditions in recent decades, driven by climate variability and long-term warming trends posing serious challenges to water security, ecosystem stability and socio-economic resilience. Therefore, understanding historical drought dynamics, together with reliable drought prediction, is essential for strengthening drought monitoring and mitigation strategies in arid environments and for reducing drought-related risks. However, accurate drought prediction at fine resolution scale remains challenging due to the sparse distribution of meteorological stations. This study investigates the performance of the Standardized Precipitation Index (SPI) and the Standardized Precipitation Evapotranspiration Index (SPEI) at 3-, 6- and 12-month timescales using precipitation data from the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and potential evapotranspiration derived from the TerraClimate dataset, respectively, for pixel-level drought assessment over the period 1992-2024. The historical dynamics were studied using Mann-Kendall trend, Sen’s slope and hotspot analysis. Random Forest (RF) was employed to assess its applicability for drought prediction in arid environments using satellite data, owing to its widespread adoption in global drought-prediction studies. The analysis demonstrates that the RF model exhibits high predictive performance under the studied conditions, with robust performance for SPEI-6 (R² = 0.92, RMSE = 0.12, NSE = 0.92) and satisfactory results for SPEI-12 (R² = 0.77, RMSE = 0.22, NSE = 0.77). These findings confirm enhanced predictability of seasonal to long-term drought variability across the Arabian Peninsula using a satellite-driven RF framework. The results showed the dominance of antecedent SPEI variables (>90%) indicating that cumulative moisture deficits and rising atmospheric evaporative demand primarily govern seasonal to long-term drought evolution over the Arabian Peninsula. In contrast, the consistently low contribution of SPI based indices (<3%) underscores the limited standalone role of precipitation variability in sustaining drought conditions in this arid region. Consistent with these predictive results, spatial trend analysis reveals pronounced heterogeneity in drought evolution across the Arabian Peninsula, with SPI exhibiting mixed and weak precipitation-driven signals, whereas SPEI shows widespread and statistically significant drying, particularly at 6- and 12-month timescales. This divergence further confirms that increasing evaporative demand and regional warming are the primary drivers of long-term drought intensification, reinforcing the dominant role of evapotranspiration processes identified by the machine-learning models. Therefore, the integration of satellite-derived pixel-level datasets with the RF model provides an effective framework for drought prediction across the Arabian Peninsula, offering valuable insights for water resource managers and policymakers to support the development of robust early warning systems and targeted mitigation strategies.