Artificial intelligence-driven drought prediction using long-term meteorological and vegetation indices

Drought is a persistent challenge in the Bundelkhand region of Central India, driven by erratic monsoons, warming, and rain-fed agriculture. This study develops an integrated framework for drought monitoring and forecasting by combining long-term meteorological data (1985–2024), satellite-based vegetation indices, and artificial intelligence (AI) techniques. The Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI) were computed at 3- and 12-month timescales using Thornthwaite and FAO-56 Penman-Monteith methods. Drought severity and long-term trends were analyzed using the Mann-Kendall test. Vegetation response was assessed through correlations between NDVI, VCI, and SPEI-3. Three ensemble machine learning models—Random Forest (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost), were implemented for district-level drought classification and early warning. Results revealed strong spatial variability, identifying Jalaun (9.8%), Mahoba (8.8%), and Hamirpur (7.5%) as the most drought-prone districts. Mann-Kendall analysis indicated significant drying trends, with SPEI outperforming SPI. Vegetation indices exhibited weak but significant correlations with SPEI-3 (r ≈ 0.15–0.25). Feature importance analysis revealed that meteorological variables, particularly relative humidity and temperature, dominate drought prediction. AI models achieved overall accuracy exceeding 82%, while district-level accuracy exceeded 94%. This framework offers sensitive drought early warning for Bundelkhand and adaptable agricultural applications worldwide.