New multivariate composite remote sensing drought index based on machine learning and geospatial techniques, insights from Northern Iraq

Study region Northern Iraq, is widely known as the breadbasket of Iraq, famous for its cereal production. It is an arid to semi-arid region that frequently experiences significant droughts, impacting agriculture, water resources, and ecosystems. Study focus This study develops and evaluates five machine learning models (Random Forest (RF), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), Gradient Boosting Machine (GBM), and Artificial Neural Networks (ANN)) for predicting the Standardized Precipitation Evapotranspiration Index (SPEI) at 3-month (SPEI-03) and 6-month (SPEI-06) timescales across Iraq. The models were trained and tested with satellite-based and bias-corrected gridded data covering 2001–2023. Seventeen variables from meteorological, vegetation, soil, and topographic sources were used across four predictor scenarios. New hydrological insights for the region Results demonstrate that RF and XGB models consistently outperformed other models in estimating short-term (SPEI03) and medium-term (SPEI06) drought conditions, with R² up to 0.90 and NSE of 0.89. SHAP analysis revealed that precipitation is the dominant driver of short-term droughts, while temperature, vegetation indices, and soil moisture have greater influence on medium-term droughts. The proposed modeling framework improves the understanding of regional drought dynamics and offers a robust, data-driven tool to strengthen early warning systems and support drought risk management for local stakeholders.