Climate-sensitive assessment of drought dynamics using pattern-mining and machine-learning approaches across Iran’s hydro-climatic zones

Résumé fourni par la source

Study region Iran. Study focus Drought dynamics vary markedly across climatic regimes, and the reliability of conventional drought indicators may decline in hydro-climatically heterogeneous regions. This study presents a climate-sensitive, multi-scale comparative assessment of drought behavior across four major climatic classes of Iran using five conventional drought indices (SPI, SPEI, MSDI, DPI, and RDI), two machine-learning models (SVM and ANF-PSO), and a Pattern Mining Engine (PME). Using 44 years of observations, the study examined how drought intensity, duration, frequency, and event structure vary with climatic class and temporal scale, and whether PME provides more stable behavior under contrasting hydro-climatic conditions. Model performance was evaluated using accuracy statistics, Taylor diagrams, correlation analysis, and uncertainty diagnostics. New Hydrological Insights for the Region Drought behavior and model performance were strongly climate-dependent. Indices incorporating evaporative demand, particularly SPEI and RDI, were more reliable in very dry and dry climates, whereas precipitation-based indices showed comparatively more stable behavior in humid conditions. No conventional index performed uniformly well across all climatic classes. Within this comparative framework, PME demonstrated a clear advantage by consistently capturing the dominant spatial and temporal structure of drought across all climatic classes and time scales, while maintaining lower sensitivity to climatic variability. Unlike conventional indices and baseline ML models, PME integrates multivariate hydro-meteorological interactions through pattern-based representation, enabling more robust identification of drought conditions under heterogeneous environments. In contrast, MSDI showed less stable behavior, especially in semidry and humid regions. These findings highlight the added value of PME as a climate-sensitive framework for improving drought characterization and support more reliable basin-scale monitoring, early warning, and water-resources planning.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.