Explainable machine learning and extreme value analysis for precipitation projections in coastal Eastern Mediterranean Türkiye

Abstract

Extreme precipitation events across the vulnerable Eastern Mediterranean coastal region of Türkiye pose severe socio-economic and infrastructural threats. However, a unified framework capable of simultaneously integrating the underlying physical drivers, statistical recurrence properties, and future trajectories of these extremes remains notably lacking for this subregion. This study introduces an integrated hybrid pipeline combining multi-source data fusion, K-Means spatial regionalization, explainable XGBoost classification, extreme value analysis, and CMIP6 climate projections. Quality-controlled daily observational records from 24 meteorological stations (2010–2024) identified a statistically significant, spatially coherent precipitation regime shift centered around 2020, marked by a domain-mean annual anomaly of −20.5% relative to the baseline decade. Unsupervised K-Means clustering successfully partitioned the domain into three physically interpretable subregions: Coastal, Inland, and Mountainous. The temporally validated XGBoost classifier achieved strong diagnostic performance with an AUC-ROC of 0.835. TreeSHAP explainability analysis revealed that near-surface relative humidity is the overwhelmingly dominant thermodynamic trigger for extreme precipitation events (mean \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\text {SHAP}|=0.557$$\end{document}), with a non-linear activation threshold near 65–70% consistent with Clausius–Clapeyron constraints. This control intensifies with elevation, reflecting orographic enhancement across the Taurus range. Furthermore, Generalized Extreme Value (GEV) modeling over a 44-year CHIRPS historical baseline quantifies severe hazard concentrations along the eastern inland corridor, yielding a domain-mean 100-year return level of 102.4 mm day\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-1}$$\end{document}. Finally, CMIP6 projections under SSP2-4.5 and SSP5-8.5 scenarios indicate robust declines in extreme event frequencies (−19.5% and −12.1%, respectively) by 2041–2060, highlighting a transitional shift toward a “fewer but potentially more intense” convective precipitation regime.