Machine learning-based hydrological trend analysis for water scarcity assessment in Erbil, Kurdistan Region of Iraq

ABSTRACT This study explores how long-term changes in temperature and precipitation have shaped river discharge in Erbil, a city in the Kurdistan Region of Iraq, using historical climate data from the World Bank Group Climate Knowledge Portal and discharge data from the Eski Kalak gauge station. A machine learning approach, extreme gradient boosting (XGBoost), was applied to extrapolate future discharge trends for 2024–2050 based on historical climate observations. The future results represent trend-based extrapolations rather than physically based climate model projections. The findings show a clear decline in precipitation and river flow over the past century, while temperatures continue to rise. The precipitation–discharge statistical relationship was weak (r = 0.038), suggesting that rising temperature and associated evapotranspiration processes may contribute to discharge variability alongside other climatic and anthropogenic influences, supported by a moderate negative discharge–temperature correlation (r = −0.255). The effect of upstream activities, including water diversions and dam construction, also cannot be disregarded and likely contributes to disruption of natural flow regimes. The XGBoost model achieved strong predictive performance (testing: R2 = 0.89, NSE = 0.88, RMSE = 17.3 m3/s), and the future-risk assessment indicates a rising likelihood of below-average discharge.