Machine Learning-Based Water Level Forecast in a Dam Reservoir: A Case Study of Karaçomak Dam in the Kızılırmak Basin, Türkiye

Reliable dam reservoir operation is crucial for the sustainable management of water resources under climate change-induced uncertainties. This study evaluates four machine learning algorithms—linear regression, decision tree, random forest, and XGBoost—for forecasting daily water levels in a dam reservoir in the Western Black Sea Region of Türkiye. A dataset of 5964 daily hydro-meteorological observations spanning 17 years (2008–2024) was used, and model performances were assessed using MAE, RMSE, and R 2 metrics after hyperparameter optimization and cross-validation. The linear regression model showed weak predictive capability (R 2 = 0.574; RMSE = 2.898 hm 3 ), while the decision tree model achieved good accuracy but limited generalization (R 2 = 0.983; RMSE = 0.590 hm 3 ). In contrast, ensemble models delivered superior accuracy. Random forest produced balanced results (R 2 = 0.983; RMSE = 0.585 hm 3 ; MAE = 0.046 hm 3 ), while XGBoost achieved comparable accuracy (R 2 = 0.983) with a slightly lower RMSE (0.580 hm 3 ). Statistical tests ( p > 0.05) confirmed no significant differences between predicted and observed values. These findings demonstrate the reliability of ensemble learning methods for dam reservoir water level forecasting and suggest that random forest and XGBoost can be integrated into decision support systems to improve water allocation among agricultural, urban, and ecological demands.