This study aims to predict the water level of the Çamgazi Dam (Adıyaman_Türkiye) reservoir using an advanced convolutional neural network and nonlinear regression approaches on hydrological and meteorological parameters. A long-term hydro-meteorological dataset was employed, and the developed AI model was evaluated with multiple performance metrics. Results show R² ≥ 0.96 for all parameters, indicating high accuracy in predicting water level (elevation), reservoir volume, dimensionless V/h³ ratio, as well as meteorological variables such as temperature, station pressure, relative humidity, and monthly rainfall. The nonlinear regression model relating the V/h³ ratio to selected meteorological ratios achieved R² = 0.849, enabling physically interpretable insights into inter-variable interactions. Under extreme rainfall conditions, the model exhibits systematic under or overestimation and greater scatter at high relative humidity, identified as key limitations. This study highlights the potential of combining AI-based approaches with regression-based empirical models to contribute to related research and planning efforts.
Keywords: Convolutional neural network, Dam, Reservoir level, Water level prediction
