Modelling the Impact of Climate Change on the Reservoir Filling Rates of Dams Used for Drinking Water Supply Through Artificial Neural Networks

Abstract

Climate change has increasingly altered global precipitation patterns and hydrological cycles, posing significant risks to the storage performance and operational reliability of dams. Changes in rainfall regimes, prolonged droughts, and rising evapotranspiration rates directly affect reservoir inflows, making accurate prediction of these variations essential for sustainable water supply management. This study aims to model the effects of climate change on the storage levels of drinking water supply dams located in Ankara, Istanbul, and Izmir using various Artificial Neural Network (ANN) techniques. Among the ANN structures evaluated, dynamic time series, curve fitting, and feedforward distributed time delay networks demonstrated limited performance on the dataset; however, the feedforward backpropagation network demonstrated successful performance. This network type was optimized for the dataset through various modifications to parameters such as training, adaptive learning, performance, and transfer functions, as well as the number of layers and neurons. The findings were supported by regression analyses conducted for dams supplying drinking water. The differences between expected and observed precipitation amounts illustrate the impacts of climate change and water losses in the reservoirs. The models developed in the study exhibited R² values ranging between 0.85 and 0.99, with the highest value calculated as 0.994. The results quantitatively demonstrate the divergence between projected and observed precipitation on an individual dam basis, thereby elucidating the susceptibility of drinking water reservoirs to the impacts of climate change. In this context, the study offers robust, data-driven insights that can inform evidence-based revisions of water management and adaptation strategies for major metropolitan regions.