Demirbaş et al. (2026) Modelling the Impact of Climate Change on the Reservoir Filling Rates of Dams Used for Drinking Water Supply Through Artificial Neural Networks

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-03-01
• Authors: Furkan Demirbaş, Emine Elmaslar Özbaş, Mehmet Sarıkap, Miraç Nur Ciner, Hüseyin Yurtseven, Huseyin Kurtuluş Ozcan
• DOI: 10.1007/s11269-026-04527-0

Research Groups

• Engineering Faculty, Environmental Engineering Department, Istanbul University-Cerrahpasa, Avcılar, Istanbul, Türkiye
• Project and Technology Office, Istanbul University-Cerrahpasa, Avcılar, Istanbul, Türkiye
• Forestry Faculty, Surveying and Cadastre Department, Istanbul University-Cerrahpasa, Bahçeköy, Istanbul, Türkiye

Short Summary

This study models the impact of climate change on the reservoir filling rates of drinking water supply dams in Ankara, Istanbul, and Izmir, Türkiye, using Artificial Neural Networks (ANNs). It quantifies the divergence between expected and observed precipitation, revealing significant water losses and the susceptibility of these urban water systems to climate change.

Objective

• 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.

Study Configuration

• Spatial Scale: Drinking water supply dams and their delineated catchment areas in Ankara, Istanbul, and Izmir, Türkiye. Specific dams include Akyar, Çamlıdere, Çubuk 2, Eğrekkaya, Kurtboğazı (Ankara); Alibey, Büyükçekmece, Ömerli, Sazlıdere, Terkos (Istanbul); Balçova, Tahtalı (Izmir). Catchment areas range from 53.28 km² to 1304.88 km².
• Temporal Scale: Daily meteorological data collected between 2011 and 2020. Specific periods vary per dam, covering 3.5-year, 5-year, or 10-year durations.

Methodology and Data

• Models used: Artificial Neural Networks (ANNs). Four network types were tested: Dynamic Time Series, Curve Fitting, Feedforward Distributed Time Delay, and Feedforward Backpropagation Networks. The Feedforward Backpropagation Network was selected due to its superior performance, optimized with 18 neurons in the hidden layer, the Levenberg–Marquardt (trainlm) algorithm for training, Mean Squared Error (MSE) as the performance function, and tansig/purelin transfer functions for hidden/output layers, respectively.
• Data sources: Meteorological data from 149 official observation stations (45 in Ankara, 50 in Istanbul, 54 in Izmir) located within or near each reservoir catchment. Input parameters included average actual pressure (hPa), average relative humidity (%), daily maximum wind speed (m/s), daily average temperature (°C), and daily average wind speed (m/s). The output variable was daily total precipitation (mm). Shuttle Radar Topography Mission (SRTM)-based Digital Elevation Models (DEMs) with a 1 arc-second (~30 m) resolution were used for basin delineation with ArcGIS Desktop 10.8.1.

Main Results

• The optimized Feedforward Backpropagation ANN model achieved R² values ranging from 0.85 to 0.99, with a maximum R² of 0.994 for the Ömerli Dam.
• Significant differences were observed between expected (ANN-predicted) and actual precipitation, indicating substantial water losses attributed to climate change. For instance, the Eğrekkaya Dam showed an expected precipitation of 40,384.29 mm over 10 years compared to an actual 5767.05 mm (14% occurrence).
• Quantified water losses in dam reservoirs due to climate change were substantial:
  • Büyükçekmece Dam: 446.51 x 10^6 m³ (equivalent to three dam capacities) over a 5-year period.
  • Sazlıdere Dam: 213.55 x 10^6 m³ (equivalent to two dam capacities) over a 5-year period.
  • Eğrekkaya Dam: 120.12 x 10^6 m³ (equivalent to one dam capacity) over a 10-year period.
  • Terkos Dam: 1069.30 x 10^6 m³ (equivalent to one dam capacity) over a 10-year period.
• Other tested ANN architectures (Dynamic Time Series, Curve Fitting, Feedforward Distributed Time Delay networks) exhibited lower average R² values (between 0.58 and 0.65) and higher RMSE, confirming the superior performance of the selected Feedforward Backpropagation Network.

Contributions

• Provides accurate predictions of climate change-induced precipitation variations for drinking water supply dams in major Turkish metropolitan areas (Ankara, Istanbul, Izmir) using ANN models.
• Develops a methodological framework that integrates extensive meteorological datasets with ANN-based reservoir filling rate modeling, applicable to other urban water systems.
• Quantitatively demonstrates the significant divergence between projected and observed precipitation on an individual dam basis, highlighting the vulnerability of drinking water reservoirs to climate change impacts.
• Offers robust, data-driven insights to inform evidence-based revisions of water management and adaptation strategies for major metropolitan regions.

Funding

• The Scientific and Technological Research Council of Türkiye (TÜBİTAK).

Citation

@article{Demirbaş2026Modelling,
  author = {Demirbaş, Furkan and Özbaş, Emine Elmaslar and Sarıkap, Mehmet and Ciner, Miraç Nur and Yurtseven, Hüseyin and Ozcan, Huseyin Kurtuluş},
  title = {Modelling the Impact of Climate Change on the Reservoir Filling Rates of Dams Used for Drinking Water Supply Through Artificial Neural Networks},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-026-04527-0},
  url = {https://doi.org/10.1007/s11269-026-04527-0}
}