نشریه کاربرد سنجش از دور و سیستم اطلاعات جغرافیایی در علوم محیطی
Spatio‑temporal monitoring and analysis of water surface changes in Mahabad Dam using an integrated data‑driven approach based on Sentinel‑1 data and climatic variables
نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه سنجش از دور و GIS، دانشکده برنامه ریزی و علوم محیطی، دانشگاه تبریز، تبریز، ایران
2 گروه سنجش از دور و سیستم اطلاعات جغرافیایی، دانشکده برنامه ریزی و علوم محیطی، دانشگاه تبریز
3 گروه مهندسی آب، دانشکده کشاورزی، دانشگاه تبریز، تبریز، ایران
چکیده
Accurate monitoring of reservoir water surface area is essential for sustainable water management, especially in arid and semi‑arid regions that are highly sensitive to climatic fluctuations. Sentinel‑1 synthetic aperture radar (SAR) imagery, with its all‑weather and day‑night acquisition capability, provides a reliable basis for tracking surface water dynamics. In this study, 360 ascending Sentinel‑1 images in VV and VH polarizations were used to extract the water surface area of the Mahabad Dam. Water bodies were identified using the Support Vector Machine (SVM) classifier, which effectively distinguishes water from non‑water features based on radar backscatter. To enhance prediction accuracy, the XGBoost model was applied to integrate SAR‑derived water area with climatic variables such as precipitation and temperature. This approach enabled modeling of nonlinear relationships affecting reservoir variations. Model performance was evaluated using RMSE, MAE, R², NSE, and WI indices. Scenario 3 provided the most accurate results, with RMSE of 0.526, MAE of 0.464, R² of 0.911, and WI of 0.977, indicating strong agreement between predicted and observed values. Scenario 4 showed the weakest performance. Overall, integrating Sentinel‑1 SAR data with machine learning methods such as XGBoost offers an efficient framework for monitoring and predicting reservoir surface area changes, supporting improved water management and drought mitigation in climate‑sensitive regions.
تازه های تحقیق
To evaluate the performance of the proposed model for predicting water surface area, five statistical metrics were used: RMSE, MAE, R², NSE, and the Willmott Index (WI). Lower RMSE and MAE values represent smaller prediction errors, while higher R², NSE, and WI indicate stronger agreement between predicted and observed values. Ten scenarios were developed using different combinations of SVM-derived water surface information and climatic variables, including evaporation (E), precipitation (P), temperature (T), and relative humidity (RH). The statistical performance of all scenarios on the training dataset is summarized in Table 5.
The testing results show that the model performs well across most scenarios, with R² values ranging from 0.848 to 0.911 and WI between 0.955 and 0.977. Among all scenarios, Scenario 3 (SVM + E) achieved the best performance, with the lowest RMSE (0.526) and MAE (0.464), and the highest R², NSE (0.911), and WI (0.977). The results show thaevaporation is the most influential climatic factor for predicting water surface variations. Scenario 1 (SVM + T), Scenario 5 (SVM + E + T), and Scenario 6 (SVM + E + RH) also produced strong results, while Scenario 4 (SVM + P) showed the weakest performance, indicating that precipitation alone is insufficient for accurate prediction. Scenarios involving multiple variables provided stable but not superior results compared to simpler combinations.
Figure 8 shows the distribution of predicted values across all scenarios, highlighting the superior performance of Scenario 3, which exhibits the closest alignment with observed values and the lowest dispersion. Scenarios 1, 5, and 6 also demonstrate good accuracy, while Scenario 4 appears as the least reliable. Figure 9 compares the distributions of predicted and observed water surface areas. Scenarios involving evaporation show almost identical mean (μ) and standard deviation (σ) values between predicted and observed data, indicating that the model effectively captures both the central tendency and variability of surface water dynamics. Slight deviations in σ for weaker scenarios reflect a reduced ability to represent extreme fluctuations. Overall, Figures 8, 9 and 10 confirm the robustness of the proposed approach and emphasize the significant role of evaporation in improving water surface area prediction.
کلیدواژهها
موضوعات
عنوان مقاله [English]
Spatio‑temporal monitoring and analysis of water surface changes in Mahabad Dam using an integrated data‑driven approach based on Sentinel‑1 data and climatic variables
نویسندگان [English]
- Samaneh Bagheri 1
- Sadra Karimzadeh 2
- Bakhtiar Feizizadeh 1
- Saeedd Samadianfard 3
1 Department of Remote Sensing and GIS, Faculty of Planning and Environmental Sciences, University of Tabriz, Tabriz
2 Department of Remote Sensing and GIS, University of Tabriz, Iran
3 Department of Water Engineering, Faculty of Agriculture, University of Tabriz, Tabriz, Iran
چکیده [English]
Accurate monitoring of reservoir water surface area is essential for sustainable water management, especially in arid and semi‑arid regions that are highly sensitive to climatic fluctuations. Sentinel‑1 synthetic aperture radar (SAR) imagery, with its all‑weather and day‑night acquisition capability, provides a reliable basis for tracking surface water dynamics. In this study, 360 ascending Sentinel‑1 images in VV and VH polarizations were used to extract the water surface area of the Mahabad Dam. Water bodies were identified using the Support Vector Machine (SVM) classifier, which effectively distinguishes water from non‑water features based on radar backscatter. To enhance prediction accuracy, the XGBoost model was applied to integrate SAR‑derived water area with climatic variables such as precipitation and temperature. This approach enabled modeling of nonlinear relationships affecting reservoir variations. Model performance was evaluated using RMSE, MAE, R², NSE, and WI indices. Scenario 3 provided the most accurate results, with RMSE of 0.526, MAE of 0.464, R² of 0.911, and WI of 0.977, indicating strong agreement between predicted and observed values. Scenario 4 showed the weakest performance. Overall, integrating Sentinel‑1 SAR data with machine learning methods such as XGBoost offers an efficient framework for monitoring and predicting reservoir surface area changes, supporting improved water management and drought mitigation in climate‑sensitive regions.
کلیدواژهها [English]
- Mahabad Dam
- Sentinel 1
- XGBoost
- Support Vector Machine
- Water Surface Area
Accurate monitoring of reservoir water surface area is essential for sustainable water resource management, particularly in arid and semi‑arid regions where water availability is highly sensitive to climatic variability and anthropogenic pressures. Sentinel‑1 synthetic aperture radar (SAR) data, with its all‑weather and day‑night imaging capability, provides a reliable framework for investigating surface water dynamics while overcoming the limitations of optical imagery. In this study, 360 monthly averaged ascending Sentinel‑1 images acquired in VV and VH polarizations from January 2017 to December 2024 were processed to analyze the spatio‑temporal variations in the water surface area of Mahabad Dam. Ascending orbit images were selected due to clearer reservoir boundary delineation, whereas descending images were excluded because of reduced visibility associated with acquisition geometry. Water body extraction was performed using a Support Vector Machine (SVM) classifier with a radial basis function (RBF) kernel. Prior to classification, a 3 × 3 Lee filter was applied to reduce speckle noise. The model was trained using 1,000 reference samples (500 water and 500 non‑water), manually selected based on visual interpretation of SAR imagery and ancillary data. Classification accuracy was assessed monthly using Overall Accuracy (OA), User’s Accuracy (UA), and the Kappa coefficient, ensuring the reliability of the extracted water surface area. To further enhance the analysis, the XGBoost machine learning model was employed to integrate SAR‑derived water surface area with key climatic variables, including precipitation and temperature. The gradient boosting structure of XGBoost enabled effective modeling of complex, nonlinear relationships between climatic drivers and reservoir dynamics. Model evaluation indicated that Scenario 3 achieved the highest performance (RMSE = 0.526, MAE = 0.464, R² = 0.911, NSE = 0.911, WI = 0.977), while Scenario 4 showed the weakest performance. Overall, the results demonstrate that the synergistic integration of SAR remote sensing data and advanced machine learning techniques provides a robust, scalable, and efficient framework for monitoring and forecasting reservoir water surface area dynamics. This framework offers valuable insights for improving water resource management, supporting drought mitigation and adaptation strategies, and contributing to sustainable watershed planning in data‑scarce and climate‑sensitive regions. Nevertheless, the study is limited by its focus on a single reservoir and a restricted set of climatic variables. Future research should extend this approach by incorporating additional hydro‑meteorological factors, exploring alternative machine learning models, and applying the framework to multiple reservoirs to enhance its generalizability.
)
دوره 6، شماره 19
تیر 1405
- تاریخ دریافت: 08 فروردین 1405
- تاریخ بازنگری: 01 اردیبهشت 1405
- تاریخ پذیرش: 05 اردیبهشت 1405