Water surface area monitoring in Golestan reservoir using Sentinel-2 images

Extended Abstract
Introduction
Continuous monitoring of surface area variations in dam reservoirs is a fundamental requirement for operational management, storage volume estimation, and reservoir hydrological analysis. Reservoir geometry evolves over time due to processes such as sedimentation and morphological alterations, hence reliance on initial area-elevation or storage-elevation curves may introduce increasing uncertainty into engineering calculations. The usual and traditional method for updating these curves is to carry out reservoir hydrograph or bathymetric surveys, which are very costly and time-consuming processes and requires specialized equipment, extensive field operations, and post-processing analyses. Furthermore, environmental constraints (wind, sunlight, air humidity, etc.) also affect the accuracy of the surveys. In this context, remote sensing technologies combined with cloud-based processing platforms provide an efficient and reproducible framework for extracting reservoir surface area. Therefore, the objective of this study is to estimate and monitor the water surface area of the Golestan dam reservoir located in Golestan province, Iran, during the time period of 2015 to 2025 using Sentinel-2 satellite imagery processed in the Google Earth Engine (GEE) cloud computing environment using Python. Sentinel-2 resources cover a time span from 2015 to present and have a repeat cycle of 5 days, which it is shorter than time span of available imagery. Also, the Sentinel-2 bands have a spatial resolution of 10m compared to Landsat-8 which has a spatial resolution of 30m. The computational results of the algorithm developed in this research have been compared with the results of the reservoir hydrographic operation in 2015, and the reasons for the differences have been analyzed. Moreover, to validate the results of this algorithm, data from the DAHITI database (Technical University of Munich) was used for two reservoirs in the United States, Aquilla and Choke Canyon.
 
Materials and Methods
Following atmospheric correction on more than 110 Sentinel-2 images downloaded from the three reservoirs studied in this research on specific dates during the years 2015 to 2025, the Normalized Difference Water Index (NDWI) was calculated to delineate water bodies by using the reflectance values of the green and near-infrared (NIR) bands. This index is dimensionless and ranges from -1 to 1, with higher values indicating the presence of water and lower values indicating other land cover types e.g. soil and vegetation. After the extraction of water bodies, binary water/non-water maps were generated. Reservoir surface areas were subsequently extracted for multiple dates, and a time series of water area variations was constructed. For validation, the derived surface areas from Aquilla and Choke Canyon dam reservoirs were temporally matched with DAHITI-based water extent data, and statistical error metrics (MAE, RMSE and R²) were calculated to quantitatively evaluate the method’s performance. This computational process was also carried out for Golestan dam reservoir. The area-elevation curve constructed for the year 2015 was compared with the curve obtained from the hydrographic survey of this reservoir for upper (near spillway crest level), middle and lower (near reservoir bed level) water stages. Following this comparative analysis, the area-elevation and storage-elevation curves for this reservoir were constructed for some time intervals between the years 2015 and 2025.
 
Results and Discussion
Overall, the quantitative results demonstrate the capability of the proposed Sentinel-2/GEE framework for long-term monitoring of reservoir geometry and sedimentation-induced changes. Validation against reference reservoirs showed a mean area estimation error of approximately 2.3%, confirming the reliability of the developed algorithm. Application of the method to Golestan Reservoir revealed a systematic decline in water level, with a long-term trend of approximately 0.07 m year⁻¹. Comparison between the satellite-derived and hydrographic area–elevation curves indicated that the discrepancy between the two approaches is strongly water-level dependent, with RMSE values increasing from about 0.56 km² at intermediate elevations to more than 3 km² near the upper operating water levels. The results further showed that the reservoir bed elevation increased by approximately 6.5 m during the 2015–2025 period, indicating substantial sediment accumulation. Consequently, the useful storage capacity of the reservoir decreased from about 86.5 Mm³ at the beginning of operation to approximately 50.5 Mm³ in 2015 and 39.5 Mm³ in 2025, corresponding to an overall storage loss of nearly 42%. The results obtained in Golestan reservoir indicate that the proposed approach provides reliable estimates of reservoir surface area. Temporal analysis of constructed area-elevation curves reveals a gradual decrease in surface area at certain water levels, potentially attributable to sedimentation processes. Comparison was made with the existing hydrographic data (2015) of Golestan Dam. The reduction in reservoir area was also elevation-dependent, reaching about 48% at the 57 m water level, while remaining limited to approximately 8.3% near the spillway crest elevation (62 m). These findings highlight the significant influence of sediment deposition on the geometric characteristics of the reservoir and demonstrate the effectiveness of satellite-based monitoring for updating reservoir area–elevation and storage–elevation relationships.
 
Conclusion
The results of this study show three fundamental aspects in evaluating the performance of reservoir geometry reconstruction by satellite imagery method. Firstly, the temporal and structural coverage superiority of the Sentinel-2 image-based approach in the Google Earth Engine environment, which enables continuous monitoring of reservoir geometric curves, thereby reducing the inherent temporal gaps of field study methods. Secondly, the significant dependence of the differences between the GEE algorithm and hydrographic surveys on the reservoir water level, especially in the ranges of spillway level and minimum operating level, indicates that the reservoir's geometric sensitivity in these ranges is high, and assuming a static area-elevation curve of the reservoir in these two ranges may lead to considerable deviations. Thirdly, periodic updating of geometric curves (area-elevation and volume- elevation) in reservoirs with high sedimentation rates is essential as a prerequisite for the reliability of hydraulic, sedimentary, and hydrological analyses of the reservoir and must be taken into consideration. From this perspective, the proposed approach can be considered a complementary framework for traditional field survey, a framework that, by integrating remote sensing data and field data, not only reduces the costs and time of operations but also decreases the uncertainty caused by the static geometry of the reservoir and provides a reliable basis for management analyses. The findings collectively highlight the necessity of updating reservoir geometric information and demonstrate the value of satellite-based observations as a complementary and up-to-date tool for continuous geometric monitoring, storage volume estimation, and reconstruction of area-elevation and storage-elevation relationships.