|
Monitoring the water level of Jareh Dam reservoir by integrating Sentinel-2 images and Random Forest algorithm
|
Javad Zahiri *   |
| Agricultural Sciences and Natural Resources University of Khuzestan |
|
|
Abstract: (9 Views) |
| Reservoirs play a key role in providing drinking water, agriculture, hydropower generation, and flood control by enabling the storage, regulation, and distribution of freshwater. However, sustainable reservoir operation depends on accurate data on water level and surface area. Traditional field measurement methods, despite their high accuracy, are not efficient for widespread and continuous reservoir monitoring—especially in arid and semi-arid regions—due to high costs, logistical constraints, and lack of access to remote areas. In this study, an integrated framework was developed for daily reservoir water level estimation at the Jarreh Dam reservoir by combining Sentinel-2 imagery, the ALOS PALSAR Digital Elevation Model (DEM), and the Random Forest algorithm. The NDWI and MNDWI spectral indices were used to extract the water body and calculate surface water area. Water edge elevation was extracted by applying the DEM to the reservoir boundary pixels. Three different scenarios were designed: satellite-derived features (modeled water level from DEM, NDWI, MNDWI, and reservoir area), along with measured water level data from 5 days and 10 days prior. The data were split into training (80%) and testing (20%) sets and then standardized. Results showed that the Random Forest model in the first scenario achieved acceptable performance (test RMSE = 1.33 m). However, adding water level data from 5 days prior (second scenario) significantly improved performance, reducing the test RMSE to 0.74 m and achieving NSE value of 0.92. The third scenario showed no meaningful improvement over the second scenario. Uncertainty assessment using out-of-bag error indicated that the second and third scenarios had narrower uncertainty bands (approximately 1.40 m for the 90% confidence interval) compared to the first scenario (1.65 m). However, the first scenario enables reservoir water level estimation using solely remote sensing data, without the need for in-situ measurements, and thus holds particular practical significance for regions lacking gauge stations. The sensitivity analysis revealed that the model is most dependent on measured water levels, followed by water level and reservoir area extracted from satellite imagery, which play the most substantial role in determining model outputs. This study demonstrates that integrating Sentinel-2 data, a DEM, and the Random Forest algorithm can provide an efficient, low-cost, and accessible tool for continuous monitoring of reservoir water levels. |
|
Article number: 7 |
| Keywords: Machine Learning, Remote Sensing, Spectral Indices, Uncertainty Analysis, Water Resources Management. |
|
|
|
Type of Study: Research |
Subject:
Special
Received: 2026/06/3 | Revised: 2026/07/5 | Accepted: 2026/07/5 | ePublished ahead of print: 2026/07/5
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
