Akram et al. (2026) An Earth observation framework for reservoir water level estimation in a data-scarce transboundary basin for flood preparedness

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-22
• Authors: M. Rizwan Akram, Arshad Ali, Naeem Aftab
• DOI: 10.1016/j.ejrh.2026.104005

Research Groups

• Punjab Irrigation Department, Punjab, Pakistan
• Google Earth Engine Team

Short Summary

This study develops an observation framework using Sentinel-1 SAR imagery in Google Earth Engine to estimate Pong Dam water levels from reservoir extents. The refined DEM-constrained backscatter substantially outperformed single-polarization, confirming its reliability as a water level proxy for Pong Dam.

Objective

• Develop a long Sentinel-1 SAR time series of Pong Dam surface water extent using dual-polarization and DEM-guided thresholding.
• Establish a gauge-validated relationship between satellite-derived surface water area and reservoir water level.
• Develop and compare multiple regression and machine learning predictive models for near-real-time water level estimation to support flood preparedness in Pakistan.

Study Configuration

• Spatial Scale: Pong Dam, a large reservoir on the Beas River in Himachal Pradesh, India, with a catchment area of 12,561 km².
• Temporal Scale: The study period spans from 2018 to 2025, with historical water level records compiled from multiple sources.

Methodology and Data

• Models used:
  • Polynomial regression (degrees 2 and 3)
  • Random Forest (200 trees)
  • XGBoost (300 estimators, tuned depth and learning rate)
  • Support Vector Regression (SVR) with an RBF kernel after feature standardization
  • k-nearest neighbours (k = 5)
• Data sources:
  • Sentinel-1 Synthetic Aperture Radar imagery
  • Google Earth Engine
  • SRTM DEM

Main Results

• The refined DEM-constrained backscatter substantially outperformed single-polarization, confirming its reliability as a water level proxy for Pong Dam.
• The relationship between the reservoir surface area and in-situ water level was further analysed using multiple predictive models.
• The SVR (RBF kernel) and polynomial regression (degree 2) models were selected for operational deployment.

Contributions

• This study advances the use of freely available, cloud-penetrating satellite observations to close the hydrological monitoring gap in the Beas–Sutlej transboundary system.
• The framework provides an operationally validated, ground-data-independent approach to reservoir monitoring, supporting flood early warning in downstream Pakistan.

Funding

• Not specified

Citation

@article{Akram2026Earth,
  author = {Akram, M. Rizwan and Ali, Arshad and Aftab, Naeem},
  title = {An Earth observation framework for reservoir water level estimation in a data-scarce transboundary basin for flood preparedness},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.104005},
  url = {https://doi.org/10.1016/j.ejrh.2026.104005}
}