Daliakopoulos (2026) Monitoring reservoir storage using remote sensing and large language models

Identification

• Journal: Journal of Environmental Management
• Year: 2026
• Date: 2026-03-25
• Authors: Ioannis Ν. Daliakopoulos
• DOI: 10.1016/j.jenvman.2026.129400

Research Groups

• Department of Agriculture, Hellenic Mediterranean University, Heraklion, Greece
• Institute of Energy, Environment & Climate Change, Hellenic Mediterranean University, Heraklion, Greece

Short Summary

This study presents an innovative framework for monitoring reservoir storage using Sentinel-1 Synthetic-aperture radar (SAR) imagery, validated against quantitative storage values extracted from online media with the assistance of large language models (LLMs), demonstrating a scalable solution for data-scarce regions.

Objective

• To develop and validate a framework for monitoring reservoir storage using remote sensing (SAR) data, leveraging large language models (LLMs) to extract reference storage values from online media, particularly in regions lacking traditional gauging data.

Study Configuration

• Spatial Scale: Aposelemis Dam, Crete, Greece
• Temporal Scale: 10-year time series

Methodology and Data

• Models used: Google Earth Engine (for SAR processing), Custom Search Engine (CSE), Large Language Model (GPT-4o-mini), Kling-Gupta Efficiency (KGE), Root Mean Square Error (RMSE)
• Data sources: Sentinel-1 Synthetic-aperture radar (SAR) imagery, publicly available online media (for reference storage values)

Main Results

• The curated CSE/LLM pipeline successfully yielded 82 distinct date-storage pairs for validation.
• The SAR-derived storage estimates achieved high performance, with Kling-Gupta Efficiency (KGE) values ranging from 0.96 to 0.98 and Root Mean Square Error (RMSE) between 1.5 hm³ and 2.0 hm³ across orbit-specific configurations.
• Residual analysis indicated limited bias but highlighted sensitivity to SAR parameter choice (backscatter threshold and median-filter radius) and the quality of reference data.
• The combined approach of SAR with LLM-assisted, human-guided validation data provides a scalable and effective pathway for reservoir monitoring, especially in data-scarce regions.

Contributions

• Introduction of a novel framework that integrates Synthetic-aperture radar (SAR) remote sensing with large language model (LLM)-assisted data extraction from online media for reservoir storage monitoring.
• Development of a method to generate independent, quantitative reservoir storage reference time series from publicly available online media, overcoming limitations of sparse or non-public gauging data.
• Demonstration of a robust and scalable approach for validating satellite-derived water extent maps against non-traditional reference data, enabling reservoir monitoring in data-scarce regions.

Funding

• Not explicitly stated in the provided text.

Citation

@article{Daliakopoulos2026Monitoring,
  author = {Daliakopoulos, Ioannis Ν.},
  title = {Monitoring reservoir storage using remote sensing and large language models},
  journal = {Journal of Environmental Management},
  year = {2026},
  doi = {10.1016/j.jenvman.2026.129400},
  url = {https://doi.org/10.1016/j.jenvman.2026.129400}
}