Satellite-Based Reservoir Water Monitoring for Irrigated Agriculture in Uruguay

Irrigated agriculture accounts for over 70% of global water consumption, with rice being the most significant irrigated crop in Uruguay, covering 140,000 to 160,000 hectares annually. Approximately half of the irrigation water comes from reservoirs, while the remainder is pumped from rivers and lagoons. However, continuous monitoring of water volumes and flows in irrigation systems is constrained by the high costs of traditional methods, limiting water use planning, efficiency improvements, and equitable water distribution.

Satellite imagery has emerged as a cost-effective tool for natural resource monitoring. Since 2010, platforms like Google Earth Engine have provided free access to geospatial data, enabling environmental analysis without the need for advanced software or hardware. Sentinel-2 (S2) is part of the European Union’s Copernicus Earth Observation program. These satellites are equipped with multiband passive sensors offering 10-30m spatial resolution and a 5-day revisit period, allow the calculation of water indexes like NDWI and MNDWI to measure water surfaces and estimate volumes. However, their performance is influenced by climatic and atmospheric conditions. Sentinel-1 (S1) satellites, with radar sensors providing 10m spatial resolution and a 6-day revisit period, offer all-weather, day-and-night monitoring.

This study was conducted between 2018 and 2024 focused on the "India Muerta" reservoir in Uruguay, using S2 and S1 imagery processed via Google Earth Engine through Google Colab Python scripts. Water surfaces were generated at 20 cm intervals based on the reservoir's digital elevation model and field sensor data, creating a multiband raster. 

For S2 image collection, a filter of at least 80% cloud-free coverage was used, applying additional filtering to ensure 70% cloud-shadow-free pixels over the area of interest. NDWI thresholds (-0.4 to 0.4) were tested to minimize errors and improve accuracy, while S1 imagery used Otsu algorithm to fit the most accurate reflectance thresholds for water detection.

The results showed that variable S2 NDWI thresholds outperformed the S1 Otsu-based detection method, achieving higher accuracy (R² = 0.88 vs. 0.77), lower mean absolute error (MAE = 7.92 vs. 13.43), and lower root mean square error (RMSE = 12.76 vs. 17.15). These findings highlight the benefits of adaptive NDWI thresholds for accurately estimating inundated areas and water volumes compared to radar-based methods.

Satellite-based reservoir monitoring provides critical data for both policymakers and farmers. For governments, it facilitates the identification and planning of reservoirs, ensuring equitable water use. For farmers, it offers a reliable tool for optimizing irrigation and improving water management. Furthermore, it helps managing  irrigation shortages  and addresses water scarcity challenges in present and future irrigated agriculture. This approach represents a cost-effective alternative to traditional monitoring methods, bridging the gap in continuous water resource management in many regions.