AI and satellite data join forces to estimate water quality in Amazon reservoirs with high accuracy and minimal fieldwork.
Smart Water Monitoring in Remote Areas
How can we assess water quality in remote, vast, and sensitive areas without collecting physical samples every time? The answer comes from an intelligent system that combines satellite data and artificial neural networks to estimate key parameters such as chlorophyll-a, transparency, and total suspended solids (TSS) in water reservoirs.
Satellite Imaging Across the Hydrological Cycle
At the heart of the method lies the use of multispectral images from the Landsat 7 satellite, covering the full annual hydrological cycle (flood, emptying, dry, refill). The visible and near-infrared bands are calibrated and atmospherically corrected to accurately estimate surface water reflectance.
Training Neural Networks with Real and Remote Data
These images feed into an artificial neural network trained on more than 30 satellite images and validated with laboratory measurements collected between 2007 and 2014. The model, tailored for each of the seven sampling points in a large Amazonian hydroelectric reservoir, was tested using robust Leave-One-Out cross-validation techniques.
Accuracy Across Seasonal Variations
The results? Performance improves during dry periods when cloud cover is minimal and satellite imagery is clearer. However, even during other hydrological cycles, the average estimation error remains low, making the system reliable for environmental monitoring.
Scalable and Non-Invasive Environmental Surveillance
This approach offers a dual advantage: it provides accurate estimates without the need for constant on-site measurements and enables large-scale continuous monitoring of reservoirs that are critical for ecological balance and water resource management. A concrete example of how artificial intelligence and satellite observation can together transform the way we observe and protect our ecosystems.
Autori
H. A. Nascimento Silva, G. Laneve, A. Rosato, M. Panella
