Modeling Water Quality Parameters in Laga Dadi Reservoir, Ethiopia: An Integrated Remote Sensing and Machine Learning Approach

Abstract Reservoirs that supply drinking water need close monitoring. That's especially true in places where routine field sampling can't keep up with how fast conditions shift. We looked at Laga Dadi Reservoir in Ethiopia, using satellite imagery and cloud-based processing to estimate three water quality indicators: dissolved oxygen, nitrate, and turbidity. Field sampling is still the gold standard for accuracy. But it's slow, costly, and hard to scale across a reservoir this large. To get around that, we combined laboratory-analyzed field samples with Sentinel-2 imagery processed in Google Earth Engine. The water body was delineated using object-based classification in eCognition, and cloud-free composites were built to keep the imagery clean. Regression models in RStudio picked up strong relationships between spectral reflectance and each parameter. Nitrate performed best (R² = 0.99, RMSE = 0.0025 mg/L), followed by dissolved oxygen (R² = 0.968, RMSE = 0.042 mg/L) and turbidity (R² = 0.914, RMSE = 19.34 NTU). Remote sensing and machine learning proved to be a practical, lower-cost way to monitor reservoirs where field sampling can't keep up and it has direct relevance for Ethiopia and similar regions to manage their water supplies.