Water quality estimation using machine learning and deep learning neural networks

The study explores the use of machine learning (ML) models to estimate chlorophyll-a levels in waterbodies using two different approaches. The first method employed real-time observations of water quality factors, such as temperature, turbidity, and salinity, to estimate the amount of chlorophyll-a present. A feed forward artificial neural network (ANN) was used for modeling, and the best-fitted network exhibited R2values of 0.92 for the spring season and 0.55 for the fall season for Atchafalaya and Terrebonne River Basins. The second approach involved predicting water quality characteristics, such as turbidity, chlorophyll-a concentration, temperature, and salinity, from Sentinel-3 hyperspectral satellite images with 21 bands using deep convolutional neural networks (CNNs). The results demonstrate that deep CNNs can predict the aforementioned water quality parameters, with R2 values ranging between 0.6 and 0.9. These findings indicate that ML can be applied for ecological surveys, water quality surveillance, and water resource management.