Predicting Water Quality Using Ensemble Machine Learning Models and Remote Sensing Data

Abstract

Sustainable management of water resources can hardly be done without water quality prediction. Conventional field-based surveillance approaches are, in most cases, subject to spatio-temporal barriers. A combination of ensemble machine learning (ML) methods and remote sensing data provides a practical and resilient mechanism of forecasting water quality aspects at large spatial scales. In this paper, an ensemble-based predictive model is proposed, which can make use of the remote sensing indicators and the historical water quality data to estimate such important parameters as Biological Oxygen Demand (BOD), Dissolved Oxygen (DO), pH and turbidity. The ensemble modeling method bettered and outweighed single models in accuracy, robustness and generalization and entailed random forest (RF) gradient boosting (GB) and extreme gradient boosting (XGBoost). The technique proposed was tested over satellite images and the in-situ water quality observations of several river basins of India. The findings show that predictive variables in remote sensing like surface temperature, NDVI, and land use are strongly correlated with the water quality indicator hence predictive capacity.