An Intelligent Multimodal System for Regional Water Quality Monitoring and Management Based on Remote Sensing and GIS

Abstract

Water pollution is a major threat to the waterways and ecosystems. The tradi tional approaches to monitor the quality of water involve manual sampling and subsequent laboratory analysis. These conventional monitoring methods are lim ited by cost, accessibility, and reliability which in turn makes real-time monitoring impractical. Furthermore, the dependence on stationary monitoring stations lim its the comprehensive assessment of water quality across large bodies of water. Moreover, the sources of pollution are diverse, including agricultural and urban runoff, sewage, and industrial discharges. In this regard, a multimodal data collec tion approach for monitoring water quality in the Rawal watershed is proposed in the thesis. This approach includes the use of remote sensing, reanalysis, and a GIS approach to collect physicochemical, hydrological and topographical, and air pol lutants and meteorological parameters. Machine learning and deep learning tech niques are employed to analyze the impact of physicochemical parameters on water quality and to forecast parameter concentrations. The results show that forecast ing multi–step parameter concentrations is possible with bi–directional LSTM with a Root Mean Square Error (RMSE) of 0.2 (mg/L) and 281.741 (μS/cm) for dis solved oxygen and electric conductivity, respectively. Regression analysis predicts physicochemical water quality feature concentrations with LightGBM, Support Vector Regressor, and Multilayer Perceptron in the range of 0.015 to 0.18 RMSE. Moreover, the impact of hydrological and air pollutant parameters on water qual ity is analyzed using correlation and regression analysis. The results reveal that parameters such as turbidity, temperature, pH, and dissolved oxygen relate to soil type parameters, and the NO2 air pollutant is correlated to turbidity, to tal dissolved solids, pH, and dissolved oxygen parameters. Furthermore multimodal indexing method, Enhanced Water Quality Index (EWQI), based on a hybrid remote sensing and semi–supervised machine learning approach is pro posed to remove the uncertainties in traditional indexing by a 100% classification rate and high–weighting parameters that include electric conductivity, secchi disk depth, dissolved oxygen, lithology, and geology. The Rawal watershed is given an overall “Medium” water quality classification, which can vary by season. In conclusion, the proposed techniques can enhance the socio–economic environment that depends on an appropriate standard of water quality for its development and can serve as a guideline for applications in other drinking water reservoirs. The multimodal approach to water quality monitoring can provide a global coverage of sample collection, reduce the cost and time required for water quality monitoring, and improve the accuracy and reliability of the results obtained.