Smart Monitoring of Water Supply Treatment Plant Processes using Remote Sensing and Artificial Neural Networks

Abstract

The increasing global population continues to be a major threat to the world’s ecological resources including lakes and rivers as people continue to clear the environment to create settlement spaces. Effective monitoring and management of a country’s water resources is critical for sustainable water supply systems. Nonetheless, the conventional water quality monitoring (WQM) approach is laborious, time consuming, and costly. The main objective of this study was to evaluate the potential application of estimated water quality parameters (WQPs) from Landsat-8 OLI satellite data to determine the optimum coagulant dose for water treatment using Artificial Neural Network (ANN) models. The specific objectives of the study were: to estimate the concentration of Turbidity, total suspended solids (TSS), and Chlorophyll-a (Chl a) from Landsat-8 OLI in correlation with in situ water quality data using empirical multivariate regression modelling (EMRM); to assess the spatial distribution and variability of the estimated and in situ WQPs for the selected period, and to use ANN modelling in predicting treated WQPs, and for determining the optimum coagulant dose for a water supply treatment plant. The study used satellite images and EMRM to estimate Chl-a, TSS, and Turbidity concentrations at different points in a water supply reservoir using same season data. Ordinary Kriging was used in the development of spatial maps showing the distribution and variability of the Landsat-predicted and in situ WQPs. The extracted spectral reflectance values from satellite images were then used as input for the first ANN model to predict treated WQPs, and in the second ANN model to predict the optimum coagulant dose required for water treatment. The results of the study show that, for all the samples, Turbidity, TSS, and Chl-a were estimated with R2 values of 0.76, 0.81, and 0.81, respectively. The ANN model 1 for the prediction of treated WQPs had dependable accuracy with R2 values of 0.99, 1.00, and 0.87 in predicting Turbidity, TSS, and Chl-a, respectively. Respectively, the in situ turbidity, TSS, and Chl-a for data collected in November 2020, December 2020 and January 2021 was (7.38 NTU, 7.08 NTU, and 8.62 NTU), (271.15 mg/L, 281.42 mg/L and 281.17 mg/L), and (37.17 mg/L, 50.86 mg/L and 44.75 mg/L) against a Landsat estimated turbidity, TSS, and Chl-a of (7.44 NTU, 6.25 NTU, and 7.99 NTU), (268.17 mg/L, 279.89 mg/L 285.07 mg/L), (37.44 mg/L, 52.35 mg/L and 49.73 mg/L) for the specific months respectively. On the other hand, the ANN model 2 also had a high accuracy in predicting the optimum coagulant dose an R2 value of 0.99. The actual coagulant dose for all the sampling days was 40 mg/l against the second ANN model’s optimum coagulant dose of 39.95 mg/l for all the sampling days. Based on the results, the study concluded that satellite data products can be used for the retrieval of reservoir WQPs with reasonable accuracy. Furthermore, the ANN models also highlight the possibility of using extracted spectral reflectance values for water quality predictions, and optimizing water treatment plant operations. Since the study used same season data, it did not account for the temporal variability in WQPs. It is recommended that the model accuracies, and dependability be improved by using fairly extensive datasets collected at different seasons of the year. The concept will also increase the likelihood of using the models for water quality predictions in other reservoirs within, and outside the catchment.