An Optimized Bayesian Algorithm for Continuous Water Quality Monitoring at Different Locations in a Dam Reservoir Using Spatial and Temporal Pixels of Satellite Imagery

Introduction: Water quality monitoring in different locations of a dam reservoir is a challenging task. Manual and laboratory methods of water quality estimation require more manpower and time. Dam water quality needs to be monitored to increase agricultural productivity and reduce biodiversity loss. Objectives: In this paper, water quality in Nagi and Nagathi dam reservoirs located in Bihar, India are continuously monitored using spatial and temporal LandSat satellite image pixels. Methods: Pixels are correlated with laboratory measured water quality parameters such as pH value and Dissolved oxygen. For accurate measurement of water quality, Landsat image pixels are perspectively projected using the proposed Transverse Dyadic Wavelet Transform (TDyWT) algorithms. The pixels are enhanced with the proposed three numbers of optimised residual Deep learning algorithms (DnCNN): (i) Particle Swarm Optimization-DnCNN (PSO-DnCNN) (ii) Red-Kite Optimization Algorithm-DnCNN (ROA-DnCNN), and (iii) Fuzzy-DnCNN. The statistical parameters of these enhanced water and moss pixels, such as contrast, entropy, band values, PSNR and SNR are correlated with the laboratory measure values of DO and pH using Bayesian optimised Multilevel Regression (BO-MR) approach. Results: The BO-MR predicted water quality parameters in the different locations of the dam have an average accuracy of about 91% and 97%, when compared to ground truth verification. Conclusions: Continuous water quality prediction in Nagi and Nagathi dams located in Bihar, India is performed using the spatial and temporal LandSat satellite image pixels. PH value and Dissolved oxygen values are predicted using Landsat image pixels using BO-MR. TDyWT algorithms and Particle Swarm Optimization-DnCNN (PSO-DnCNN) , Red-Kite Optimization Algorithm (ROA-DnCNN) and Fuzzy-DnCNN enhanced pixels are used for the prediction of the DO and pH values from pixels of water and moss, respectively. In the proposed algorithms, the combination of TDyWT and ROA-DnCNN have higher accuracy in DO and pH prediction such as 91% and 97%, respectively. when compared to ground truth verification. Problems in continuous monitoring is solved using the proposed method. Further, to obtain accurate and representative data, frequent sample collection is not required. However, sampling collection is often done in accessible locations only in the reservoir, and miss out on localized pollution or anomalies in traditional methods are avoided in the proposed method. Further, human errors in conducting analyses and recording the observations/data can affect the quality/accuracy of the results during traditional methods and this is avoided in the proposed ROA-DnCNN method of water quality measurement. Furthermore, other water quality measurements can be predicted.