Enhanced Water Quality Monitoring in Dam Reservoirs Using Optimized Vision Transformer and Bayesian-Optimized Support Vector Regression

Abstract

Introduction:  Monitoring water quality at various locations in a dam reservoir is a complex task that requires significant time and manpower when using traditional manual and laboratory methods. Regular monitoring is essential for enhancing agricultural productivity and preventing biodiversity loss.

Objectives:  This paper focuses on the continuous monitoring of water quality in the Nagi and Nagathi dam reservoirs in Bihar, India, utilizing spatial and temporal Landsat satellite image pixels.

Methods:  To improve prediction accuracy, Landsat images are perspectively projected using the Transverse Dyadic Wavelet Transform (TDyWT) and enhanced with Particle Swarm Optimized Vision Transformer (PSOViT) and Adam Optimized Vision Transformer (AdamViT) algorithms. Statistical features such as mean, entropy, PSNR, and band values extracted from enhanced water and moss regions are correlated with laboratory-measured values using Bayesian Optimized Support Vector Regression (BO-SVR). The proposed methods, PSOViT-SVR and AdamViT-SVR, are employed to predict four water quality parameters namely pH, dissolved oxygen (DO), total dissolved solids (TDS), and Conductivity at different locations within the Nagi and Nagathi dams.  

Results: Prediction accuracy of proposed method AdamViT-SVR followed by PSOViT-SVR is higher compared to existing methods.

Conclusions: Proposed methods achieved an average accuracy of 96% to predict pH, DO, TDS and Conductivity parameters when compared to ground truth verification.