River Water Temperature Modelling Under Climate Change Using Support Vector Regression

Abstract

Accurate River Water Temperature (RWT) is of great significance for the river water quality control and management. Various mathematical models were developed to study the stream temperatures based on the heat advection, transportation and equilibrium temperature concepts, which necessitates basic river hydro-meteorology, geometry, effluents and other vegetative data. In contrast, several data-driven models were developed, which are based on the most prominent input variables which define the RWT, such as air temperature and streamflows. The present study applied a Multiple Linear Regression Model (MLRM) and Support Vector Regression (SVR) model for the prediction of RWT at daily scale along Shimoga, Tunga-Bhadra river, a tributary of Krishna River, Karnataka, India. The results indicate that SVR model showed the best performance when compared with the linear regression model. The projected RWT under climate change was studied with the downscaled outputs from a statistical downscaling model, Canonical Correlation Analysis (CCA). The SVR model provides a promising reliable tool to predict the RWT and for analyzing the possible future projections under climate change.