Hydroclimatic streamflow prediction using Least Square-Support Vector Regression

Abstract

In last two decades, many Artificial Intelligence-based models have gained recognition in the field of hydrologic forecasting. In this study, potential of least square support vector regression model is explored in the context of streamflow prediction using hydroclimatic inputs. This study is conducted in Narmada river basin up to Sandia gauging station and Mahanadi river basin up to Basantpur gauging station. Four different hydroclimatic variables, namely rainfall, maximum temperature, minimum temperature and streamflow values of previous day are used as input variables. Prediction performances are assessed in terms of different statistical measures, namely – correlation coefficient, root mean square error and Nash–Sutcliffe efficiency. Whereas these statistics indicate an overall impressive result, in particular, low and medium ranges of streamflows are found to have better correspondence between observed and predicted values.

Keywords:

• kernel-based learning
• least square support vector regression (LS-SVR)
• streamflow modelling
• Narmada River
• Mahanadi River

Acknowledgement

This article belongs to the selected papers presented at the ‘Hydro-2012’ conference held at IIT Bombay on 7–8 December 2012 and short-listed by Editor for publication in this Journal after re-review and revisions where necessary.