Estimation of seasonal and annual river flow volume based on temperature and rainfall by multiple linear and Bayesian quantile regressions

Abstract

Investigation of river flow volume in different conditions as a function of temperature and rainfall variables can be quite effective in understanding the hydrological and hydro-climatic conditions of the watershed. Multiple linear regression models were applied in estimating river flow in several studies due to their straightforwardness and appropriate interpretation of results. In this study, to overcome the limitations of the multiple linear regression model, the Bayesian quantile regression model was used to estimate the river flow volume as a function of rainfall and temperature, and the results were compared. The data and information used for the Qareh-Sou basin in northern Iran are of substantial environmental and socio-economic importance. Five data series, including spring, summer, autumn, winter, and annual series, were created and used for this study. It was found that the Bayesian quantile regression model has considerable flexibility to model the volume of flow for different quantiles, predominantly upper and lower quantiles, and can be used to model high and low flows. With increasing the values of quantiles, a limited decreasing pattern in the effect of rainfall on the volume of flow was identified, which can be due to increasing the effect of other factors in the formation of extreme flows of the river. For summer data in high quantiles, the effect of rainfall on river flow volume shows an increasing pattern. This pattern is different from the other studied series, which may be due to the low base flow in summer. The results confirm that the application of Bayesian quantile regression compared to multiple linear regression leads to much more valuable information on the impact of rainfall and temperature on river flow volume.