Time-lagged prediction of rainfall-runoff processes through black-box and regression techniques

Abstract

This paper focuses on the study of comparison among various black box and regression-based techniques in determining the rainfall-runoff relations. Without considering the temperature, topography or other parameters of the study area, simply using the data of rainfall and runoff to predict the future runoff is the key characteristic of these techniques. Ignoring the parameters of temperature, topography may lead to inaccuracy in forecasting future runoff, but the use of black-box or regression-based techniques provide a simple and fast way to determine the runoff. Three methods were compared in this study: autoregressive integrated moving average (ARIMA), regression, and artificial neural network (ANN). ARIMA model is a traditional approach to handling the rainfall-runoff model, which is highly fitted to the time series data. The main characteristic of this model is that, instead of considering two parameters, rainfall and runoff, it only uses one parameter, which is runoff, to forecast future runoff value. The multiple-linear regression model and the non-linear (quadratic) regression model comprise both parameters. The only difference between this two regression models is the order of parameter. Furthermore, artificial neural network (ANN) model are also studied to compare each method’s strengths and weakness.