Monthly precipitation forecasting using rescaling errors

Abstract

Although various models have been developed for prediction and forecasting of time series in various engineering fields, there is no perfect model to forecast hydrologic time series. In recent decades, Artificial Neural Networks (ANNs) have been very common for prediction and forecasting of hydrologic time series because of their practicality in applications. This study proposed a post-process in an ANN model to improve the forecasting performance by rescaling the errors based on a correlation between observations trained data. The model proposed in this study was examined using precipitation data achieved from different four stations in the United States, and compared with the feedforward networks. It was observed that all error measures used in this study were improved through a rescaling post-process in the model for all stations. The strong point of the model lies in that the correlation approach is very easy to apply.