Artificial neural networks as routine for error correction with an application in Singapore regional model

Abstract

This research presents an error correction scheme based on artificial neural networks, and demonstrates its application on water level forecast for the Singapore water. The error correction scheme combines the numerical model outputs with the in situ measurements on a two-step basis: (1) predicting the model errors at the measurement stations and (2) distributing the predicted errors to the nonmeasure-ment stations. Artificial neural networks are used in both error prediction and error distribution as the mapping function approximators. The efficiency of this scheme is tested on six water level stations in the Singapore regional model domain with four prediction horizons. The results show that this error correction scheme produces high-precision forecasts, and improves the forecast accuracy at both measurement and nonmeasurement stations. © 2011 Springer Science+Business Media, LLC.