Neuro-fuzzy systems approach to infill missing rainfall data for Klang River Catchment, Malaysia

Abstract

Rainfall data can be regarded as the most essential input for various applications in hydrological sciences. Continuous rainfall data with adequate length is the main requirement to solve complex hydrological problems. Mostly in developing countries hydrologists are still facing problems of missing rainfall data with inadequate length. Researchers have been applying a number of statistical and data driven approaches to overcome this insufficiency. This study is an application of neuro-fuzzy system to infill the missing rainfall data for Klang River catchment. Pettitt test, standard normal homogeneity test (SNHT) and Von Neumann Ratio (VNR) tests were performed to check the homogeneity of rainfall data. The neuro-fuzzy model performances were assessed both in calibration and validation stages based on statistical measures such as coefficient of determination (R2), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). To evaluate the performance of the neuro-fuzzy system model, it was compared with a traditional modeling technique known as autoregressive model with exogenous inputs (ARX). The neuro-fuzzy system model gave better performances in both stages for the best input combinations. The missing rainfall data was predicted using the input combination with best performances. The results of this study showed the effectiveness of the neuro-fuzzy systems and it is recommended as a prominent tool for filling the missing data.