Genetic programming for the downscaling of extreme rainfall events on the East Coast of Peninsular Malaysia

Abstract

A genetic programming (GP)-based logistic regression method is proposed in the present study for the downscaling of extreme rainfall indices on the east coast of Peninsular Malaysia, which is considered one of the zones in Malaysia most vulnerable to climate change. A National Centre for Environmental Prediction reanalysis dataset at 42 grid points surrounding the study area was used to select the predictors. GP models were developed for the downscaling of three extreme rainfall indices: days with larger than or equal to the 90th percentile of rainfall during the north-east monsoon; consecutive wet days; and consecutive dry days in a year. Daily rainfall data for the time periods 1961-1990 and 1991-2000 were used for the calibration and validation of models, respectively. The results are compared with those obtained using the multilayer perceptron neural network (ANN) and linear regression-based statistical downscaling model (SDSM). It was found that models derived using GP can predict both annual and seasonal extreme rainfall indices more accurately compared to ANN and SDSM.