Assessing future rainfall uncertainties of climate change in Taiwan with a bootstrapped neural network-based downscaling model

Abstract

To investigate the impacts of climate change on Taiwan, a downscaling model (DSM) was used due to the large grid size of general circulation models (GCMs). DSM is a data-driven model based on the Radial Basis Function Neural Network (RBFNN). A Genetic Algorithm (GA) was adapted for parameter optimization, and the bootstrap method was employed to assess uncertainty. Two weather stations at similar latitudes but separated by mountains with altitudes of above 3000 m were selected as examples. Three GCMs were chosen for the model building and the assessment of near future (2050–2060) and far future (2080–2090) climate change impacts of three future scenarios A1B, A2 and B1. The results suggest that in the future, rainfall will tend to increase in winter but decrease in summer, with a similar average rainfall. In addition, our results suggest that in the future, typhoons might arrive later in the season.