Author(s): M. T. Vu; T. Aribarg; S. Supratid; S. Y. Liong; S. V. Raghavan
Linked Author(s):
Keywords: Statistical downscaling; Global Climate Models; ANN; PCA; Bangkok
Abstract: The damage caused by the 2011 devastated flood event in Bangkok raised a valid worrisome concern as to whether future rainfall will be much more severe with the changing climate. This study focuses rainfall projection through downscaling of global climate models (GCM), specifically for Bangkok (Thailand), using the Artificial Neural Network (ANN). ANN is an established technique with a flexible mathematical structure that is capable of identifying complex non-linear relationships between input and output data. The study illustrates the applications of the feed forward back propagation using large-scale predictor variables derived from both the ERA-Interim reanalysis data and present day/future GCM data. The predictors are first selected over different grid boxes surrounding the Greater Bangkok region and then filtered through using Principal Component Analysis to arrive at the most correlated predictors for ANN training. An emission scenario A1B from the Assessment Report 4 (AR4) and its closest counterpart, RCP4.5, from AR5 were considered; 3GCMs (CCSM, ECHAM, MIROC) were downscaled. The projection shows a marked increase in the annual rainfall over Bangkok by the end of the 21st century. The same was observed for the extreme rainfalls. The statistical indices show wetter climate in the wet season and drier climate in the dry season. These findings are tremendously useful to the policy makers to revisit whether the current drainage network system is sufficient to meet the projected heavier rainfall and to plan ahead for a range of flood adaptation and mitigation measures. The study also compared results of the downscaling with ANN with those resulting from the dynamical downscaling method, driven by the same GCMs, conducted in an earlier study.
Year: 2014