Future Rainfall Projections in Cisadane Watershed using Statistical Downscaling Technique on Global Climate Model (GCM) Outcome

Abstract

The changes of pattern and duration of the rainy season due to climate change affect water availability for agriculture sector, particularly in the Cisadane watersheds. Climate change projections particularly on local scale rainfall in the region are greatly needed. In this case the output of Global Circulation Model (GCM) can be used to project the local scale rainfall by using downscaling techniques. The result showed that the downscaling models developed by using simple regression analysis has a relatively small determination and correlation coefficients respectively ranging from 0 to 0.13 and from –0.183 to 0.366. In order to obtain better results, Principle Component Analysis (PCA) was used to transform and reduce the variables from the GCM outputs namely CSIRO, GFDL, and CGCM3. Six principle component (PC1-PC6) were selected from each GCM data, and were then used to build multi-linear models with observation data. This analysis is called principal component regression (PCR). It is found that PCR reproduced better results for almost all GCMs in comparison with the simple regression results. This study indicates that CSIRO produces better downscaling model compared to other GCMs used in this study. This can be seen from the highest coefficient of determination (0.533) and correlation (0.73) on Citeko Station resulted by the model, which was followed by CGCM3 and GFDL at the same station. Model validation using root mean square error (RMSE) and correlation showed that the resulted downscaling models based on PCR are significant at 95% confidence level. Future projections on each GCM based on the average of all stations showed an increase in mean rainfall values compared to the baseline data, where the highest percentage of increase (around 21.63%) was found in CGCM3. Meanwhile, based on the average of all GCMs, Station Citeko was projected to have the highest increase in future rainfall (11.6%). The average percentage of all increases between current and future rainfall is 9.04%. If seperated based on season, each model showed an increase of mean rainfall value during wet season and a decrease during dry season, except for CGCM3.