Statictical Downscaling Modeling with Quantile Regression to Estimate Extreme Precipitation (A Case Study in Bangkir Station, Indramayu).

Abstract

Statistical downscaling (SD) is a technique used to model the relationship between global-scale data and local-scale data with statistics model. The global-scale outcomes of Global Circulation Model (GCM) are used as independent variables in SD. Various methods of SD include multiple regression analysis, principal component regression analysis and artificial neural networks. However, these methods can not accurately predict extreme events. Quantile regression can be used to detect extreme conditions, both extreme dry and extreme wet. The aim of this study was to predict the extreme event and its probabability. The data of independent variables used were monthly rainfall of the district Indramayu Bangkir station. Quantile regression method was used to predict extreme rainfall and logistic regression to estimate the chances of extreme events. Quantile regression models formed had correct prediction rate in the 90th quantile in February. The probability of extreme rainfall events in quantile 75 was high in November, December, January, February and March. However, the occurrence probability in quantile 90 and 95, was only high in February. The prediction and probability of extreme rainfall based on quantile regression models and logistic regression showed similar trend with the data pattern observed in extreme conditions.