Statistical downscaling with generalized Pareto distribution (Study case: Extreme rainfall estimation)

Abstract

Indonesia has tropical climate with small variation of temperature but quite large variation of rainfall. So the rainfall which is an essential climate element related to climate change has to be observed. Climate change may increase the incidence of extreme rainfall that affects flooding in farmland. In order to anticipate the occurrence of extreme rainfall, the information of rainfall forecast is required. Statistical Downscaling (SD) is a technique to model the relationship between global scale data and local scale data. Global Circulation Model (GCM) output is global scale data and rainfall is local scale data. GCM has characteristic non-linear, high dimension, and multicolinierity. These problem can be overcome by principal component analysis (PCA). One of the primary methods for estimating extreme rainfall is generalize Pareto distribution (GPD) regression based on a threshold. The objective of this study is SD modeling based on GPD to predict extreme rainfall. The result show that GPD models can predict extreme rainfall well. Monthly rainfall prediction in January and December show a higher value than the actual data, but predictions follow actual data pettern well, especially during extreme rainfall. February has the highest rainfall that occurred in 2008 with a value of 439 mm/month. This value can be estimated either by prediction on quantile 0.95.