Quantile regression with partial least squares in statistical downscaling for estimation of extreme rainfall

Abstract

Extreme rainfall can lead to flooding and causes various disadvantages, such as crop failure in the field of agriculture. An analysis to examine extreme events is required to minimize these adverse impacts. Extreme rainfall can be analyzed by statistical downscaling (SD) model which is the functional relationship between local scale variable as the response variable (rainfall) and global scale variables as the explanatory variables (precipitation of global circulation). The purpose of this research is to develop SD model with quantile regression to predict extreme rainfall and partial least square to reduce dimension of explanatory variables. Extreme rainfall modeling is developed using linear, quadratic and cubic quantile regression respectively at 75 th , 90 th , and 95 th quantiles. However, these models were not give good results. The models were improved by dummy variables. It was shown that cubic quantile regression model with the addition of dummy variables can predict extreme rainfall especially at 95 th quantile in February, which was closed to the