Analysis of Precipitation Pattern in Selected Cities of Pakistan using Penalized Quantile Regression

Abstract

the present study is aimed to downscale the precipitation using Quantile Regression and penalized Quantile regression. Statistical downscaling of precipitation is required as a major aspect of numerous environmental change studies. Statistical downscaling based on regression models needs one to sample from the conditional distribution to preserve the variance of observed precipitation. By using the given techniques will eliminates the need for some of the assumptions required in standard linear regression, including the assumption of normallydistributed errors with constant variance. The Quantile Regression model are flexible in a sense that they are selecting different set of predictors for different part of distribution. Two penalties LASSO and SCAD are used to Quantile Regression for selection of predictor variables. We have illustrated this method through an application to two weather stations in Pakistan. In this study we have used idea of “amalgated” Quantile Regression to improve the prediction performance.