Open AccessLiu-type regression in statistical downscaling models for forecasting monthly rainfall salt as producer regions in Pangkep regency
Author(s) -
St Sahriman,
Anisa Anisa,
Vieri Koerniawan
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1341/9/092021
Subject(s) - multicollinearity , statistics , mathematics , mean squared error , variance inflation factor , regression analysis , downscaling , regression , collinearity , precipitation , geography , meteorology
Liu-Type Regression (LTR) is one of the statistical methods to overcome multicollinearity in multiple regression models. LTR is the development of Ridge regression and Liu estimator. When there is a strong collinearity, selected k parameter in the ridge regression does not fully overcome the multicollinearity. This study aimed to estimate the rainfall data in Pangkep Regency (as response variable) with LTR approach on Statistical Downscaling (SD) models. Precipitation (as predictor variables) is the result of a simulation of a grid on the Global Circulation Model (GCM). This study uses a size 8 8 grid of GCM (64 predictor variables) over an area of Pangkep Regency so that there is a high multicollinearity. Three dummy variables were determined from k-means cluster technique used as predictor variables to overcome the heterogeneity of residual variance. LTR model with dummy variables are able to explain the diversity of rainfall data properly. The value of R 2 produced ranges 85.23% -88.99% with Root Mean Square Error (RMSE) ranges 117.732-136.377. Validation of the model generates a high correlation value between the actual rainfall and alleged rainfall period of 2017 (about 0.977-0.979). The value of Root Mean Square Error Prediction (RMSEP) produced lower (about 57.625-61.120). SD analysis was also performed with and without the dummy variable in the Ridge regression and LTR. In general, LRT models with dummy (k = 0.652, d = -0.799) is the best model based on the value of R 2 , RMSE, correlation, and RMSEP.