Statistical Downscaling Using the Kibria–Lukman Regression with Dummy Variables for Rainfall Forecasting

Abstract

Global Circulation Models (GCM) are widely used to project climate variables at the global scale. However, the relatively coarse spatial resolution of GCM outputs makes direct GCM-based climate forecasting generally less accurate at the local scale. To bridge this scale mismatch, this study applies statistical downscaling (SD). The main challenge in GCM-based SD is the large number of predictor (grid) variables that are strongly correlated, which induces multicollinearity and can reduce the stability of coefficient estimation. To address this issue, Kibria-Lukman Regression (KLR) is used, which is a shrinkage method that combines the variance-reduction property of Ridge Regression (RR) with the bias-control concept of Liu Regression (LR). This study aims to obtain the best SD model and to produce local rainfall forecasting in Pangkep Regency during the 2023–2024 testing period. The research stages included: (1) developing an SD model using KLR with the addition of dummy variables as additional predictors; and (2) evaluating the forecasts using testing data. The results showed that the KLR model with dummy variables provided the best performance, with an RMSE of 73.002 and a coefficient of determination (R²) of 94.069%. At the validation stage, the model also produced a high pattern agreement (correlation of 0.936) and a relatively low forecasting error (RMSEP of 94.614), and it outperformed other multicollinearity-handling approaches. Thus, the proposed model has the potential to serve as a tool for local-scale rainfall forecasting to support salt production planning in Pangkep Regency

Keywords

• Dummy variable
• Global Circulation Model
• Kibria–Lukman regression
• Multicollinearity
• Rainfall
• Statistical downscaling