Statistical Downscaling Model with Jackknife Ridge Regression and Modified Jackknife Ridge Regression to Forecast Rainfall

Abstract

Statistical downscaling (SD) is a transfer function that connects local scale rainfall data with global scale rainfall. Global-scale rainfall can be obtained from the Global Circulation Model (GCM) output. GCM simulates climate variables in the form of large-scale grids, causing a high correlation between the grids (multicollinearity). The methods used in SD modeling to overcome multicollinearity are Jackknife Ridge Regression (JRR) and Modified Jackknife Ridge Regression (MJR). The method is the development of the Ridge Regression (RR) method. This study aims to predict local rainfall data in Pangkep Regency (response variables) based on local scale GCM output rainfall data (predictor variables) with the JRR and MJR approaches. In addition, K-means cluster technique is used in determining dummy variables to overcome the heterogeneity of the various remaining models. Results using training data (1990-2017 period) show that the MJR method is better at explaining the diversity of data based on a higher R2 value (68%) and a lower Root Mean Square Error / RMSE value (165.57) than the JRR method (R2 amount is 67 and RMSE amount is 167.72). Model validation using data testing (2018 period) also shows the same results, namely MJR is better than JRR. Other than that, the addition of dummy variables can improve the accuracy of the model in estimating rainfall data. Adding a dummy variable to the model results in a high R2 (range between 94% -95%) with a lower RMSE value (range between 66.60-67.69).