Temperature Forecast Using Ridge Regression as Model Output Statistics

Abstract

Over the past few years, BMKG (Meteorological, Climatological and Geophysical Agency) in Indonesia has used numerical weather forecasting techniques, namely Numerical Weather Prediction (NWP). However, the NWP forecast still has a high bias because it is only measured on a global scale and unable to capture the dynamics of atmosphere (Wilks, 2007). Hence, this study implements Ridge Regression as Model Output Statistics (MOS) for temperature forecast. This study uses the maximum temperature (Tmax) and minimum temperature (Tmin) observation at 4 stations in Indonesia as the response variables and NWP as the predictor variable. The results show that the performance of the model based on Root Mean Square Error of Prediction (RMSEP) is considered to be good and intermediate.  The RMSEP for Tmax in all stations is intermediate (0.9-1.2), Tmin in all stations is good (0.5-0.8). The prediction result from Ridge Regression is more accurate than the NWP model and able to correct up to 90.49% of the biased NWP for Tmax forecasting.