The Comparison of Bayesian Model Averaging with Gaussian and Gamma Components for Probabilistic Precipitation Forecasting

摘要

Ensemble forecasting has relatively good predictive abilities, especially in the field of climatology. However, the results of ensemble predictions are often underdispersive or overdispersive. Therefore, it is necessary to calibrate the ensemble forecasting. The Bayesian model averaging (BMA) method with gaussian or gamma distribution is commonly used to calibrate ensemble forecasting. This research reveals that there have been extreme precipitation in the observed periods. This has an impact on the pattern of rainfall that is asymmetric but has a longer tail on one side. This research examined the monthly rainfall data in Juanda Station in East Java Province generated by the North American Multi-Model Ensemble (NMME) models and further calibrated them with BMA. The purpose of this research is to assess the performance of the calibration results using the BMA Gaussian and Gamma. Both calibration results were evaluated using the continuous range probability score (CRPS) and the percentage of captured observations. The calibration with BMA Gaussian produced an average CRPS of 8.27 with 58.16% coverage, while with BMA-Gamma an average CRPS of 7.23 with 62.11% coverage was obtained. This result suggests using BMA-Gamma to generate more accurate probabilistic forecasts.