Calibration and combination of seasonal precipitation forecasts over South America using Ensemble Regression

Abstract

Models participating in the North American Multi Model Ensemble project were calibrated and combined to produce reliable precipitation probabilistic forecast over South America. Ensemble Regression method (EREG) was chosen as it is computationally affordable and uses all the information from the ensemble. Two different approaches based on EREG were applied to combine forecasts while different ways to weight the relative contribution of each model to the ensemble were used. All the consolidated forecast obtained were confronted against the simple multi-model ensemble. This work assessed the performance of the predictions initialized in November to forecast the austral summer (December–January–February) for the period 1982–2010 using different probabilistic measures. Results show that the consolidated forecasts produce more skillful forecast than the simple multi-model ensemble, although no major differences were found between the combination and weighting approaches considered. The regions that presented better results are well-known to be impacted by El Niño Southern Oscillation.