Enhancing Subseasonal Temperature Prediction by Bridging a Statistical Model With Dynamical Arctic Oscillation Forecasting

Abstract

This study proposes a hybrid approach to improving subseasonal prediction skills by bridging a conventional statistical model and a dynamical ensemble forecast system. Based on the perfect prognosis method, the phase of the Arctic Oscillation (AO) from the European Centre for Medium-range Weather Forecasts ensemble forecast system is used as a predictor in a composite based statistical model to predict the wintertime surface air temperature in the Northern Hemisphere. The hybrid model, which employs AO phases predicted by the dynamical model for weeks 1-4, generally outperforms the conventional statistical model for lead times of weeks 2-6. The improved skill score is due to the high accuracy of the AO forecast from the dynamical model and the strong lagged connection between the AO and temperature. This study thus lays the groundwork for the potential use of combining climate variability, statistical relation, and dynamical forecasting. Plain Language Summary Climate prediction from 1 week in advance to a season is challenging despite increasing social demand and scientific interest for accurate and dependable predictions. Statistical climate prediction relies on observed relationship within the climate data, while dynamical climate prediction utilizes numerical models that is built based on physical principles. Here, we suggest combining those two types of models to enhance the prediction accuracy for the surface air temperature at weeks 2-6 ahead. The hybrid model is built upon a statistical phase model using one of major climate variabilities of Northern Hemisphere winter, so called the Arctic Oscillation (AO). Conventional AO phase model relies on the observed lagged relationship between the AO and the surface air temperature. In the hybrid model, information of forecasted AO from a dynamical model is provided to a statistical model to improve the accuracy and reliability of the conventional statistical model for all forecast lead times. This approach thus lays the groundwork for the potential use of combining climate variability, statistical relation, and dynamical forecasting.