Selección de miembros en un ensemble de predicción estacional

Abstract

At the moment, AEMET is working on the development of an objective seasonal forecast system to improve the current subjective procedure. We designed a prototype to improve the skill of seasonal forecast models in the Mediterranean region by selecting meaningful members from the Copernicus multimodel ensemble. We found that using the most recent model run did not always provide us with a higher skill than older initializations. Using an old run, on the other hand, allowed us to evaluate how the different scenarios proposed by the model were evolving and compare them with observations. We explored two alternative ways: one clustering the scenarios for the target season and comparing them with the most recent observed data to find which cluster was closer to observations in its evolution. The second one consisted in selecting a fixed number of members for the hindcast and the forecast, choosing those closer to the observations. Although the skill did not increase in the whole region, there were local improvements which could be useful for future predictions.