Premium
Comment on ‘Can multi‐model combination really enhance the prediction skill of probabilistic ensemble forecasts?’
Author(s) -
Weigel A. P.,
Bowler N. E.
Publication year - 2009
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1002/qj.381
Subject(s) - probabilistic logic , ensemble forecasting , computer science , context (archaeology) , range (aeronautics) , gaussian , probabilistic forecasting , stochastic modelling , econometrics , machine learning , artificial intelligence , mathematics , statistics , geography , composite material , materials science , physics , archaeology , quantum mechanics
This note refers to the study of Weigel et al. (2008), where the success of multi‐model ensemble combination has been evaluated with a Gaussian stochastic toy model. The authors concluded that multi‐models can outperform the best participating single models, but only if the single model ensembles are under‐dispersive. Here we introduce two improved versions of the toy model of Weigel et al. For one of these models the combination of well‐dispersed (i.e. reliable) forecasts can improve the prediction skill, but for the other model this possibility is excluded. It is argued that, for normally distributed variables, the first model may be applicable in the context of short‐ and medium‐range forecasting, but the latter may be more appropriate to seasonal forecasting. Copyright © Royal Meteorological Society and Crown Copyright, 2009.