The rationale behind the success of multi-model ensembles in seasonal forecasting - I. Basic concept

Abstract

The DEMETER multi‐model ensemble system is used to investigate the rationale behind the multi‐model concept. A comprehensive documentation of the differences in the single and multi‐model performance in the DEMETER hindcast data set is given. Both deterministic and probabilistic diagnostics are used and a variety of analyses demonstrate the improvements achieved by using multi‐model instead of single‐model ensembles. In order to understand the reason behind the multi‐model superiority, basic scenarios describing how the multi‐model approach can improve over single‐model skill are discussed. It is demonstrated that multi‐model superiority is caused not only by error compensation but in particular by its greater consistency and reliability.

Weighting of NMME temperature and precipitation forecasts across Europe

“…Overall, the skill of our multimodels is similar to that of other multi-model weighting techniques such as equal weights Hagedorn et al, 2005;Slater et al, 2017), multiple linear regression , other Bayesian-based approaches (Rajagopalan et al, 2002;Robertson et al, 2004;Weigel et al, 2008), optimal weights (Wanders and Wood, 2016;Weigel et al, 2008) or genetic algorithms (Ahn and Lee, 2016). However, it is difficult to compare these multi-models in detail as most have been applied over different spatial and temporal resolutions, and often verified using different evaluation metrics.…”

“…The predictive skill of these equally weighted multi-models tends to be greater than or equal to the skill of the best model within the ensemble Hagedorn et al, 2005;Ma et al, 2015a;Slater et al, 2017;Thober et al, 2015;Wood et al, 2015). Generally, multi-model ensembles can outperform single-model ensembles when the individual models are overconfident, so the multi-model widens the ensemble spread and reduces the average ensemble-mean error (Weigel et al, 2008).…”

“…Existing NMME multi-model approaches have mostly used equal weighting schemes, giving the same weight to each single-model ensemble (i.e., the mean of each model's members) or to all of the individual members, irrespective of their skill Hagedorn et al, 2005;Slater et al, 2017;Tian et al, 2014). The predictive skill of these equally weighted multi-models tends to be greater than or equal to the skill of the best model within the ensemble Hagedorn et al, 2005;Ma et al, 2015a;Slater et al, 2017;Thober et al, 2015;Wood et al, 2015).…”

Decadal Climate Prediction: Opportunities and Challenges

“…This sensitivity is associated with the uncertainty in subgrid scale parameterized physics and model numerics. The recognition of the importance of this sensitivity has led to a number of efforts that have demonstrated that a multi-model ensemble strategy is the best current approach for adequately resolving forecast uncertainty and the forecast probability distribution in seasonal-tointerannual predictions ( [65], [66], [67], [68], [69]). Another recently proposed methodology is to use stochastic-dynamic parameterization techniques which perturb parameterizations in such a way as to improve on the benefits of a multi-model ensemble by using a single model [70].…”

Change in the precipitation intensity of the East Asian summer monsoon projected by CMIP3 models

“…Multi-Model Ensemble (MME) average can be expected to outperform individual models in case of present-day climate simulations (Lambert and Boer 2001;Gleckler et al 2008;Reichler and Kim 2008) as well as seasonal forecast (Palmer et al 2004;Hagedorn et al 2005). In case of global warming projections, IPCC (2007) summarizes the performance of models and future change of climate in terms of MME approach.…”