A climate model projection weighting scheme accounting for performance and interdependence

Abstract

Uncertainties of climate projections are routinely assessed by considering simulations from different models. Observations are used to evaluate models, yet there is a debate about whether and how to explicitly weight model projections by agreement with observations. Here we present a straightforward weighting scheme that accounts both for the large differences in model performance and for model interdependencies, and we test reliability in a perfect model setup. We provide weighted multimodel projections of Arctic sea ice and temperature as a case study to demonstrate that, for some questions at least, it is meaningless to treat all models equally. The constrained ensemble shows reduced spread and a more rapid sea ice decline than the unweighted ensemble. We argue that the growing number of models with different characteristics and considerable interdependence finally justifies abandoning strict model democracy, and we provide guidance on when and how this can be achieved robustly.

The neglected Indo‐Gangetic Plains low‐level jet and its importance for moisture transport and precipitation during the peak summer monsoon

“…Thus, the correct expression of the IAM is limited by the ability of atmospheric general circulation models (AGCMs) to capture regional-scale circulation [Ashfaq et al, 2009;Johnson et al, 2015;Knutti et al, 2017]. This has proven to be challenging, but the availability of higher-resolution modeling has substantially improved the representation of the IAM in climate model simulations [Delworth et al, 2012;Ma et al, 2014;Johnson et al, 2015;Li et al, 2015;Shields et al, 2016].…”

A method to encapsulate model structural uncertainty in ensemble projections of future climate: EPIC v1.0

“…those generated by El Niño and La Niña events, into the stochastic weather model. The implementation of a model weighting scheme, such as that of Knutti et al (2017), for the training data could increase the applicability of the model.…”

“…Their method combines information from a perturbed physics ensemble (PPE), multi-model ensembles to capture model structural uncertainties, and observations. Since GCMs have been shown to not be structurally independent (Masson and Knutti, 2011;Knutti et al, 2013), multi-model ensembles benefit from model weighting to improve the ensemble performance (Knutti et al, 2017). The limitations of these methods are that large computer resources are required to run the ensembles of simulations required, which limits the ability to apply this method across many different greenhouse gas (GHG) emissions scenarios.…”

Providing future climate projections using multiple models and methods: insights from the Philippines

“…However, Weigel et al (2010) show that weighting does not necessarily improve predictive skill and models can perform well or poorly depending on the choice of metrics, stating that Bequal weighting may be the safer and more transparent way to combine models^. Yet Knutti et al (2017) argue that equal weighting may no longer be justifiable when considering both model performance and interdependence. A different approach to weighting is documented in McSweeney et al (2015) where particularly poor performing models are excluded and all other models considered equally plausible.…”