18 April 2019

The Effect of Ensemble Weighting

Most of the time, weighting does not and cannot change ensemble mean projections.

Science

We explored the effect of weighting a regional climate model (RCM) ensemble using metrics developed for RCM performance evaluation. In most cases, when metrics are used for weights they do not improve ensemble mean bias or change ensemble mean projections; additionally, we showed that weights often would not be able to change the mean projections regardless of weighting scheme.  

Impact

The question of whether or not to weight a model ensemble often comes up when working with ensembles. We've proved that one should think twice before weighting, as it may not make a difference.  

Summary

We explored the effect of weighting a regional climate model (RCM) ensemble using metrics developed for RCM performance evaluation. In most cases, when metrics are used for weights they do not improve ensemble mean bias or change ensemble mean projections. Additionally, we showed that any weighting scheme would often not be able to significantly change the mean projections regardless of weighting scheme. We also note that the universally-applicable RCM performance metrics used in the weighting scheme often did not significantly differentiate the models, and that the model differentiation produced by the metrics did not always agree with an in-depth, process-level analysis. 

Contact
Melissa S. Bukovsky
National Center for Atmospheric Research (NCAR)
Publications
Bukovsky, MS, JA Thompson, and LO Mearns.  2019.  "Weighting a Regional Climate Model Ensemble: Does It Make a Difference? Can It Make a Difference?"  Climate Research 77(1): 23-43, pp. 23-43.  https://doi.org/10.3354/cr01541.