A barycenter-based approach for the multi-model ensembling of subseasonal forecasts

Le Coz, Camille; Tantet, Alexis; Flamary, Rémi; Plougonven, Riwal

doi:10.5194/gmd-19-8167-2026

Articles | Volume 19, issue 17

https://doi.org/10.5194/gmd-19-8167-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-19-8167-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 19, issue 17

Model evaluation paper

|

03 Sep 2026

Model evaluation paper |

| 03 Sep 2026

A barycenter-based approach for the multi-model ensembling of subseasonal forecasts

Camille Le Coz, Alexis Tantet, Rémi Flamary, and Riwal Plougonven

Data sets

Pre-processed data for "A barycenter-based approach for the multi-model ensembling of subseasonal forecasts" Camille Le Coz et al. https://doi.org/10.5281/zenodo.15038871

MERRA-2 statD_2d_slv_Nx: 2d,Daily,Aggregated Statistics,Single-Level,Assimilation,Single-Level Diagnostics V5.12.4 (M2SDNXSLV) Global Modeling and Assimilation Office (GMAO) https://doi.org/10.5067/9SC1VNTWGWV3

MERRA-2 inst1_2d_asm_Nx: 2d,1-Hourly,Instantaneous,Single-Level,Assimilation,Single-Level Diagnostics V5.12.4 (M2I1NXASM) Global Modeling and Assimilation Office (GMAO) https://doi.org/10.5067/3Z173KIE2TPD

MERRA-2 inst3_3d_asm_Np: 3d,3-Hourly,Instantaneous,Pressure-Level,Assimilation,Assimilated Meteorological Fields V5.12.4 (M2I3NPASM) Global Modeling and Assimilation Office (GMAO) https://doi.org/10.5067/QBZ6MG944HW0

Model code and software

Code for "A barycenter-based approach for the multi-model ensembling of subseasonal forecasts" Camille Le Coz et al. https://doi.org/10.5281/zenodo.15058503

Short summary

We explore an alternative framework for constructing multi-model ensembles by formulating ensemble combination as a barycenter problem. We compare the L₂ barycenter (equivalent to pooling) with the Wasserstein barycenter (more precisely its Gaussian approximation). Both have the same ensemble mean but differ in how they represent forecasts uncertainty. In terms of Continuous Ranked Probability Score, the Wasserstein barycenter outperforms more often while performing similarly on average.