Framework for an Ocean‐Connected Supermodel of the Earth System

François Counillon,Noel Keenlyside,Shuo Wang,Marion Devilliers,Alok Gupta,Shunya Koseki,Mao‐Lin Shen

DOI: https://doi.org/10.1029/2022ms003310

IF: 8.469

2023-03-01

Journal of Advances in Modeling Earth Systems

Abstract:A supermodel connects different models interactively so that their systematic errors compensate and achieve a model with superior performance. It differs from the standard non‐interactive multi‐model ensembles (NI), which combines model outputs a‐posteriori. Supermodels with Earth system models (ESMs) has not been developed because it is technically challenging to combine models with different state space. Here, we formulate the first supermodel framework for ESMs and use data assimilation to synchronise models. The ocean of three ESMs is synchronised every month by assimilating pseudo sea surface temperature (SST) observations generated by them on a common grid to handle discrepancies in grid and resolution. We compare the performance of two supermodel approaches to that of the NI. In the first (EW), the models are connected to the equal‐weight multi‐model mean, while in the second (SINGLE), they are connected to a single model. Both versions achieve synchronisation in the ocean and in the atmosphere, where the ocean drives the variability. The time variability of the supermodel multi‐model mean SST is reduced compared to observations, most where synchronisation is not achieved and is lower‐bounded by NI. The damping is larger in EW, for which variability in the individual models is also damped. Hence, under partial synchronisation, the unsynchronized variability gets damped in the multi‐model average pseudo‐observations, causing a deflation during the assimilation. The SST bias in individual models of EW is reduced compared to that of NI, and so is its multi‐model mean in the synchronised regions. A trained supermodel remains to be tested. Supermodeling is a novel approach in which different models are run interactively and are connected to achieve a model with superior performance. The method exploits model diversity and reduces model error by training the connection terms using observations. Structural differences between the models and the amount of data exchange among models are technical challenges and have limited the applicability of supermodeling to Earth system models (ESMs). Here, we show that data assimilation can handle these issues. A demonstration is done with three ESMs having different resolution and structural differences and using a limited amount of data exchange (monthly ocean surface temperature). Synchronisation is achieved in several key regions. There, the ocean surface temperature error is smaller than by taking the a‐posteriori average of the non‐interactive multi‐model ensemble, an approach commonly used for handling Coupled Model Intercomparison Project simulations. It also modifies atmospheric teleconnections. Connecting the model via their weighted mean (one of the two approaches tested) causes a spurious deflation of variability. This study opens the application and training of supermodeling to ESMs. Data assimilation can be used to synchronise different Earth system models Partial synchronisation is achieved and bias is reduced in key regions by exchanging monthly sea surface temperature Synchronising the models toward a weighted‐mean damps variability Data assimilation can be used to synchronise different Earth system models Partial synchronisation is achieved and bias is reduced in key regions by exchanging monthly sea surface temperature Synchronising the models toward a weighted‐mean damps variability

meteorology & atmospheric sciences