Potential predictability of key marine biogeochemical fields: what can be learnt from cross-generation model comparison ?

The ‘perfect model’ approach allows to determine the uppermost limit of predictability by considering that models are unbiased and able to replicate all the real-world processes. While the perfect model approach and its utility in predictability studies has been highlighted for the physical components of coupled models, its use has not been questioned so far for Earth System components where a number of threshold processes are sensitive to model bias.

Here, we assess the potential predictability of the CNRM-CERFACS Earth System Models (ESMs) of first and second generation (CNRM-ESM1 and CNRM-ESM2-1) to investigate how far model performance may impact the multi-year (1-10 years ahead) predictability of ocean carbon fluxes, net primary productivity and oxygen minimum zone. We elaborate our analyses on 2 sets of decadal ‘perfect model’ predictions, carried out with both ESMs. These predictions consist in a set of initialised ensemble members (from 5 to 10 members) every five years from a multi-centennial-long preindustrial simulation performed with the corresponding ESM.

We show that the regions where ocean carbon fluxes display predictability are robust across the different generations of models. Nevertheless, our results also show that the predictability horizon substantially differs between both generations of models, in particular over those regions where both ESMs versions exhibit very different biases.