Estimates of artificial ensemble mean ocean properties from individual simulations and observations to better isolate the atmospheric influence on ocean variability

Ensemble ocean model experiments can be useful to understand the extent to which internal variability has exerted an influence on a given observation, or indeed modelled event. This is important in contexts such as that of ongoing climate change, for example, where this probabilistic information can be useful for the purposes of detection and attribution. However, ensemble simulation has certain disadvantages, including its very high computational and energetic cost, the technical skill required to implement such modelling strategies, and the inherent dependence of the results on model physics. The aim of this study is to address these drawbacks by directly estimating the ensemble mean using statistical methods applied to individual model simulations, or observations. The effects of internal variability should be strongly reduced in these artificial ensemble mean estimates, enabling better insight into the direct effects of atmospheric forcing on the chosen ocean variables.

In previous work, we showed that the ensemble mean sea surface height can be estimated with good accuracy by filtering an individual member of the ensemble. Here, we extend this result to sea surface temperature (SST), which requires a more complicated spatiotemporal filter to estimate the ensemble mean, but again shows good agreement with the true ensemble mean SST at very low computational cost. However, examination of the full 3D temperature fields show a more complicated spectral coherence signature, suggesting that application of the filtering method to these 3D fields would be more challenging. In a second step, a neural network is thus trained to reproduce 3D ocean temperature fields using SST and sea surface height as inputs. By combining the filtered fields with the neural network, first estimates are made of the ensemble mean 3D temperature field, based on observations. Comparisons with the true ensemble mean 3D fields are encouraging, and suggest that the method may be useful as a cheap alternative to numerical simulation to better identify the atmospheric influence on ocean variability.