Since June 2018, all operational configurations of the Integrated Forecasting System (IFS) have been coupled to an ocean/sea-ice model, which has substantially increased the computational cost of research and development activities at ECMWF. The purpose of this memorandum is to assess whether some aspects of model development and testing at ECMWF can be undertaken with lower resolution ocean
configurations without comprising the scientific integrity of the results. Although the higher resolution ocean configuration outperforms its lower resolution counterpart, there are many aspects of model testing and development for which the absolute model performance is less relevant than the difference (i.e. Δ) between two experiments. This study considers whether Δ estimated using a lower resolution ocean model (i.e. ΔLRO) can be considered an appropriate proxy for Δ estimated using a higher resolution
ocean model (i.e. ΔHRO) for a variety of deterministic and probabilistic metrics at different lead times. In general, ΔLRO is an extremely good proxy for ΔHRO in medium-range forecasts (days 1-15), provided that ocean initial conditions used in lower and higher resolution systems are as consistent as possible. ΔLRO is also good proxy for ΔHRO in extended-range forecasts (weeks 1-4), particularly for metrics that are computed from forecast anomalies relative to a hindcast climatology. At extended-range lead times, we find a stronger sensitivity to changes in atmospheric resolution (50 km to 31 km) than changes in ocean resolution (100 km to 25 km). At seasonal (and longer) timescales, ΔLRO is a useful proxy for ΔHRO for some metrics (e.g. changes to the model climatology), but the approximation begins to break down due to a divergence of SST biases in lower and higher resolution systems. Finally, although lower resolution ocean configurations will continue to be a useful tool for research and development purposes at ECMWF, there will always be cases where it is necessary to use the highest resolution ocean model that is affordable.