When designing a set of climate projections for a given amount of supercomputing, the choice of resolution affects the ensemble size that can be afforded. The larger sample size helps to quantify uncertainties more robustly and provide better statistics to understand the projections. The trade-off is that the use of coarser resolution risks some compromise on quality in representing processes that could play a crucial role in the climate response. But what exactly is the added value of the enhanced resolution? We use two tools, which are applicable to a broader set of applications, to assess this for processes (teleconnections) that drive the year-to-year variability of late winter circulation over the North Atlantic. First, a causal network, which captures the teleconnections that relate key drivers of variability to late winter North Atlantic Oscillation (NAO), is elicited from experts in seasonal forecasting. Causal Inference theory is then used to estimate the teleconnection strengths. Second, we use two parallel perturbed parameter ensembles (PPEs) of coupled control simulations based on HadGEM3-GC3 at low and medium atmosphere/ocean resolutions. Combining the two tools allows us to identify teleconnections which have strengths (a) controlled by parameters, and (b) that are altered in a systematic way across parameter space by the enhanced resolution. Overall, the medium resolution is marginally better than the lower resolution, notably in the way El Niño-Southern Oscillation (ENSO) affects the Indian Ocean Dipole. For many of the teleconnections, most model configurations have weaker relationships than observed. Several teleconnections are shown to have modest parametric effects that produce correlated changes across the two PPEs. However, the ensemble size would need to be increased to better identify the key parameters and processes.
Read full abstract