Constraining Arctic Climate Projections: A Process-Based Approach to Model Weighting

Significant uncertainties in projections of various ocean and sea ice variables stem from a variety of sources, including different modeling approaches, imperfect representations of physical processes, and natural variability. Multi-model ensembles like CMIP6 are essential for assessing the range of uncertainty, however they rely on "model democracy," which assumes all models are equally plausible and independent of one-another.Various constraining and weighting approaches are in use to minimize model uncertainties. Most of these approaches focus on state quantities, often relying solely on historical simulations of the target variable itself as the primary diagnostic. Here, we want to use more process-based diagnostics to incorporate physical mechanisms and interactions that govern the system dynamics. Previous assessments of the historical Arctic's energy budget in CMIP6 have shown tight connections between oceanic heat transports and key Arctic state quantities like sea ice and the ocean's warming rate, with substantial biases prevailing from the ocean to the Arctic surface. Using our new StraitFlux tools, which enable fast and precise calculations of oceanic transports for diverse climate models, we can quite efficiently incorporate oceanic transports into existing model weighting algorithms. By evaluating model performance against observational data and assessing their independence of one-another, we aim to identify and mitigate biases in Arctic projections. We use this approach to weight and constrain key Arctic variables, such as sea ice, for a large ensemble of CMIP6 models. For example, weighting the Arctic September sea ice extent ensemble reduces the spread in the first year of an ice-free Arctic and indicates a general tendency to an earlier ice-free Arctic than when using model democracy. Those results agree very well with past studies using different weighting diagnostics, demonstrating the robustness of the weighting approach.