Ice Sheet Speed-dating: Using Expert Elicitation to identify “good” simulations of the LGM North American Ice Sheets

After running a large ensemble of palaeo ice sheet model runs, it is common to either rank the simulations or determine which simulations are an acceptable match to observations and which are ruled out. This task requires human judgement, which is usually left only to the authors of the research. Tools have been developed to compare ice sheet simulations to empirical reconstructions numerically, but even this approach requires some human input on values for match thresholds.

An alternative is to use expert elicitation to identify “good” ice sheet simulations. Expert elicitation normally captures expert’s beliefs in the form of a probability distribution; for something as complicated as ice sheet geometry this is much too arduous a task. Instead, we propose to elicit binary classifications of “good” and “bad” and find descriptions of plausible ice sheets through probabilistic inverse modelling. Experts can consider empirical ice sheet reconstructions, but also “soft-knowledge” about the sectors of the ice sheet it is most important to match, margin shapes considered to be glaciologically plausible, and an idea of the reasonable best-reconstruction a model will be able to provide. By seeking the input of many experts, it is possible to both lighten the task load of determining the quality of 100-1000s of simulations, and gain a wisdom of the crowd benefit to the results. Just like any other method of ranking ice sheet simulations, this method requires human judgement; in this case more explicitly than usual.

We are seeking expert input to rank an existing ensemble of North American Ice Sheet simulations. By asking experts at EGU 2023 to spend 3-5 minutes sorting simulations using an online interface we will build up an average community view on which LGM North American Ice Sheet simulations are “good”. This will provide a community resource to compare future ice sheet simulations against that is a justifiable representation of academic expert knowledge, adding to the current arsenal of model-data intercomparison tools.