The Importance of Being Uncertain: The Influence of Initial Conditions and Parameter Settings in Ocean Models

Our ability to make accurate predictions depends is limited by uncertainty of the model itself. The intrinsic uncertainty in the physical processes in ocean models can be introduced through, though not exclusively, in uncertainty in the initial conditions and values for empirical process parameters. This paper describes how we might quantify these contributions to overall model uncertainty. To illustrate, we use the metrics of ocean heat content and ocean mixing. Importantly, we test our assumptions about the uncertainty of the initial conditions and process parameter values by designing a robust, but limited, sampling scheme of the inputs and use these selected inputs to run a limited set of global Community Earth System Model (CESM) simulations. Using Bayesian statistics, the selective inputs and their associated outputs (the metrics) are then used to interpolate outcomes of the metrics to produce a broad distribution of uncertainty in a given metric. We show the relative importance of these types of uncertainty as compared to underlying internal variability of model.