Baird Langenbrunner and J. David Neelin
Journal of Advances in Modeling Earth Systems,submitted Sept. 2016.
Supplement (6.4 MB).
Abstract Global climate models (GCMs) are examples of high-dimensional input-output systems, where model output is a function of many variables, and where an update in model physics commonly improves performance in one objective function (performance metric) at the expense of degrading another. Here, concepts from multiobjective optimization in the engineering literature are used to investigate parameter sensitivity and optimization in the face of such tradeoffs. Metamodels introduced in Part I are leveraged in the context of multiobjective optimization in order to improve GCM simulation of the tropical Pacific climate, focusing on precipitation, column water vapor, and skin temperature. An evolutionary algorithm is used to solve for Pareto fronts, which are surfaces in objective function space along which tradeoffs in model performance occur. Pareto fronts allow the modeler to visualize these tradeoffs quickly and identify the physics at play. In some cases Pareto fronts are small, implying that tradeoffs are minimal, optimal parameter combinations are more straightforward to choose, and the GCM is well-functioning. In all cases considered here, the control run was found not to be Pareto-optimal (i.e., not located on the Pareto front), highlighting an opportunity for model improvement through objectively informed parameter selection. Taylor diagrams illustrate that these improvements occur primarily in field magnitude and not spatial correlation, and they help identify parameter updates that improve fields fundamental to tropical moist processes-namely precipitation and skin temperature-without significantly impacting others. These results provide an example of how basic elements of multiobjective optimization can facilitate pragmatic GCM tuning processes.
Citation B. Langenbrunner and J. D. Neelin, 2015: Multiobjective constraints for climate model parameter choices, Part II: Pragmatic Pareto fronts in CESM1. Journal of Advances in Modeling Earth Systems, submitted Sept. 2016.