Evaluating Chemistry-Climate Models using Metrics that Matter

Abstract:

Global atmospheric chemistry models are becoming a standard component of Earth system models. These chemistry-climate models have become more complete in terms of species and processes, more computationally expensive in terms of higher resolution, more complex in terms of running coupled historical simulations within an Earth system model, and thus far more difficult to evaluate in terms of skill in simulating the atmosphere's chemistry, e.g., the aerosols and greenhouse gases. With the large number of gas and aerosol species and the range in global conditions, one can become buried in the possible comparisons of models with observations. Collectively, we can respond to this intellectual challenge by building a set of skill-based metrics that test model performance in areas directly relevant to the major questions asked of the chemistry models, effectively, metrics that matter. Such metrics will focus on the chemical budget terms of greenhouse gases (methane, nitrous oxide, tropospheric ozone) and the physical-chemical processes controlling aerosol-cloud interactions. They must be statistical, have clean metrics (e.g., Taylor diagrams), and be applicable to climate models as opposed to hindcast modeling of specific meteorological events like many aircraft missions. The selection of metrics requires some design analysis, for example, matching Logan's classic ozonesonde observations is tied to what model property? local chemistry? stratospheric influx? long-range transport? Like the emergent constraint technique used in physical climate we need to develop a key set of observed/observable trace species or photochemical processes that relate to an important model budget or process that cannot be directly measured. Such chemistry-climate metrics need to become standard through successive model intercomparison projects (MIPs) to demark model evolution and improvement. This presentation will examine relational metrics that use seasonal and interannual variability in key species, as well as high-order metrics based on the nonlinearity of atmospheric chemistry. For example, we can compare budget terms calculated with observed species covariance versus those with the modeled ones.