Using the Resolution Dependence of Modeled Extreme Event Fidelity to Drive Model Development: Model Devaluation within the CASCADE SFA

Wednesday, May 14, 2014 - 07:00
Add to Calendar

A number of recent analyses have shown strong resolution dependence (e.g., of clouds, convection, and extreme precipitation) in the CESM that appear to originate in the atmospheric physics parameterization suite, and we have recently shown that the resolution dependence of clouds and precipitation is physically inconsistent with the scaling properties of the atmosphere. We broadly classify these resolution dependence issues as configuration-inconsistencies, which includes inconsistencies related to vertical resolution, temporal resolution, and numerical implementation details such as sub-stepping. We argue that configuration-consistency should be a fundamental target for model development that will ultimately lead to broad improvements in model fidelity and that will allow broad use of variable-mesh capabilities. Toward this model development paradigm, we provide an overview of a systematic, cyclic, hindcast based framework through which to evaluate the configuration-consistency of the ACME model: the InitiaLized-ensemble Identify, Analyze, Develop (ILIAD) framework. In the Initialized-ensemble stage of the framework, we run a large ensemble of short-term hindcast simulations with the ACME model (we are presently using CESM)--in a manner similar to CAPT (Phillips et al., 2004)--at multiple resolutions. We use a hindcast-based framework in order to assess whether the model produces extreme events at inappropriate times (e.g., extreme events resulting from forcing that would not produce extreme events in reality) or whether it misses extreme events that should occur. Such information is invaluable for assessing why model fidelity changes with resolution. In the Identify stage, we evaluate whether and how the fidelity of the simulations changes with horizontal resolution. In the Analysis stage, we apply systematic and process-based analysis methods to determine the model component(s) responsible for any configuration-inconsistency. In the Develop stage, we will ultimately collaborate with developers in the SciDAC Multiscale project and the ACME SFA to design and test modifications to the ACME model aimed at improving configuration-consistency. We present preliminary results applying the ILIAD framework to CESM to examine the resolution dependence of modeled extreme precipitation fidelity.

Presentation file(s):