Case Study Applying GIS Tools to Verifying Forecasts over a Mountainous Domain

The meteorological community makes extensive use of the Model Evaluation Tools (MET) developed by the National Center for Atmospheric Research for numerical weather prediction model verification through grid-to-point, grid-to-grid and object-based domain level analyses. MET Point-Stat has been used for grid-to-point verification; MET Grid-Stat has been used to perform grid-to-grid neighborhood fuzzy verification to account for the uncertainty inherent in high resolution forecasting; and, MET Method for Object-based Diagnostic Evaluation (MODE) has been used to apply object-based spatial techniques for the verification of high resolution forecast grids for continuous meteorological variables.

High resolution modeling requires more focused spatial and temporal verification over parts of the domain, especially when that terrain is mountainous. With a Geographical Information System (GIS), researchers can now consider terrain type/slope and land use effects and other spatial and temporal variables as explanatory metrics in model assessments. By augmenting a GIS with python code, one can also perform objective analysis, i.e., one based on Barnes (1994a, b, and c). A GIS, augmented by objective analysis techniques and coupled with high resolution point and gridded observation sets, allow location-based approaches that permit discovery of spatial and temporal scales where models do not sufficiently resolve the desired phenomena.

In this paper, we augment a case study given at this year's annual AMS meeting to show an ability to introduce measures of atmospheric state produced via, for example, a Barnes Objective Analysis. We use this technique in a study of bias (mean error) for a 9/3/1 km triple nested configuration of WRF-ARW whose innermost fine scale domain was centered over Southern California. Southern California contains a mixture of urban, sub-urban, agricultural and mountainous terrain types along with a rich array of observational data with which to illustrate our ability to conduct sub-domain verification.