P71 Assessing weather model and decision tool product performance

Raby, John, and Gail Vaucher, United States Army Research Laboratory

The critical task of battlefield weather forecasting has transitioned from a human forecaster located in-theater to computerized Numerical Weather Prediction (NWP) with the human forecaster perhaps located far from the area of interest. Weather forecast validation has always been required, with the emphasis now shifting to a greater extent from the skill scores of human forecasters to the accuracy of the NWP models.

The Army Research Laboratory (ARL) utilizes a high-resolution configuration of the WRF model that employs a data ingest method called Four Dimensional Data Assimilation. This modeling system (WRE-N) can ingest weather observations to improve the accuracy of the forecasts. To show the value-added of these forecasts over those produced by the standard WRF initialization grid, it is necessary to make statistical comparisons between weather observations and the WRE-N forecast fields.

The validation of high resolution weather models has proven to be a difficult challenge when accounting for the small time and space scales involved which limits the applicability of traditional verification techniques. Newer fuzzy and spatial techniques require gridded observation data which are not available at the high resolutions in which the Army is interested. The accuracy of the high resolution ARL Nowcast (WRE-N) is assessed by making comparisons of WRE-N forecasts with point-based and gridded weather observations. The NCAR Model Evaluation Tools (MET) are used to evaluate model performance using traditional, fuzzy and spatial verification techniques. A technique has been developed for applying the best available gridded observation datasets for use with fuzzy and spatial verification tools.

The FY13 goal is to develop a technique which will enable the use of available gridded observation datasets for fuzzy and spatial verification of the ARL WRE-N forecasts which operate at Army-relevant resolutions of 1km or less.