The study is based on an evaluation of the Pennsylvania State University/National Center for Atmospheric Research Mesoscale Model MM5 and Model Output Statistics (MOS) System used for fine-scale weather prediction over Northern Utah during the 2002 Olympic and Paralympic Winter Games. MOS forecasts were generated for low-, mid-, and high-elevation sites at Olympic venues and along transportation corridors, some less than 1 km apart. Our evaluation suggests that model skill does not improve as grid spacing is decreased below 12 km. One contributor to large model errors was an inability to accurately simulate the strength and duration of persistent and nocturnal cold-pool events. Results also showed that statistical techniques (e.g., MOS) provided substantially better forecasts than purely numerical predictions even at high resolution. For example, MOS temperature, relative humidity, wind speed, and wind direction forecasts were more skillful than forecasts produced by the parent MM5 at 12- and 4-km grid spacing, and equally or more skillful than manual forecasts issued by the Olympic Forecast Team. An extension of this validation work using 200 stations over the complex terrain of Northern Utah, including 65 precipitation sites, is currently underway and will also be presented.
The National Weather Service will be producing high-resolution gridded forecasts beginning in October 2003 (e.g., the Interactive Forecast Preparation System (IFPS)). Results from our study suggest that increasing the resolution of numerical models to match the 2.5-km grid spacing of IFPS forecasts issued by the NWS Western Region, will add little, if any, value over the coarser resolution model. Instead, statistical techniques based on high-density observations, such as those provided by MesoWest cooperative networks, should be used to improve gridded forecast products created by the IFPS and other applications.