An evaluation of high-resolution modeling and statistical forecast techniques over complex terrain

The accuracy of weather forecasts produced during the 2002 Olympic and Paralympic Games (23 Jan--25 Mar 2002) by a multiply nested version of the PSU-NCAR Mesoscale Model (MM5) and associated model output statistics (MOS) system is evaluated using observations collected by the MesoWest cooperative network. Using traditional verification measures, the accuracy of MM5 wind and precipitation forecasts improved as model grid spacing decreased from 12 to 4 km. In contrast, temperature forecasts did not improve with decreasing model grid spacing due to the inability of the MM5 to properly simulate the evolution of persistent or nocturnal cold pools over lowland regions. Improved parameterization of the stable boundary layer processes may enable model skill to improve as grid spacing is decreased from 12 to 4 km. The need to improve model physics was also illustrated by the fact that MM5 MOS temperature, relative humidity, wind speed, and wind direction forecasts were considerably more accurate than those produced by either the 12- or 4-km MM5.; These results suggest that, contrary to studies in regions of broader topographic features, decreasing model grid spacing (<10 km) does improve model skill over the finescale topography of the Intermountain West. These improvements may be valuable for a variety of environmental applications including fire weather forecasting, air quality prediction, and transport and dispersion modeling. Nevertheless, large biases frequently limit the direct application of numerical model output for weather prediction of surface variables. Thus, it is recommended that traditional MOS or other statistical techniques based on high-density observations from the MesoWest cooperative networks be used to improve surface weather forecasts, including those generated by the National Weather Service Interactive Forecast Preparation System (IFPS).