Does increasing the resolution of numerical forecasts improve forecast accuracy over fine-scale Intermountain orography?

Forecasts produced for the 2002 Olympic and Paralympic Winter Games (23 Jan - 25 Mar 2002) by multiply nested version of the PSU-NCAR fifth generation Mesoscale Model (MM5) are examined to determine if decreasing horizontal grid spacing to 4-km improves forecast accuracy over the fine-scale topography of the Intermountain West. The verification is based on high-density observations collected by the MesoWest cooperative networks, including 200 wind and temperature sites and 100 precipitation sites across northern Utah.

In contrast to studies over the broader topography of the Pacific Northwest, wind and precipitation forecasts produced by the 4-km MM5 domain were more accurate (based on traditional measures) than those of its parent 12-km domain. The most significant improvements in wind speed forecasts occurred at night in valleys and lowland locations where the topography of the 4-km domain enabled more accurate simulation of nocturnal flows. Wind direction forecast improvements were most substantial at mountain sites where better resolution of the topography by the 4-km domain more accurately reflected the exposure of these locations to the free atmosphere. The 4-km domain also produced quantitative precipitation forecasts that were equally (small events) or more (large events) skillful than the 12-km domain. Precipitation bias errors varied substantially between the two domains since the representation of the narrow, steeply-sloped basin-and-range topography improved dramatically at 4-km grid spacing.

Curiously, the overall accuracy of temperature forecasts by the 4-km domain was not significantly better than that of the 12-km domain. This was due to an inability of the MM5 to properly simulate nocturnal and persistent cold pools within mountain valleys and the lowlands upstream of the Wasatch Mountains. Paradoxically, the added resolution of the 4-km domain exacerbated this model deficiency, resulting in poorer skill scores. At upper-elevations, which are typically above the cold pools, the 4-km domain was substantially more skillful.

These results illustrate that decreasing grid spacing (<10 km grid spacing) does improve wind and precipitation forecasts over fine-scale Intermountain topography. It is hypothesized that temperature would also show improvement if cold pool structure and processes were better represented. Implications for weather prediction over the Intermountain West are described.