MM5 precipitation verification over the Pacific Northwest during the 1997-99 cool seasons

Precipitation forecasts from the Pennsylvania State University-National Center for Atmospheric Research fifth-generation Mesoscale Model (MM5) are verified for the 1997-98 and 1998-99 cool seasons over the Pacific Northwest. The MM5 precipitation at 36-, 12-, and 4-km horizontal resolution is compared with over 250 NOAA cooperative observer, snow telemetry (SNOTEL), avalanche, and National Weather Service sites in order to evaluate the effects of increasing horizontal resolution and to document spatial variations in model skill. A noticeable improvement in bias, equitable threat, and root-mean-square (rms) error scores occurs as horizontal resolution is increased from 36- to 12-km resolution; however, going from 12- to 4-km resolution improvements in skill are restricted to the heavy precipitation events (>5.08 cm in 24 h). For light to moderate precipitation events, both the 12- and 4-km domains have significant overprediction over the upper windward slopes of the higher terrain. In contrast, for heavy precipitation events there is no widespread overprediction, and all resolutions underpredict precipitation in the lowlands and major gaps in the Cascades. The cool-season undercatchment for selected storage gauges (non-SNOTEL) are estimated using collocated wind, temperature, and precipitation data with undercatchment regression relationships. The undercatchment generally ranges from 25%-40% for exposed high-elevation sites (>1500 m), 15%-20% within major gaps in the Cascades (∼1000 m), to 5%-15% in the lowlands. Undercatch corrections mitigate but do not eliminate the 4-km overprediction problem at low to moderate thresholds. Since synoptic forecast errors preferentially penalize high-resolution forecasts, less accurate large-scale forecasts are eliminated by setting strict criteria for the 850-mb wind errors at the Quilayutte, Washington, and Salem, Oregon, soundings sites. For all horizontal resolutions there is a significant increase in model skill with such screening (i.e., at 36-km resolution with screening, rms errors are equal to those at 12 km without screening); however, even with screening the 4-km forecasts still have larger rms errors than 12-km predictions for many precipitation thresholds.