Forecasts of boundary layer winds, temperature, and humidity from the Pennsylvania State University/NCAR Mesoscale Model (MM5) are evaluated over the Pacific Northwest in order to examine the impact of the boundary layer parameterization on model errors. An extensive network of regularly reporting radiosonde and surface observations is used to evaluate the mean errors and bias of fields from simulations initiated every 12 hours as part of an operational forecasting system. The goal is to improve predictions of quantities important to fire meteorology. The spatial variation of model skill and the sensitivity to horizontal resolution are discussed.

The boundary layer parameterization is an implementation of the MRF (Hong-Pan) scheme, with surface layer properties determined using the similarity theory approach of the Blackadar parameterization. Boundary layer properties with the model domain are highly affected by complex orography and the large spatial variability of surface properties, and the development of terrain-influenced mesoscale flows. The network of observation stations is dispersed throughout windward, mountainous, and intermountain regions, and covers a wide range of climate zones including coastal rainforest, high mountain, and desert. Attempts to separate the PBL parameterization errors from the large-scale forecast errors are described. Dominant failure and success modes are identified, and their relative frequency, severity, and suspected causes discussed.