Abstract
While numerical weather prediction models have made considerable progress regarding forecast skill, less attention has been paid to the planetary boundary layer. This study leverages High-Resolution Rapid Refresh (HRRR) forecasts on native levels, 1-s radiosonde data, and (primarily airport) surface observations across the conterminous United States. We construct temporally and spatially averaged composites of wind speed and potential temperature in the lowest 1 km for selected months to identify systematic errors in both forecasts and observations in this critical layer. We find near-surface temperature and wind speed predictions to be skillful, although wind biases were negatively correlated with observed speed and temperature biases revealed a robust relationship with station elevation. Above ≈250 m above ground level, below which radiosonde wind data were apparently contaminated by processing, biases were small for wind speed and potential temperature at the analysis time (which incorporates sonde data) but became substantial by the 24-h forecast. Wind biases were positive through the layer for both 0000 and 1200 UTC, and morning potential temperature profiles were marked by excessively steep lapse rates that persisted across seasons and (again) exaggerated at higher elevation sites. While the source or cause of these systematic errors are not fully understood, this analysis highlights areas for potential model improvement and the need for a continued and accessible archive of the data that make analyses like this possible.
Significance Statement
We employed high vertical resolution radiosonde and near-surface observations to evaluate an operational numerical weather prediction model, the High-Resolution Rapid Refresh (HRRR), focusing on the lowest 1 km above ground level. This layer of the atmosphere plays a critical role in overall forecast skill and uncertainty as processes within in must be parameterized. Our analysis critically assessed not only the model forecasts but also the observations, and we determined that radiosonde wind information to be contaminated below about 250 m above the ground. Our verification revealed consistent biases with respect to wind speed and also between temperature and elevation, thereby identifying several areas for model improvement and highlighting the value of high-resolution observations in the boundary layer.
© 2020 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).