Journal Article
Weather and Forecasting, 2021
Authors
Robert M. Banta, Yelena L. Pichugina, Lisa S. Darby, W. Alan Brewer, Joseph B. Olson, Jaymes S. Kenyon, S. Baidar, S.G. Benjamin, H.J.S. Fernando, K.O. Lantz, J.K. Lundquist, B.J. McCarty, T. Marke, S.P. Sandberg, J. Sharp, W.J. Shaw, D.D. Turner, J.M. Wilczak, R. Worsnop, M.T. Stoelinga
Abstract
AbstractComplex-terrain locations often have repeatable near-surface wind patterns, such as synoptic gap flows and local thermally forced flows. An example is the Columbia River Valley in east-central Oregon-Washington, a significant wind-energy-generation region and the site of the Second Wind-Forecast Improvement Project (WFIP2). Data from three Doppler lidars deployed during WFIP2 define and characterize summertime wind regimes and their large-scale contexts, and provide insight into NWP model errors by examining differences in the ability of a model [NOAA’s High-Resolution Rapid-Refresh (HRRR-version1)] to forecast wind-speed profiles for different regimes. Seven regimes were identified based on daily time series of the lidar-measured rotor-layer winds, which then suggested two broad categories. First, in three regimes the primary dynamic forcing was the large-scale pressure gradient. Second, in two regimes the dominant forcing was the diurnal heating-cooling cycle (regional sea-breeze-type dynamics), including the marine intrusion previously described, which generates strong nocturnal winds over the region. The other two included a hybrid regime and a non-conforming regime. For the large-scale pressure-gradient regimes, HRRR had wind-speed biases of ~1 m s−1 and RMSEs of 2-3 m s−1. Errors were much larger for the thermally forced regimes, owing to the premature demise of the strong nocturnal flow in HRRR. Thus, the more dominant the role of surface heating in generating the flow, the larger the errors. Major errors could result from surface heating of the atmosphere, boundary-layer responses to that heating, and associated terrain interactions. Measurement/modeling research programs should be aimed at determining which modeled processes produce the largest errors, so those processes can be improved and errors reduced.
