AbstractOne challenge with wind-power forecasts is the accurate prediction of rapid changes in wind speed (ramps). To evaluate the Weather Research and Forecasting (WRF) model's ability to predict such events, model simulations, conducted over an area of complex terrain in May 2011, are used. The sensitivity of the model's performance to the choice among three planetary boundary layer (PBL) schemes [Mellor–Yamada–Janjić (MYJ), University of Washington (UW), and Yonsei University (YSU)] is investigated. The simulated near-hub-height winds (62 m), vertical wind speed profiles, and ramps are evaluated against measurements obtained from tower-mounted anemometers, a Doppler sodar, and a radar wind profiler deployed during the Columbia Basin Wind Energy Study (CBWES). The predicted winds at near–hub height have nonnegligible biases in monthly mean under stable conditions. Under stable conditions, the simulation with the UW scheme better predicts upward ramps and the MYJ scheme is the most successful in simulating downward ramps. Under unstable conditions, simulations using the YSU and UW schemes show good performance in predicting upward ramps and downward ramps, with the YSU scheme being slightly better at predicting ramps with durations longer than 1 h. The largest differences in mean wind speed profiles among simulations using the three PBL schemes occur during upward ramps under stable conditions, which were frequently associated with low-level jets. The UW scheme has the best overall performance in ramp prediction over the CBWES site when evaluated using prediction accuracy and capture-rate statistics, but no single PBL parameterization is clearly superior to the others when all atmospheric conditions are considered.