J3.2 Seasonal Dependence of RAP and HRRR NWP Forecast Errors During the Second Wind Forecast Improvement Project (WFIP2) Field Campaign

Wednesday, 25 January 2017: 10:45 AM
606 (Washington State Convention Center )
James Wilczak, NOAA, Boulder, CO; and I. V. Djalalova, L. Bianco, J. B. Olson, J. Kenyon, R. Banta, L. K. Berg, J. Bickford, T. A. Bonin, A. Brewer, M. Brewer, S. Bustos-Ortola, A. Choukulkar, C. T. Clack, K. L. Clawson, A. Clifton, J. W. Cline, D. Cook, R. Coulter, R. Eckman, H. J. S. Fernando, K. Friedrich, E. P. Grimit, L. S. Leo, J. K. Lundquist, T. J. Martin, M. Marquis, J. McCaa, K. McCaffrey, P. Muradyan, M. Pekour, Y. Pichugina, W. J. Shaw, M. T. Stoelinga, and S. Wharton

The Second Wind Forecast Improvement Project (WFIP2) is a Department of Energy and NOAA-led program whose goal is to improve the accuracy of NWP forecasts of wind speed in complex terrain.  WFIP2 includes a field campaign held in the vicinity of the Columbia River Basin in the Pacific Northwest of the U.S., which began in October 2015, and will continue through March, 2017.  As part of WFIP2 a large suite of in-situ and remote sensing instrumentation has been deployed.  NWP forecast errors from the 13 km resolution Rapid Refresh (RAP), 3km High Resolution Rapid Refresh (HRRR), and 0.75km HRRR-Nest are analyzed on a seasonal basis using the observations collected during WFIP2.  These observations include a network of three 449 MHz radar wind profilers (RWP’s) with RASS, eight 915 MHz RWP’s with RASS, 18 sodars, 4 profiling microwave radiometers, vertical profiles from 8 lidars, a network of 10 microbarographs, and many surface meteorological stations.  Diurnal time height cross-sections of the model’s mean bias and RMSE for wind speed, direction, temperature, and humidity, as well as boundary layer depth, are evaluated for each of the three models, providing a holistic view of model accuracy at simulating boundary layer structure.  Model errors are analyzed as a function of season (3 month averages) and location, and show the impact of increasing model resolution on forecast skill.  Seasonal averages of model biases and RMSE provide more robust results than do shorter case study episodes, and can be used to verify that model errors found in shorter case study episodes are in fact representative. The results are used to identify specific model weaknesses and the corresponding parameterization schemes that are in greatest need of improvement.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner