Grimit, Eric P., 3TIER, Inc., Philip Regulski,
Clifford F. Mass, and Gregory J. Hakim, University of Washington
In 2011, 3TIER and the
University of Washington entered an arrangement for the joint evaluation an
operational version of its 64-member ensemble Kalman filter (EnKF) data
assimilation system, which uses DART software in combination with version 3.0.1
of the WRF model. The initial aim
of the project was to evaluate the impact that this could have on reducing wind
power forecast errors compared to existing operational NWP models generated by
public sources and at 3TIER. A
synoptically active period over the Pacific Northwest (April 10-30, 2011), with
generally moderate to strong westerly and northwesterly flow, was chosen for a
retrospective forecast experiment due to the presence of significant timing and
intensity errors in the existing operational NWP models. The 21-day period was re-simulated
using 3-hourly data assimilation cycles with UWÕs regional observation data
sets. 3TIER then executed its wind
power forecast system based on the ensemble mean forecast at three
representative wind facilities in the region.
Timing and intensity
errors associated with fronts and other major weather features that are present
in control WRF runs with no data assimilation (cold-start) are improved more
often than not. Overall, a
reduction of about 0.5 m/s in mean absolute wind speed forecast errors is
achieved at two sites. A noticeable
improvement in wind power ramp event detection, as defined by 3x3 contingency
table scores, is observed at all three facilities. Using the RUC model as a benchmark forecast that employs
data assimilation, the verification scores achieved by the UW-3TIER system are
far superior.