An Overview of CAPS Storm-Scale Ensemble Forecast for the 2014 NOAA HWT Spring Forecasting Experiment
WRF-ARW V3.5.1 was used, with different microphysics and PBL schemes assigned for different members. In addition to the four existing two-moment microphysics schemes that are included in v3.5.1 (Thompson, Milbrandt-Yau, Morrison, and WDM6), a newly developed P3 (Predicted Particle Properties) microphysics by Morrison and Milbrandt was implemented and included in 2014 SSEF ARW ensemble members. A modified Milbrandt-Yau scheme addressing overly prediction of ice cloud anvil was also included. Model simulated radar reflectivity is computed within each individual microphysics algorithm. PBL schemes used include MYJ, MYNN, QNSE, YSU, as well as a modified YSU by Greg Thompson in an attempt to correct the overly dry and warm PBL issue of YSU. The AFWA HAILCAST algorithm was implemented in ARW members to predict hail size. Post-season verification shows that, in terms of ETS scores for QPF, different microphysics schemes including the new P3 scheme do not perform very differently, even though their synthetic GOES IR imagery can have big difference.
A major push in 2014 is the experimental EnKF based forecasting that includes a one hour EnKF cycling at 15 min interval from 2300 UTC to 0000 UTC following a 5-h 40-member ensemble forecast initiated from 1800 UTC, over the same CONUS domain as other regular SSEF. In order to provide an ensemble background for EnKF, a separate 4-km ensemble of 5-h forecasts, starting at 1800 UTC, with 40 WRF-ARW members is produced over the CONUS domain. This ensemble is configured with initial perturbations and mixed physics options to provide input for EnKF analysis. Each member uses WSM6 microphysics with different parameter settings. No radar data is analyzed for this set of runs. All members also include random perturbations with recursive filtering of ~20 km horizontal correlations scales, with relatively small perturbations (0.5K for potential temperature and 5% for relative humidity). EnKF analysis (cycling), with radar data and other conventional data, is performed from 23 to 00 UTC every 15 min over the CONUS domain, using as background the 40-member ensemble. A 12- member ensemble forecast (24h) follows using the 00 UTC EnKF analyses. In addition, two deterministic forecasts, one from the ensemble mean analysis and another from 3DVAR analysis, are also produced. Results show that the probability matched mean 3-hourly accumulated precipitation from the EnKF-based analysis outscores the HWT (3DVAR-based) for higher thresholds.