The Center for Analysis and Prediction of Storms (CAPS) at University of Oklahoma produced storm-scale ensemble forecasts (SSEF) at convection-allowing 3-km horizontal grid spacing from April 18 through June 3 to support the NOAA 2016 Hazardous Weather Testbed (HWT) Spring Forecasting Experiment (NOAA/HWT SFE2016, formally from May 2 to June 3), and from June 20 through July 22 to support the NOAA 2016 Hydrometeorological Testbed (HMT) Flash Flood and Intensive Rainfall (FFaIR) Experiment (NOAA/HMT FFaIR2016, with the July 4th week off). Both WRF-ARW and NMMB numerical weather prediction models were run over a full CONUS domain, utilizing ARPS 3DVAR and Cloud Analysis to analyze all Doppler Weather Radar data and MADIS with Mesonet1 data feed, to form a 24-member SSEF ensemble (15 for HMT) of 60-h (36-h for NMMB and single physics ARW members) forecast initiated at 0000 UTC. Separately, an experimental cycled GSI plus EnKF data assimilation was performed on the same 3-km CONUS grid, with a 40-member WRF ensemble initiated at 1800 UTC, assimilating the RAP/HRRR GSI data stream (without Mesonet1) hourly from 1900 to 0000 UTC and all WSR-88D radar data over a one hour period (2300-0000 UTC) at 15 minute intervals. The final cycle (0000 UTC) analyses were used to initiate a 12-member WRF SSEF ensemble of 60-h long. In NOAA/HWT SFE2016, the CAPS non-cycled 3DVAR-based SSEF and the cycled GSI+EnKF based SSEF also contribute into a larger Community Leveraged Unified Ensemble (CLUE) coordinated among various groups including NSSL, SPC, CAPS, NCAR, UND, EMC, GSD, and DTC, in an effort to provide guidance to the design of near-future operational SSEF systems.
The NCEP Multi-Radar/Multi-Sensor (MRMS) QPE data were used to compute verification scores of CAPS SSEF QPFs and PQPFs. Results suggest that NMMB members underperform ARW's in overall in terms of ETS and ROC scores, while among the ARW members the single physics sub-ensemble (single physics with IC/LBC perturbations) is comaprable in QPF scores to the mixed sub-ensemble (multi-physics with IC/LBC perturbations) but with somewhat smaller spread than the later. Assimilating GSI data stream over a five hour period benefits the GSI+EnKF based ensemble, producing a much improved QPF skill compared to last year when only EnKF radar data cycling was performed. Subjective comparison of cases suggests that GSI+EnKF forecast MAY BE as good as 3DVAR’s or slightly better. A 12-day verification indicates mixed ETS scores between the two ensembles.
Exceedances of Flash Flood guidance (FFG) and Recurrence Intervals (RIs) in a neighborhood framework (with a 40-km radius) were computed for the NOAA/HMT FFaIR2016. Example products for the June 23 West Virginia Flood case will be examined in the conference.