92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Thursday, 26 January 2012: 3:45 PM
Evaluation of Ensembles of Opportunity Available During Hazardous Weather Testbed 2011 Spring Experiment
Room 238 (New Orleans Convention Center )
Tara L. Jensen, NCAR/RAL, Boulder, CO; and B. J. Etherton, S. J. Weiss, I. L. Jirak, D. R. Novak, F. E. Barthold, C. J. Melick, F. Kong, M. Xue, M. C. Coniglio, J. S. Kain, M. Harrold, and G. DiMego

The US operational numerical weather prediction (NWP) community is working diligently to satisfy the need for probabilistic forecasts in support of National Center for Environmental Prediction (NCEP) forecast centers. The general plan is to increase compute resources over the next decade to accomodate NWP ensemble goals. In the meantime, forecast centers like the Storm Prediction Center (SPC) and Hydrometeorological Prediction Centers (HPC) are looking for useful alternatives. Several ideas were explored during the Hazardous Weather Testbed - Experimental Forecast Program (HWT-EFP) 2011 Spring Experiment, including a seven member storm-scale ensemble of opportunity and several sub-ensembles formed from the Center for Analysis and Prediction of Storms (CAPS) Storm-Scale Ensemble Forecast system (SSEF). Additionally, NCEP/Environmental Modeling Center (NCEP/EMC) provided a different approach to the ensembling challenge through the Hybrid-Regional Ensemble Forecasts (HREF).

The HWT SSEO was constructed at SPC on N-AWIPS and consisted of seven convection-allowing models available at the SPC and HPC. All had 4-km grid spacing. Five were from the current 00 UTC run and two were time-lagged from 12 hrs earlier. Six of the seven members are run at NCEP/EMC and one was provided by the National Severe Storms Laboratory (NSSL).

The CAPS SSEF was a 4-km fifty model ensemble with nearly half (twenty-one) containing mixed initial condition (IC)/physics perturbations (seventeen ARW and four NMM). In all mixed IC members, the background initial condition will come from interpolation of the 12 km NAM analysis. Mesoscale atmospheric perturbations will be introduced in the initial and lateral-boundary conditions of the mixed IC/physics members by extracting perturbations from EMC‘s operational Short Range Ensemble Forecast (SREF) system and applying them to the 21 members. Convective-scale perturbations will be introduced in the initial conditions of all but two members (one ARW and one ARPS) by assimilating reflectivity and velocity data from the national NEXRAD radar network and a cloud analysis as part of a CAPS 3DVAR system. These twenty-one members were combines with three additional members to product the standard suite of SSEF ensemble products for this year. The other twenty-nine models were used for physics and data assimilation experiments. Additionally, there was a fifteen member SSEF ensemble with the same make-up as the 2010 season and a smaller five member ensemble for comparison.

The HREF is a new addition to the NCEP/EMC experimental production suite. It provides a hybrid approach of explicit prediction combined with statistical downscaling. It applies downscaled SREF perturbations, when differenced from EMC HiResWindow WRF-NMM and WRF-ARW, to those HiResWindow WRF-NMM and WRF-ARW runs. This provided a forty-four member ensemble at a 4-km resolution.

These ensembles of opportunity were evaluated subjectively by forecasters participating in the Spring Experiment and objectively by the Developmental Testbed Center (DTC) Ensemble Testbed (DET). Preliminary results indicate that the HWT SSEO preformed better than the larger CAPS SSEF ensemble in some cases but suffered when a particular member was not available. This would suggest that such an ensemble may provide promise to the operational centers as long as all models are available routinely. The HREF exhibited a slight improvement in skill over the NCEP/EMC SREF, for extreme precipitation events, but not as much skill as the CAPS SSEF ensembles. This could imply that this hybrid approach should be examined more closely. Finally, while the ensemble mean of the smaller CAPS sub-ensembles show slightly more skill due to less amplitude reduction in the averaging process, the 15 and 24 member ensembles scored similarly and much higher than the SREF and other operational baselines for probabilistic scores. These results may suggest 15 members is sufficient for high-resolution ensembles.

While these results cannot be considered statistically significant due to sample size limitations, they do provide some interesting points of discussion. Results from the DET objective evaluation and a discussion of the implications use by operational centers will be presented during this presentation. 201896 modified by on 8-8-2011-->

Supplementary URL: