1. AHW Ensemble Data Assimilation and Forecasting System Ryan D. Torn, Univ. Albany, SUNY Chris Davis, Wei Wang, Steven Cavallo, Chris Snyder, James Done, NCAR/MMM

2. Overview Since participation in HFIP HRH test, we have been using cycling EnKF approach to create initial conditions for AHW model Wanted initial conditions that: –Have a good estimate of environment –Have a “decent” estimate of TC structure –Does not lead to significant initialization problem Since then, we have upgraded the system based on observed flaws in both model and initial conditions

3. Assimilation System WRF ARW (v3.3), 36 km horizontal resolution over basin, 96 ensemble members, DART assimilation system. Observations assimilated each six hours from surface and marine stations (P sfc ), rawinsondes, dropsondes > 100 km from TC, ACARS, sat. winds, TC position, MSLP, GPS RO Initialized system this year on 29 July, continuous cycling using GFS LBC No vortex bogusing or repositioning, all updates to TC due to observations

4. AHW Model Setup WSM6 Microphysics (includes graupel) Tiedtke cumulus parameterization (includes shallow convection) YSU PBL, NOAH land surface model Updated Ck/Cd formulation in Davis et al. 2010 Pollard 1D Column ocean model SSTs from NCEP 1/12 degree analysis HYCOM Mixed-layer depths

5. Data Assimilation Nesting Strategy Each time NHC declares an INVEST area, generate a 12 km resolution two-way interactive nest that moves with the system until NHC stops tracking it (1600 km x 1600 km nest) Observations are assimilated on the nested domain each 6 h Nest movement determined by extrapolating NHC positions over the previous 6 h Works better than vortex-following nests, which have largest covariances associated with differences in land position

6. Nest Example Earl Fiona Gaston INVEST

7. Cycling Nest Experiments No Nest (HRH test setup) Fixed 12 km two-way nest (1000 km x 1000 km) generated for each TC in domain at each analysis time. Discarded at end of model advance (2009 real-time setup) Moving 12 km two-way nest (1000 km x 1000 km) generated when TD is declared. Nest is cycled along with the coarse domain. Motion based on NHC advisory position over previous 6 hr (2010 real-time setup)

8. TC Vitals Error RMS Error Bias

9. Deterministic Forecast For each TC, choose one analysis ensemble member whose TC properties are closest to ensemble mean (see below) Remove other 12 km nests, add additional 4 km nest to 12 km for that storm (800 km on a side), runs without cumulus parameterization.

10. 2011 Retrospective Forecasts TrackMaximum Wind Speed

11. Ensemble Forecasts Currently running 15 member ensemble for NHC highest priority TC Take first 15 members of the analysis ensemble since all are equally likely All members use same lower BC, lateral BC, and model physics (will be relaxed in the future)

12. 0000 UTC 2 Aug. Ensemble

14. Lessons to Date Ensemble can produce fairly large variance in intensity without significant variance in large-scale environment parameters Biases due to deficiencies in model physics (in particular aerosol and ozone treatment) lead to many situations where truth is outside ensemble