Overview of HFIP FY10 activities and results Bob Gall HFIP Annual - PowerPoint PPT Presentation

Overview of HFIP FY10 activities and results Bob Gall HFIP Annual Review Meeting Miami Nov 9, 2010

Outline • In this presentation I will show a few preliminary results from the summer program. – More detail should come out in the team reports • The second talk later this morning will outline changes to the management of Stream 1.5 • On Wed we will discuss next steps for the HFIP program

Recent HFIP Activities/Results • A 20 member low resolution GFS (T256~60 km) ensemble using an EnKF DA system showed a 20% improvement over the higher resolution operational GFS using GSI DA at the longer lead times • The GFS ensemble appears to be providing good predictions of genesis at lead times of several days. – This statistic needs to be verified – Mike may have some results • A multi model ensemble has been run twice per day on each storm this season (Oper. HWRF, Oper. GFDL, TC-COAMPS, AHW, FSU ARW, and experimental GFDL). – Ensemble mean is bias corrected using retro resulkts from the strream 1.5 runs and the oprational archieve. – In addition there is preliminary results from the Correlation Based Consensus proposed by Krish – Initial assessment is very good – Overall statistics will be available at end of season • Stream 1.5 runs being made available to forecasters in real-time (consists of AHW at 1 km and the experimental GFDL at 7.5 km) • Several experiments are being conducted on advanced data assimilation (using all available aircraft data) and alternate initialization systems by HRD – Some statistics presented in the team reports?

Global statistics, GFS/EnKF vs. ECMWF (ensemble statistics, 5 June to 21 Sep 2010; all basins together) GFS/EnKF competitive despite lower resolution (T254 vs. ECMWF’s T639). GFS/EnKF has less spread than error this year, more similar last year. Is this due to this year’s T254 vs. 6 last year’s T382?

FORECASTED HURRICANE COUNT FORECASTED TROPICAL STORM COUNT

The GFDL Ensemble An Ensemble has been constructed from the GFDL “Operational” model with initial conditions defined as follows: 1. GPA - Unbogussed run 2. GPB - GFD5 with no asymmetries 3. GPC - GFD5 with old environmental filter 4. GPD - Increase storm size (ROCI-based) by 25% 5. GPE - Decrease storm size (ROCI-based) by 25% 6. GPF - All wind radii increased by 25% 7. GPG - All wind radii decreased by 25% 8. GPH - This combines the filter and size criteria of GPC and GPF 9. GPJ - This combines the filter and size criteria of GPC and GPG 10.GPK - For small storms sets the min RMAX to 45 km (in GFD5 it is 25km) 11.GP0 - Control run.--essentially the operational GFDL GFMN - The ensemble mean of the 10 perturbed members. This model was run for most of the 2010 season. Results will be presented later in the meeting

Paula

Paula

Hurricane Ike (2008) Ike (2008) Track Errors Sep 1-14 1200 ARFS GFDL 1000 HWRF 800 Error (nm) COTC 600 AHW1 400 GFD5 200 ENSM CBC 0 12 24 36 48 60 72 84 96 108 120 Forecast Hour Ike (2008) Intensity Errors 45 ARFS 40 GFDL 35 HWRF 30 Error (kt) COTC 25 AHW1 20 15 GFD5 10 ENSM 5 CBC 0 12 24 36 48 60 72 84 96 108 120 Forecast Hour

PSU ARW-EnKF Assimilating Airborne Radar OBS Mean Absolute Error and Ensemble Spread for all 56 cases from 2008 A1PS: PSU 1.5km single forecast initialized with EnKF analyses A4PS: PSU 4.5km single forecast initialized with EnKF analyses P400: ensemble forecast mean of 30 members in 4.5km resolution PSTD: averaged ensemble spread of P400

HFIP Intensity Baseline VT (h) N OFCL PRCL BASE 0 820 1.9 2.2 2.2 12 745 7.2 8.3 7.7 24 667 10.4 11.5 10.1 36 590 12.6 14.2 11.7 48 522 14.6 16.1 13.7 72 415 17.0 17.8 16.0 96 316 17.5 19.3 16.6 120 250 19.0 19.3 17.0

End

PSU ARW-EnKF Assimilating Airborne Radar OBS Mean Error and Ensemble Spread for all 56 cases from 2008 A1PS: PSU 1.5km single forecast initialized with EnKF analyses A4PS: PSU 4.5km single forecast initialized with EnKF analyses P400: ensemble forecast mean of 30 members in 4.5km resolution PSTD: averaged ensemble spread of P400 number on the top: sample number of cases for HWRF and EnKF

For earlier forecasts the ensemble predicted for 1200Z September the following probabilities: – 17/23 36 hour lead time – 15/23 60 hour lead time – 10/23 84 hour lead time

Model Descriptions for Mesoscale Models for ensemble forecasts Models Nesting Vertical Cumulus Microphysic PBL Land Radiation Initial and Initialization Horizontal levels Parameterizati s Surface boundary resolution on conditions (km) HWRF 2 43 Simplified Ferrier GFS Non- GFDL Slab Schwarzkopf GFS Advanced HWRF 27/9 Arakawa Local PBL Model and Fels (1991) vortex Schubert (longwave) / initialization Lacis and that uses GSI HWRF 2 42 Simplified Ferrier GFS Non- GFDL Slab GFS Hansen (1974) 3D-var 13.5/4.5 Arakawa Local PBL Model (shortwave) assimilation of Schubert Doppler radar data to run in development parallel. HWRF-X 2 42 Simplified Ferrier GFS scheme NCEP LSM RRTM GFS HWRF HRD version 9/3 Arakawa (longwave) / of HWRF Schubert Dudhia HWRF-x (shortwave) WRF ARW 2 36 New Kain WSM5 YSU 5-layer RRTM GFS EnKF method (NCAR) 12/4 Fritsch (12 km thermal (longwave) / in a 6-hour AHW1 only) diffusion soil Dudhia cycling mode model (shortwave) COAMPS-TC 3 40 Kain Fritsch Explicit Navy 1.5 Force and Harshvardardet NOGAPS 3D-Var data COTC 45/15/5 microphysics order closure restore slab et al. (1987) assimilation (15/5 km (5 class bulk land surface with synthetic following the scheme) model observations storm) GFDL 3 42 Arakawa Ferrier GFS Non- Slab Model Schwarz-kopf- GFS GFDL GFDL 30/15/7.5 Schubert Local PBL Fels scheme synthetic GFD5 bogus vortex WRF ARW 2 27 Simplified WSM5 YSU 5-layer RRTM GFS (initial GFS AFRS 12/4 Arakawa thermal (longwave) / and boundary Schubert diffusion soil Dudhia condition) model (shortwave)

Correlation based model ensembles Model increment Observed increment forecasts (Lat, Lon, values (Lat, Lon, Int) Int) for each lead for each lead time time Correlation coefficients for Training phase each model for Lat, Lon, Int at  2008 and 2009 storm cases each lead time (Total 164 cases)  The storm to be forecasted Normalize the is taken out (if it is in the coefficients using training period) to calculate available member models for Lat, Lon, Int the correlation coefficients at each lead time Utilize the above coefficients during the forecast phase and construct a new forecast

Paula

Global statistics, GFS/EnKF vs. ECMWF (ensemble statistics, 5 June to 21 Sep 2010; all basins together) GFS/EnKF competitive despite lower resolution (T254 vs. ECMWF’s T639). GFS/EnKF has less spread than error this year, more similar last year. Is this due to this year’s T254 vs. 34 last year’s T382?