Assimilation of GPS Radio Occultation Data for an Intense Atmospheric River with the NCEP GSI System Zaizhong Ma and Ying-Hwa Kuo MMM/NCAR Marty Ralph, Ellen Sukovich, and Paul Neiman NOAA/ESRL/PSD
Atmospheric River case: Nov 6-8, 2006 From Neiman et al. (2008)
Observed Daily Precipitation 24-h precipitation ending at 1200 UTC 7 November 2006 Flooding and debris flow on White River, Oregon
Experiment Setup CTRL: operational observation data; LOC: CTRL+GPS with Local operator NON: CTRL+GPS with Non-local operator Three runs: 18 UTC 0200 UTC 03 First-guess is AVN analysis Cycling experiments with 6-hr assimilation window from 3 to 9 Nov CTCL LOC NON 12 UTC 0918 UTC 09 First-guess is 6h WRF forecast …………… First-guess is 6h WRF forecast System: NCEP Gridpoint Statistical Interpolation (GSI) + WRF ARW Case: AR took place in the early of Nov.2006 Setup: Cycling Assimilation: 36km38L; Ptop: 50hPa 24h Forecast: triple nested domain, 36x12x4km
GPS RO soundings for one week (Nov. 3-9, 2006) The distribution of GPS RO soundings with the time in each 3h cycling assimilation window.
PWV of analysis at 0600 UTC 07 Nov SSM/I observation Non-Local analysisLocal minus Non-Local
The 3-h WRF forecasts fit to GPS refractivity with time. The value is cost function for CTRL (blue), LOC (red) and NON-LOC (green) runs, respectively. 3h Forecast Verification in the Cycling Assimilation
The statistics of difference for the assimilation domain from 0000 UTC 03 to 1800 UTC 09 November Bias (left panel) and Standard Deviation (middle panel) errors of 3-h WRF forecasts verified against GPS RO refractivity for CTRL (dashed curve), LOC (thin curve) and NON-LOC (thick curve). The right panel shows the total number of verifying GPS soundings at each level during one-week cycling period. Standard Deviation and Bias of 3h forecast fit to GPS Refractivity
GPS Impact on 24h WRF forecast D1 D2 D3 24h forecast starting from 1200 UTC 6, 3 domains nested. Assimilation on D1. D3 only covers Washington and Oregon states.
24-hr accumulative precipitation ending at 1200 UTC 7 Nov OBS LOC CTRL NON-LOC
24h PWV Difference between LOC (or NON) and CTRL experiments LOC - CTRL NONLOC - CTRL
Bias and Standard Deviation of 24h forecast fit to GPS Refractivity
QPF and evaluation data SITES 50 sites in WA, OR, & CA (117” precip. total) 22 sites in “wet” region (107” precip. total) 28 sites in “dry” region (10” precip. total) WA OR CA DATA 1200 UTC 6 Nov. to 1200 UTC 7 Nov Model quantitative precipitation forecast (QPF) –Forecasts made from 12 Z to 12 Z –Resolution of 4 km Quantitative precipitation estimates (QPE) –From NWRFC –Gauge-based –12 Z to 12 Z –Resolution of 4 km Verification Region
All 50 sites (wet area and dry area) 24 h COSMIC QPF (in)NWRFC (in) CTRLLOCALNONLOCALObserved Avg Precipitation Avg Bias h COSMIC QPF (in) NWRFC (in) Site IDCTRLLOCALNONLOCALObserved Astoria, ORAST Frances, WAFRAW Cinebar, WACINW Cougar, WACUGW Packwood, WAOHAW Aberdeen, WAABEW Enumclaw, WAENUW Glacier, WAGLAW Leavenworth, WALWNW Marblemount, WAMARW Seattle, WASEA Skykomish, WASKYW Stampede Pass, WASMP Quillayute, WAUIL Verlot, WAVERW Bonneville Dam, ORBONO Detroit Dam, ORDETO Lees Camp, ORLEEO Portland, ORPDX Three Lynx, ORTLYO Salem, ORSLE Summit, ORSMIO Avg ppt Avg Bias h COSMIC QPF (in) NWRFC (in) Site IDCTRLLOCALNONLOCALObserved Brookings, OR4BK Burns Airport, ORBNO Cougar Dam, ORCGRO Colville, WACQV Crater Lake, ORCRLO The Dalles, ORDLS Eugene, OREUG Spokane, WAGEG Agness, ORILHO Klamath Falls, ORLMT Meacham, ORMEH Rogue Valley, ORMFR Mazama, WAMZAW Enterprise, ORNTPO Oak Knoll, CAOKNC Omak Airport, WAOMK North Bend, OROTH Owyhee, NVOWYN Rome, ORP Pendleton, ORPDT Prairie City, ORPRCO Riddle, ORRDLO Redmond Roberts, ORRDM Glide, ORSRSO Goldendale, WASSPW Sexton Summit, ORSXT Williams, ORWLMO Yakima, WAYKM Avg Avg Bias Site Forecast and Observed Data “Wet” region sites“Dry” region sites
> 7 in/24h in/24h in/24h in/24h Indicates “wet” region in/24h Observed precipitation (inchesX100) Observed precipitation at 50 evaluation sites
Comparison of QPF bias for forecasts with (“non- local”) and without (“control”) COSMIC data Control is best Minor difference Nonlocal is best Indicates “wet” region [QPF (non-local) – QPF (control)]/observed X100% * Numerical values represent difference between the two forecasts in inches, normalized by the total observed precipitation at that site. It is expressed as a percentage. *Color fill represents which forecast had smallest bias: -green: COSMIC data improved the forecast -red: Control run without COSMIC is still best -yellow: Differences were minor ***The COSMIC data improved the QPF at sites where the heaviest rain fell. NOLOCAL performs better than LOCAL.
Summary and Conclusions COSMIC GPS RO soundings successfully assimilated with NCEP regional GSI system using both local and nonlocal observation operators. Assimilation of COSMIC data improved regional analysis and prediction of the atmospheric river event: –Better fit to independent observations Nonlocal observation operator performs better than local observation operator: –Significantly reduces dry bias in precipitation forecast –improves QPF at sites where the heaviest rain fell More case studies are needed to substantiate the results.