United States Coast Guard 1985 Evaluation of a Multi-Model Storm Surge Ensemble for the New York...
-
date post
21-Dec-2015 -
Category
Documents
-
view
217 -
download
2
Transcript of United States Coast Guard 1985 Evaluation of a Multi-Model Storm Surge Ensemble for the New York...
United States Coast Guard 1985
Evaluation of a Multi-Model Storm Surge Ensemble for the New York Metropolitan
Region
Brian A. Colle
Tom Di Liberto
Stony Brook University
ADCIRC Water-level and Flooding12-km MM5 Forecast 1200 UTC 11 December 1992
meters
Colle, B. A., F. Buonaiuto, M. J. Bowman, R. E. Wilson, et al., 2008: Simulations of past cyclone events to explore New York City’s vulnerability to coastal flooding and storm surge model capabilities , Bull. Amer. Meteor. Soc.
ADCIRC Surge Forecast
Current Real-Time Systems
Stony Brook Storm Surge Model
Stevens Institute of Technology’s Storm Surge model (NYHOPS)
NOAA Extratropical Storm Surge model
http://hudson.dl.stevens-tech.edu/maritimeforecast/
http://www.nws.noaa.gov/mdl/etsurge
http://stormy.msrc.sunysb.edu/
Real-Time Modeling Systems• All three models use different ocean models and
atmospheric forcing. Storm Surge Forecasting Systems
Institution Atmospheric Forcing Ocean Model Start TimeStony Brook 5 MM5 / 3 WRF members ADCIRC 0000 UTC
Stevens Institute of Technology NCEP - NAM model Princeton Ocean Model 12:00 AM
NOAA NCEP - GFS model NOAA Extratropical Storm Surge model 0000 UTC
• Stony Brook Storm Surge Model (SBSS) uses 5 MM5 and 3 WRF members
Stony Brook Storm Surge Model Atmospheric Ensemble MembersMembers Model Microphysics PBL Scheme Radiation Initial Condition Cumulus 9a MM5 Simple Ice MRF Cloud Radiation WRF-NMM GrellBMMY MM5 Simple Ice MY CCM2 GFS Betts Miller GRBLK MM5 Simple Ice Blackadar CCM2 NOGAPS GrellK2MRF MM5 Reisner MRF Cloud Radiation GFS Kain FritschK2MY MM5 Simple Ice MY CCM2 Canadian Model Kain Fritsch221 WRF Ferrier YSU RRTM WRF-NMM Kain FritschGFS WRF Ferrier YSU RRTM GFS model GrellNOG WRF WSM3 YSU RRTM NOGAPS Betts Miller
Real-time Surge Model Grids
Blumberg et al. 1999
SIT Grid
SBSS Grid
NOAA ET Grid
Motivation for Storm Surge Ensembles
MM5 (GRMRF)-NAM WRF(GRYSU)-GFS
0000 UTC April 16th, 2007 – SLP (contour), Temp (shaded) and wind
WRF-GFS
MM5-NAM
OBS
Data and Methods• Data: Nov. 2007 – March 2008 and Oct. 2008 – Dec. 2008 (75
in total)• Deterministic: Mean Error, Root Mean Square Error • Probabilistic: Rank (Talagrand) Histograms, Brier Score, Brier
Skill Score and Reliability Diagrams
• Bias correction was applied after the first month (Nov. 2007).– Use a regression approach of
storm surge observations (> 0 and < 0 m) versus the storm surge mean error.
• Use daily NCEP-NCAR reanalysis to look at the composite flow patterns associated with some of the errors.
Surge Mean Errors
Bias Corr-ALL
NOAA-ET
ALLBias Corr-ALL
NOAA-ET
SBSS
SIT
SIT
Surface Wind Speed Biases
NCEP-NAM
Top 10 Largest Negative Error Days
- 24
• Days determined from calculating the largest 24-48 h negative mean error from the SBSS ensemble member 9a
• Largest negative error day is 11/04/2007 when an extratropical hurricane Noel impacted the region
- 48 h - 24h
Northeast winds occur 24 hours prior to large negative error
Trough moves east/deepens 48 hours prior to large negative errors
0 h
Potential Wave Impacts
Daily Averaged Significant Wave Height, m
Da
ily M
ea
n E
rro
r, m
Significant Wave Height at buoy 44017 vs. 24-48 h Mean Error at Montauk
Top 10 Largest Positive Error Days
• Days determined from calculating the largest 24-48 h positive mean error from the SBSS ensemble member 9a
• Largest negative error period was 12/23 – 12/26/2008
- 24h- 48 h
Pressure gradient strengthens 24 hours prior to large negative error
0 h
RMSE vs Forecast Hour
Bias Corr -ALL
NOAA-ET
SBSS
ALL
SIT
Percentage Best and WorstBEST WORST
NOAA-ET
NOAA-ET
SIT
SIT
SBSS SBSS
Rank Histogram
ALL
ALL-BC
Brier Scores vs Threshold
ALL-BC
ALL
ENS3-BC
ENS3
SBSS
Brier Skill Score (vs SBU CTL)
ALL
ENS3
ENS3-BCALL-BC
SBSS
Reliability Diagrams
SBSSALL
> 0.3 m Surge
> 0.4 m Surge
SBSS
ALL
ALL
SBSS
Reliability Diagram
SBSSALLENS-3
> 0.3 m Surge
> 0.4 m Surge
ALL
ALL
SBSS
SBSS
ENS3
ENS3
Conclusions• All surge models have a slight negative bias overall, which is largest
in the NOAA-ET model. One can not use the last 7-14 days for bias correction, since bias depends on the sign of the surge.
• Stevens Institute (SIT) has greater deterministic accuracy (lower RMSEs) than all the SBSS MM5 and WRF ADCIRC members, which highlights the importance of adding different (and good) ocean models to the ensemble.
• The largest SBSS mean surge errors are dependent on the synoptic flow patterns. Positive surges with nor-easters are underpredicted on average, while offshore flow with an anticyclone to the west favors positive errors (underpredicted “blow-out” conditions).
• Most of the ensemble probabilistic skill and reliability originates from the three different ocean models on average, not from using one ocean model and multi-model atmospheric forcing (MM5 and WRF).
• Recommendation: In addition to coupling NOAA-ET to the SREF (NWS plans to do this soon), added skill can be obtained by also using different ocean (surge) models (ADCIRC, POM, ROMS, FVCOM, etc…).