Title: Use of Ensembles to Assess Uncertainties in Consequence Assessment
1Use of Ensembles to Assess Uncertainties in
Consequence Assessment
- Robert P. Addis and Robert L. Buckley
- Atmospheric Technologies Group
- National Homeland Security Directorate
- Savannah River National Laboratory
- March 15, 2006
- Palmetto Chapter of the American Meteorological
Society
- Mini-Technical Meeting, Columbia, South Carolina
2Plan
- European ENSEMBLE project brief description
- US participation in ENSEMBLE
- Describe the SRNL operational use of RAMS LPDM
- Atmospheric modeling
- Why is it so uncertain?
- Importance of ensembles in atmospheric modeling
- Understanding model uncertainties for emergency
response
- A role for ensembles
3ENSEMBLE
- European Union effort to improve modeling
capabilities in event of hazardous accidental
releases
- Follow-up to ATMES (1992), ETEX (1994), and RTMOD
(1998)
- Perform ensemble averaging of transport model
output
- Numerous countries participating (mostly
European)
- SRNL is United States participant
- Variety of outputs required on pre-selected grid
(equally spaced latitude/longitude)
4European ENSEMBLE Project
- 17 nations (mostly European)
- 19 institutes
- using 29 models
- Currently completed 17 planned exercises and
several special exercises
- New ENSEMBLE initiative
- to expand the domain to anywhere in the world
- to allow for incident specific grid resolution
5ENSEMBLE Exercises
- Ex14-17 Cernavoda, Romania ConvEx-3 May, 2005
- Ex 13 Malmo, Sweden November, 2004
- Ex 12 Murmansk, Russia
- Ex 10 11 London, UK June, 2003
- Ex 9 Bratislava, Slovak Republic February,
2003
- Ex 8 Mochovce, Slovak Republic December, 2002
- Ex 7 Glasgow, Scotland, UK October, /2002
- Ex 6 Dublin, Ireland June, 2002
- Ex 5 Stockholm, Sweden April, 2002
- Ex 4 Nantes, France February, 2002
- Ex 3 London, England, UK November, 2001
- Ex 2 Carcassonne, France October, 2001
- Ex 1 Lerwick, Scotland, UK April, 2001
6 U.S. Methodology
- Meteorological (analyzed forecast) fields
- SRNL uses Regional Atmospheric Modeling System
(RAMS) to generate meteorological fields
- Initialize bound with larger scale data (i.e.
NOAA)
- Dispersion
- SRNL uses Lagrangian Particle Dispersion Model
(LPDM) to perform transport using RAMS winds and
turbulence
- Modifications made to accommodate wet/dry
deposition in LPDM
- Scripts to automate RAMS portion.
- Input for LPDM decided the day of the exercise
7ENSEMBLE and SRNL Modeling Domains
- Curved nature of the SRNL grid is due to a
polar-stereographic projection
- SRNL Large grid
- covers most of the ENSEMBLE domain
- Resolution 60 km
- SRNL Nested grid
- covers vicinity of the release
- Resolution 15 km
8The problem with atmospheric models..
- Atmosphere is governed by non-linear partial
differential equations
- The solutions are not unique
- In models, uncertainties in initial and boundary
conditions create uncertainties in the analyzed
and forecast fields (Chaos Theory)
- Incomplete knowledge of physical processes, how
to parameterize them and how to describe scale
interactions also leads to uncertainties in model
results.
9The problem with atmospheric models..
- Edward Lorenz was running a model of convective
atmospheric rolls. He wanted to rerun it from the
middle
- He jotted down the intermediate parametric value
to 3 significant figures 0.506, and reran it.
The solution diverged.
- The computer stored the intermediate parameter
on the first run to 6 significant figures
0.506127
Initial difference 0.000127 or 0.025
10Importance of initial conditions
Forecast state
Initial state
Initial uncertainty of an atmospheric variable
T, P, u, etc. Described by a probability d
ensity function
Produces an ensemble of predicted states
Reference Roberto Buizza ECMWF
11The problem with atmospheric models..
- Atmosphere is governed by non-linear partial
differential equations
- The solutions are not unique
- In models, uncertainties in initial a and
boundary conditions create uncertainties in the
analyzed and forecast fields (Chaos Theory)
- Incomplete knowledge of physical processes, how
to parameterize them and how to describe scale
interactions also leads to uncertainties in model
results.
12(No Transcript)
13Ensembles part of the solution
Reference Roberto Buizza ECMWF
Ensembles help us understand uncertainties
14Ensemble Exercise 10
- Release from London
- June 11, 2003 noon
- Duration 15 minutes
- Rate 1E13 Bq/hr
- Cs-137
- 350 m above ground
15Decisions?
- You
- are the Prime Minister of the UK
- the Prime Minister of the Netherlands calls you
- What do you tell him?
- What did the President of France tell him?
- What did his own experts tell him?
16UK Met Office Model Run
The UK Met Office uses this information to advise
PM Blair. Mr. Blair tells PM Balkenende The
plume is well clear of the Netherlands. You have
nothing to worry about.
Amsterdam
17Decisions?
- You
- are the Prime Minister of the UK
- the Prime Minister of the Netherlands calls you
- What do you tell him?
- What did the President of France tell him?
- What did his own experts tell him?
18Meteo France Model Run
Meteo France uses this information to advise Pres
ident Chirac. President Chirac tells PM Balkene
nde The plume is definitely inundating the
Netherlands. You must take protective actions.
Amsterdam
19Decisions?
- You
- are the Prime Minister of the UK
- the Prime Minister of the Netherlands calls you
- What do you tell him?
- What did the President of France tell him?
- What did his own experts tell him?
20KNMI (Netherlands) Model Run
The Netherlands Met Office (KNMI)
uses this information to advise PM Balkenende.
What does Mr. Balkenende think about the advice
of Mr. Blair or Mr. Chirac?
Amsterdam
21Three reputable assessments..
- UK Met Office
- Meteo France
- KNMI (Royal Netherlands Met Inst)
- Three of the most respected national weather
centers in the world. Their modeling is excellent.
22(No Transcript)
23Lets look at another example
- Discrepancies between models is more the norm,
than the exception.
- Consider this example and ask what the Italian
Prime Minister Berlusconi would do depending on
which model output he was given.
- Would he respond differently if his advisers were
aware of all 4 outputs?
24(No Transcript)
25(No Transcript)
26Summary
- Nonlinear nature of the equations that govern the
atmosphere,
- combined with
- imperfect knowledge of initial boundary
conditions
- Results in uncertainties variability in
numerical atmospheric models
- Ensembles provide better insight into
- atmospheric behavior
- model performance and variability
- Ensembles provide decision makers with
- a measure of model uncertainty