Date post: | 11-May-2015 |
Category: |
Technology |
Upload: | guestee5a52 |
View: | 615 times |
Download: | 1 times |
KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universität Karlsruhe (TH) www.kit.edu
Approaches to visualisation of uncertainties to decision makers in an operational Decision Support System
W. Raskob1, F. Gering2, V. Bertsch3
1 Forschungszentrum Karlsruhe, IKET, Karlsruhe, Germany2 Federal Office for Radiation Protection, Neuherberg, Germany 3 Karlsruhe Institute of Technology , Karlsruhe, Germany
ISCRAM 2009, 10.-13-05.2009
ISCRAM 2009; Raskob; 2
Outline
Introduction
Short description of the decision support system RODOS (Real-time On-line Decision SuppOrt system)
Early phase issues
Late phase issues
Conclusions
ISCRAM 2009; Raskob; 3
ISCRAM 2009; Raskob; 4
Exposure during and after a nuclear accident
Ingested with water
Dos
e ra
te
Time
Inhaled from plume
External from plume
External from deposition
Ingested with food
Accident happened
Hours Days Weeks Months Years
Total
R. Mustonen
ISCRAM 2009; Raskob; 5
Information processing in RODOS
Meteorological and Release Data, Radiological Monitoring Data
GIS Data, National Data Base, Scenario Data0
Environmental Contamination of Air, Ground, and Food, Potential Doses
Radiological Situation: real-time Diagnosis +
Prognosis
1
Areas, Organ Doses, People affected by Countermeasures, Health Effects, Effort and Cost
Countermeasures: Strategies and Consequences
2
Ranked List of feasible Strategies of long-term countermeasures (Decision Analysis)
Evaluation of Strategies3
ISCRAM 2009; Raskob; 6
Data Uncertainties Parameter (preferential) Uncertainties
t
Early Phase: Uncertainty of the Input Data
- Meteorological Fields
- Source term
Late Phase: Uncertainties of CSY-Simulations and Uncertainties of decision parameters
- Weights
- Value functions
Emergency
Intermediate Phase: Measurement Uncertainties
Different types of uncertainty are of different importance in the different phases of emergency management
ISCRAM 2009; Raskob; 7
Issues in the early (pre-release) phase
ProblemSource term very uncertain
Results from dose assessments are uncertain due to the very uncertain source term and uncertainties in the weather forecast (besides limitation of the dispersion model and the conversion of activity to dose)
How to deal with itIn plant data used to estimate source term on best information available (ASTRID, STERPS)
Improve weather forecast and simulation models
Decisions have to be taken with very uncertain input to initiate evacuation, sheltering or distribution of stable iodine
ISCRAM 2009; Raskob; 8
Typical result of dose model
Dose for action: “sheltering” is 10 mSv
ISCRAM 2009; Raskob; 9
Comparison of on-site and prognostic weather data
Preliminary results for a NPP in hilly terrain in Germany
Statistics of differences between numerical weather forecast and Neckarwestheim data for the first 11 hours of a 48 hour prognosis
Limited set of data (less than 3 months)
0 90 180 270 360
M easured w ind d irection [deg]
0
90
180
270
360
NW
P w
ind
dir
ect
ion
[de
g],
0 t
o 1
1 h
ou
rs f
ore
cast
W ind speed a t 40 m > 3 m /s
W ind speed a t 40 m < 3 m /s
N eckarw estheim , 58 m .
ISCRAM 2009; Raskob; 10
Model parameters
Model input
Uncertainty of model parameters
Uncertainty of model input
Ensembles
model parameters
Uncertainty modelling
Ensemble-Kalman filter used to generate 100 Ensembles
Distribution of uncertain model parameters is derived a priori
ISCRAM 2009; Raskob; 11
Ensemble calculations
Main source of uncertainty for atmospheric dispersion modelling is the input data (two key variables):
Source term: log-normal distribution is assigned to the source term since a deviation of an order of magnitude is considered to be equiprobable in both directions
Wind direction: normal distribution is assigned to the mean wind direction with a standard deviation of 30°
ISCRAM 2009; Raskob; 12
Communication of results in RODOS
Two types of results considered in German RODOS
Decision relevant: colour coding is:Green: no problem
Yellow: be careful
Reddish: level is exceeded
Not decision relevant: colour code is a variety of blue
ProblemColour-blindness (red-green)
Printing
ISCRAM 2009; Raskob; 13
Visualisation of uncertainties (2D)
Two layers, one showing the mean value and the second the standard deviation (from http://www.cse.ohio-
state.edu/~bordoloi/Pubs/pdfCluster.pdf)
Weather forecast: movement of storm with trajectory and area of potential deviation from the mean trajectory (from NOAA)
ISCRAM 2009; Raskob; 14
Proposed visualisation
Proposed visualisation of the impact of data uncertainties
The area and location of the probability to exceed the dose threshold for sheltering is displayed
Decision makers have to decide which area is appropriate?
ISCRAM 2009; Raskob; 15
Issues in the later phase
Problem
Many possible countermeasures might be applicable to reduce the dose or consequences
Non quantifiable factors influence the decision
How to deal with it
Measurements and countermeasure simulations by DSS provide basis for a decision
Decisions analysing support tools provide means to deal with non quantifiable factors such as social or political aspects
Decisions have to be taken with relative certain input but other ‚soft‘ factors have to be taken into account
ISCRAM 2009; Raskob; 16
Resolving conflicting objectives, setting priorities and building consensus for the various perspectives of the many stakeholder groups
One has to ensure transparency during the decision making process
First, the problem has to be structured and analysed and second the preferences and importance of the influencing factors have to be determined
This task can be performed either as iterative process or with the help of tools (e.g. Multi Criteria Decision Analysis with Multi-Attribute Value Theory)
Decision making in the context of emergency management
ISCRAM 2009; Raskob; 17
Problem Structuring aims at hierarchically modelling the decision criteria
overall goal
overall objective
logistics
dose
sub-objectives
collective dose saved
individual dose saved
waste
work effort
attributes
strategy y
strategy x
strategy z
alternatives
ISCRAM 2009; Raskob; 18
The direct weighting dialog
The “SMART” weighting method
The “SWING” weighting dialog
Web-HIPRE provides various preference elicitation methods
Preference Elicitation
ISCRAM 2009; Raskob; 19
The composite priorities illustrate the results of the analysis and the contributions of the
different criteria to the overall results
Aggregation
The communication of the results is accompanied by sensitivity analyses in Web-HIPRE
Sensitivity Analysis
Sensitivity analyses show the effect of changing the weight of an objective and give
an overall assessment of the decision parameters
ISCRAM 2009; Raskob; 20
Example data for Web-Hipre
Example ensembles
Deviation from mean wind
direction
Deviation from mean source
term
1 0° x 1.0
2 0° x 0.01
3 + 30° x 1.0
4 0° x 100
5 - 29° x 0.02
6 - 40° x 0.007
7 + 6° x 5.1
8 - 24° x 0.9
9 + 48° x 488
10 - 4° x 1.2
Distribution normal log-normal
100 ensembles from the atmospheric dispersion calculations were used to assess the potential countermeasures/consequences
No individual uncertainty analysis is performed in the countermeasure subsystem
Preferences
ISCRAM 2009; Raskob; 21
Composite priorities
Visualisation of three results for the composite priorities (5%, mean and 95% percentiles)
Important: does the “best” option change for a given percentile
ISCRAM 2009; Raskob; 23
Conclusions and future steps
Uncertainties are part of any decision making in emergency situations
Uncertainties are not much considered in Decision Support Systems for nuclear and radiological emergencies
Time consuming calculations
Decision makers prefer deterministic results (German experience)
Ensemble method provides a good basis for determining uncertainty bands
Visualisation in terms of probability bands is one possible outcome of such an uncertainty handling in the early phase
Visualisation of distinct percentiles might be a good solution for the later phase
Visualisation will be tested in future work shops and the RODOS Users Group (RUG)
ISCRAM 2009; Raskob; 24
Thank you for your attention
Questions?
http://www.euranos.fzk.de