1. UNCERTAINTY ANALYSIS IN OZONE MODELS
Dick Derwent
rdscientific, Newbury, United Kingdom
Task Force on Measurements and Modelling
Cyprus
May 2010
This work was supported by the UK Defra under contract AQ 0704

2. PROBABILISTIC EVALUATION
How to handle uncertainty within policy analyses and assessments?
Model input data are inherently uncertain but how do these uncertainties propagate into the output?
GLUE approach
Equifinality

3. EQUIFINALITY PROBLEM
Many different parameter sets within a chosen model structure may be acceptable for reproducing observations
It may not be possible to find a single optimal representation in a complex model of a given set of observations
Keith Beven (2009) Environmental Modelling: An Uncertain Future

4. WHAT IS THE UNCERTAINTY IN OZONE MODELS DUE TO THE CHEMICAL MECHANISM ALONE?
Chemical mechanisms contain rate coefficient and product yield data covering two major areas:
the inorganic chemistry of the fast photochemical balance
the organic chemistry of the degradation of VOCs which generates ozone
Few of the thousands of reactions involved in the organic chemistry have been studied in the laboratory and many are estimated using structure-activity relationships.
In contrast, all of the inorganic reactions have been studied in the laboratory.

5. UNCERTAINTIES IN CHEMICAL MECHANISMS
Inorganic chemistry of the fast photochemical balance
Organic chemistry of VOC degradation and O3 production
RCHO O3
CO, H2
,CH4, VOCs hν hν
HO2 + RO2 OH
NO+HO2, O3
,RO2+NO
HNO3
H2O2
ROOH

6. MONTE CARLO ANALYSIS OF CHEMICAL MECHANISM UNCERTAINTIES USING A PHOTOCHEMICAL TRAJECTORY MODEL
Studied PUMA Campaign episode during June 1999
Air mass trajectories from Met Office NAME model
Set up PTM using CBM4 (100 reactions) and MCM CRI v2 (1193 reactions) chemical mechanisms
assume every rate coefficient has an uncertainty range of ± 30 %
perform Monte Carlo analysis using 10,000 model runs

7. UNCERTAINTIES IN OZONE MODELS DUE TO CHEMICAL MECHANISMS
Condensed MCM CRI mechanism
UNCERTAINTY DUE TO VOC OXIDATION ALONE
UNCERTAINTY DUE TO FAST PHOTOCHEMICAL BALANCE ALONE

8. UNCERTAINTIES DUE TO CHEMICAL MECHANISMS
UNCERTAINTY DUE TO FAST PHOTOCHEMICAL BALANCE ALONE
Standard deviation
± 10.4 ppb
No difference because both mechanisms use the same rate coefficient data for the fast photochemical balance

9. UNCERTAINTIES DUE TO CHEMICAL MECHANISMS
Standard deviation
± 4.6 ppb
± 4.4 ppb
Surprisingly little difference here. This is because most of the RO2 are reacting with NO because the system is VOC-limited.

10. MONTE CARLO ANALYSIS FOR UNCERTAINTIES IN THE CHEMICAL MECHANISMS ALONE
network observations
10,000 runs
± 30 % uncertainties in all reactions or inorganic only
50%ile
Standard Deviation
79.0
79.2
80.2
80.3
± 11.1 ppb
± 10.4 ppb
PUMA campaign 26th June 1999

11. MONTE CARLO ANALYSIS FOR UNCERTAINTIES IN THE CHEMICAL MECHANISMS ALONE
Conclusions:
there is little difference between the results from the CBM4 and the CRI v2 mechanisms
the CRI v2 gives a much more explicit treatment of VOC chemistry, an increase from 6 emitted VOCs in CBM4 to 115 in CRI v2
the vast majority of the chemical mechanism uncertainty comes from the inorganic chemistry that drives the fast photochemical balance
there is little increase in uncertainty in going from the compact condensed CBM4 mechanism to the CRI v2 mechanism based on the MCM

12. MONTE CARLO ANALYSIS OF ALL PTM MODEL UNCERTAINTIES
all dry deposition velocities: x 0 – 1
longitude and latitude of position: ± 0.45o x 1-3; ± 0.28o
NH3 VOC NOx SO2 CO CH4 emissions: x ± factor of 2
VOC speciation: x ± factor of 2
isoprene emissions: x ± factor of 4
initial conditions: x ± factor of 1.5
boundary layer depth: x ± factor of 2
relative humidity: x ± factor of 2
temperature: ± 0 – 3 oC
choice of NAME trajectory: select 1 out of 1000 independent choices
all rate coefficients: ± 30 %
These are subjective estimates of uncertainties. PUMA Campaign 1999.

13. MONTE CARLO ANALYSIS TO FIND ACCEPTABLE PARAMETER SETS
Fixed PUMA campaign input for 26th June 1999
Monte Carlo sampled input
PTM model
Generate 10,000 sets of outputs
Fail
Pass
Reject parameter sets
Check against observations
Store parameter sets
2,602 acceptable parameter sets

14. PROBABILISTIC UNCERTAINTY ANALYSIS IN A POLICY CONTEXT
PTM model
PUMA Campaign
26th June 1999
Monte Carlo analysis
PTM model with 2,602 acceptable sets of parameters
-30 % NOx control case
-30 % VOC control case
Base case
Which set of controls are more likely to reduce peak ozone to below 60 ppb?

15. PROBABILISTIC ASSESSMENT OF RESPONSES TO 30 % REDUCTIONS IN NOx AND VOC EMISSIONS
O3 decreases
O3 increases
Percentage of runs below 60 ppb
5.4%
9.9%
Decrease of 24 ppb required to take O3 down to 60 ppb
Area under curves
2602 acceptable parameter sets
PUMA Campaign 26th June 1999

16. PROBABILISTIC SOURCE ATTRIBUTION
2602 acceptable parameter sets
PUMA Campaign 26th June 1999

17. CONCLUSIONS
Probabilistic uncertainty assessments are possible using Monte Carlo sampling
Provides a way of handling equifinality
Approach is practical with simple trajectory model
How are these methods to be applied to the large and complex 3-D Eulerian grid models?