1. UNCERTAINTIES IN ATMOSPHERIC MODELLING
Dick Derwent
rdscientific, Newbury, United Kingdom
Joint TFEIP/TFMM Workshop
Dublin
22nd October 2007
This work was supported by the United Kingdom DEFRA through contract number AQ 03508

2. WHY DO WE NEED MODELS?
Observational networks tell us what is happening to air quality but not why it is happening.
Models provide a framework to link together information connecting a disparate range of issues:
emissions,
meteorological data,
atmospheric chemistry,
air quality data
Much of the presentation refers to ozone rather than PM.

3. WHY MODEL OZONE ?
To provide a vehicle for exploring ignorance and answering questions.
Where have the elevated ozone levels come from ?
How important is long-range transport of ozone?
Which VOCs are the most important to control ?

4. DAILY SOURCE ATTRIBUTION OF GROUND-LEVEL OZONE AT A RURAL LOCATION AT HARWELL, OXFORDSHIRE DURING 2006
EMEP site GB36R

5. WHY MODEL OZONE ?
To develop a means of prediction for policy-makers.
What would happen of this source of precursor emissions is controlled ?
Which sources are best to control ?
What will happen if nothing is done to control emissions ?

6. IMPACT OF THE CAFÉ THEMATIC STRATEGY ON SUMMERTIME OZONE
Results from the Unified EMEP Eulerian Model (Tarrason et al., 2005)

7. MODELS ARE INHERENTLY SIMPLIFICATIONS OF THE REAL-WORLD
When models are built, they are always simpler than the world they represent. This simplification is achieved by:
Generalisation
Distortion
Deletion
Neglect

8. THE PURSUIT OF COMPLEXITY
Models are always incomplete and efforts to make them more complete can be problematic:
adding new features and processes may introduce more uncertain parameters
complex models may contain more parameters than can be calibrated with the available observations
scientific advances will never make it possible to build the perfect model

9. MAIN AREAS OF UNCERTAINTY
The simplifications inherent in models introduce uncertainties. There are four main areas of uncertainty:
theoretical aspects – not fully understood
empirical aspects – difficult to measure
parametrical aspects – simplified concepts
temporal aspects – not stable in time

10. UNCERTAINTIES - THEORETICAL ASPECTS
How to cope with atmospheric dispersion?
How to cope with the range of spatial scales involved?
Are we dealing with long range transport of ozone or with the formation of ozone on the long range transport scale?

11. UNCERTAINTIES – EMPIRICAL ASPECTS
How to handle the 100s of VOCs, the 1000s of RO2 radicals, the 10,000s of reactive intermediates?
How to measure them, how to describe ozone formation using them?
How to construct a chemical mechanism from smog chamber data?

12. UNCERTAINTIES – PARAMETRICAL ASPECTS
Simplified concepts need parameters to describe them.
flux-based dry deposition schemes need data on vegetation and soil status, precipitation, phenology
natural biogenic emission schemes need data on radiation, temperature, plant species, phenology
solar photolysis rates need data on aerosol, cloud and stratospheric ozone column

13. UNCERTAINTIES – TEMPORAL ASPECTS
Some parameters are not stable in time and present problems when working down from annual values.
some emission processes have a large stochastic element
some emissions are strongly event-based, accidental or just random

14. HOW TO HANDLE UNCERTAINTIES ?
A wide range of possibilities exist for handling uncertainties:
Probabilistic uncertainty analysis
represent all model uncertainties probabilistically
compute distribution of output of interest
Scenario assessment or sensitivity analysis
consider ‘pessimistic’, ‘neutral’ or ‘optimistic’ scenarios for parameters

15. WHAT ARE THE MAJOR UNCERTAINTIES ?
missing processes
weather conditions
chemical mechanism
emissions from human activities
emissions from natural processes

16. UNCERTAINTY ANALYSIS IN A PHOTOCHEMICAL TRAJECTORY MODEL
VOC emissions
NOx emissions
SO2 emissions
CO emissions
methane emissions
isoprene emissions
deposition velocities
initial concentrations
boundary conditions
x,y,z trajectory position
air parcel temperature
air parcel pressure
air parcel humidity
boundary layer depth

17. UK PTM MODEL PERFORMANCE DURING JULY 2006
30 mesoscale trajectories per 15:00z

18. There are difficulties selecting a single air parcel trajectory.
Output from the HYSPLIT Trajectory Model (NOAA ARL)

19. UNCERTAINTY ANALYSIS FOR 18TH JULY 2006
Subjective uncertainty ranges adopted for:
ammonia emissions
VOC emissions
NOx emissions
SO2 emissions
CO emissions
CH4 emissions
isoprene emissions
PAR speciation
XYL speciation
TOL speciation
FORM speciation
ALD2 speciation
OLE speciation
ETH speciation
O3 dry deposition
Other species dry deposition
Initial conditions
Boundary layer depth
J values
Temperatures
Longitude of air parcel
Latitude of air parcel
D trajectories from the mesoscale NWP model

20. MONTE CARLO ANALYSIS FOR 1ST JULY 2007
100,000 model runs
233 ‘acceptable’ model runs in this range
Observations 82 ± 8 ppb

21. WHAT ARE FEATURES OF 233 ACCEPTABLE PARAMETER SETS?
Central value

22. OZONE RESPONSES TO 30% NOx REDUCTION IN THE 233 MODEL RUNS WITH ACCEPTABLE PARAMETER SETS
Mean ± 1sd
Harwell, Oxfordshire 1st July 2006
+2.5 ppb
ozone decrease
47 parameter sets
ozone increase
186 parameter sets

23. 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

24. CONCLUSIONS (largely for ozone and policy applications)
We have to face up to model input data being uncertain
uncertainty propagation is model and output dependent
uncertainties in emission inventories are no longer my main concern (except in gridding and biogenic sources)
process descriptions and parameterisations are a major concern
non-inventoried emissions such as forest fires, agricultural burning and industrial fires are a major cause of ozone and PM episodes