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

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.

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 ?

DAILY SOURCE ATTRIBUTION OF GROUND-LEVEL OZONE AT A RURAL LOCATION AT HARWELL, OXFORDSHIRE DURING 2006

0

10

20

30

40

50

60

01/0

1/20

06

01/0

2/20

06

01/0

3/20

06

01/0

4/20

06

01/0

5/20

06

01/0

6/20

06

01/0

7/20

06

01/0

8/20

06

01/0

9/20

06

01/1

0/20

06

01/1

1/20

06

01/1

2/20

06

O3,

pp

b

Europe-regional

North America

Asia

Europe-intercontinental

Extra-continental

Stratosphere

EMEP site GB36R

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 ?

IMPACT OF THE CAFÉ THEMATIC STRATEGY ON SUMMERTIME OZONE

Results from the Unified EMEP Eulerian Model (Tarrason et al., 2005)

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

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

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

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?

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?

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

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

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

WHAT ARE THE MAJOR UNCERTAINTIES ?

• missing processes

• weather conditions

• chemical mechanism

• emissions from human activities

• emissions from natural processes

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

UK PTM MODEL PERFORMANCE DURING JULY 2006

0

20

40

60

80

100

120

01/0

7/2

006

03/0

7/2

006

05/0

7/2

006

07/0

7/2

006

09/0

7/2

006

11/0

7/2

006

13/0

7/2

006

15/0

7/2

006

17/0

7/2

006

19/0

7/2

006

21/0

7/2

006

23/0

7/2

006

25/0

7/2

006

27/0

7/2

006

29/0

7/2

006

31/0

7/2

006

Ozo

ne,

pp

b

Obs

84%-ile

97%-ile

30 mesoscale trajectories per 15:00z

Output from the HYSPLIT Trajectory Model (NOAA ARL)

There are difficulties selecting a single air parcel trajectory.

UNCERTAINTY ANALYSIS FOR 18TH JULY 2006

Subjective uncertainty ranges adopted for:

ammonia emissionsVOC emissionsNOx emissionsSO2 emissionsCO emissionsCH4 emissionsisoprene emissionsPAR speciationXYL speciationTOL speciationFORM speciationALD2 speciationOLE speciationETH speciationO3 dry depositionOther species dry depositionInitial conditionsBoundary layer depthJ valuesTemperaturesLongitude of air parcelLatitude of air parcel

+ 1000 3-D trajectories from the mesoscale NWP model

MONTE CARLO ANALYSIS FOR 1ST JULY 2007

100,000 model runs

Observations 82 ± 8 ppb

233 ‘acceptable’ model runs in this range

WHAT ARE FEATURES OF 233 ACCEPTABLE PARAMETER SETS?

0 0.5 1 1.5 2

PAR fraction

XYL fraction

TOL fraction

FORM fraction

ALD2 fraction

OLE fraction

ETH fraction

other species deposition

NH3 emissions

VOC emissions

NOx emissions

SO2 emissions

CO emissions

CH4 emissions

isoprene emissions

O3 deposition velocity

initial concentrations

boundary layer depth

J-values

temperature

Multiplier

Central value

OZONE RESPONSES TO 30% NOx REDUCTION IN THE 233 MODEL RUNS WITH ACCEPTABLE PARAMETER SETS

0

5

10

15

20

25

30

35

40

45

50

-20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20

Delta O3, ppb

Nu

mb

er o

f m

od

el r

un

s

ozone decrease

47 parameter sets

ozone increase

186 parameter sets

Harwell, Oxfordshire 1st July 2006

Mean ± 1sd

+2.5 ppb

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

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