Aucun titre de diapositive

1
Uncertainties in measurement and modelling an
overview Laurence Rouïl
2
In-situ Measurement data main sources

• Regulatory observation sites (in compliance with
  the Air quality directives)
• Selected air pollutants and parameters measured
• Obligations related to the choice of the
  observation site and the standards used for the
  measurement devices
• Commitment of the Member States to comply with
  the directives in term of station number,
  location, quality assurance, reporting (EIONET
  network)
• Data reported to the European Environment Agency
  and made routinely available
• Research networks grow up in Europe
• New parameters measured non regulatory
  pollutants, aerosol speciation, size
  distribution, physico-chemical parameters
  vertical profiles (Lidars, Radiosondes), aircraft
  measurements
• Wide range of methods can be tested and compared
• Continuous measurement and/or fields campaigns
  (EUSAAR, EARLINET, GALION, EUCAARI,
  MOZAIC/IAGOS.)
• Data compiled by the project partners and made
  available under certain constraints (publication,
  restrictive use)

3
Uncertainties in measurement

• Data quality objectives (DQOs) specified in
  particular in AQ Directives
• Measurement uncertainty
• Minimum data capture
• Minimum time coverage
• Metrological uncertainty from the measurement
  devices rather well managed for regulatory
  pollutants
• Appropriate standards are developed by
  normalisation Committees (CEN, ISO) according to
  the requirements of the Air quality Directives
  (e.g. measurement uncertainty lower than 30 in
  most of the cases)
• Definition of reference methods and
  inter-laboratory tests
• Definition of common statistical procedures for
  uncertainty estimations
• Metrological uncertainty a field of
  investigation for research networks
• Intercalibration campaigns (see the EUSAAR
  project) EC/OC measurement, optical properties,
  size distribution (SPMS)....

4
Intercalibration experiments (from P. Laj)

• OC/EC (J. P. Putaud, JRC) Round-Robin
  intercomparison and development of artefact free
  sampler
• Intercomparison of identical filters from several
  EUSAAR sites operating operating with similar
  thermo-optical methods
• Need for homogeneizing methods -gt Converging
  towards a EUSAAR method for thermal-optical
  methods and EMEP references
• Size distribution (A. Wiedensohler, IFT)
  intercalibration and improvement of SMPS
• 34 CPCs (12 different models) and 16 SMPS were
  checked and calibrated
• Intercalibration clearly needed. High
  variability in terms of total number and size
• Improvement when using standard retrieval
  procedures

5
Uncertainties in measurement (ii)

• Uncertainties in measurement interpretation
• Which parameters are measured?
• Artefacts in the measurement?
• How to retrieve the expected data (concentration
  level) from the available measurement (AOD for
  instance)?
• Non validated and validated data role of the
  human expertise
• Reporting chains (EMEP, EEA) include data
  flagging to qualify the status and the quality of
  the data
• Time release of validated data must be improved
  in most cases (EMEP)
• Access to Near Real Time (NRT) unvalidated data
  offers new opportunities (monitoring of air
  pollution episodes, air quality forecasting and
  short term analysis, NRT model evaluation....)
  but can increase uncertainties.
• Uncertainty due to the measurement strategy
• Representativeness of the observations to
  reduce uncertainties in maps production and air
  quality assessment
• Performance in terms of data capture and time
  coverage

6
Example sensitivity to the spatial sampling
strategy
Initial data set (source ATMO Champagne-Ardenne,
2005)
Example NO2 background concentrations over the
region Champagne-Ardenne (France) winter 2005
Dx mesh size
7
Spatial sampling strategy
Sensitivity of the estimated map to sampling
density. The sampling mesh should not be larger
than 15 km.
Ordinary kriging Estimated maps
8
Spatial sampling strategy
With auxiliary variables, the sampling mesh can
be extended to 25-30 km..
Kriging of the residuals using population and NOx
emissions density Estimated maps
9
Uncertainties in modelling

• Estimated by comparison with measurement
• Statistical scores (bias, root mean square error,
  gross error, correlation)
• Graphical indicators (Taylors diagrams)
• Contingency tables assessing the ability of the
  model to capture situations where thresholds are
  exceeded or not
• Various sources of uncertainties
• input data emissions and meteorological fields
  (V, temperature, . . .)
• physical parameterizations (ci , K, . . .)
• numerical schemes
• Model resolution
• Sensitivity to input data propagating input
  uncertainty in the models with Monte-Carlo
  approaches

10
Méthodology

• Probability Distribution Function (PDF) for input
  parameters
• PDFs propagates in the CTMs with a Monte Carlo
  approach Hanna et al. 1998, 2001, Beekmann and
  Derognat 2003

Parameter PDF Factor
Wind speed LN 1.5
Temperature N 1
PM emissions LN 4

• Sources
• Parole d'expert
• Erreur de mesure
• Ecart aux observations

PDF concentrations
PDFs parameters
AQ model
Standard deviation measure of the
output concentration uncertainty.
11
Example CHIMERE France results
PM10 winter 2009 300 simulations
Ozone august 2009500 simulations

• Standard deviation 19 for ozone daily peak et
  33 PM10 daily average
• Lower for highest concentrations
• Uncertainty can be underestimated for PM model
  concentrations, the bias being also
  underestimated?

12
Identification of the sensitive variables for
ozone concentrations

• Temperature
• Lateral boundary conditions
• Deposition speed

13
The ensemble approach to assess model uncertainty
14
From the individual model verification....
14
15
to the multi-model analysis range of
variability a kind of model uncertainty
measurement
Biais
RMS
16
Model intercomparison and evaluation exercises
a promising approach to assess model uncertainty
The AQMEII initiative JRC (S. Galmarini),
USEPA (S.T. Rao)
The Eurodelta initiative with JRC,
CONCAWE, Next phase under the TFMM umbrella
17
Emissions, modelling and measurement ..

• Close relationship missing sources (natural) ,
  inaccurate approximation (diffusive emissions,
  wood combustion...) can explain a part of
  uncertainty in model results
• High temporal resolution for emissions can be
  crucial for forecasting or NRT monitoring
  applications
• Observation should help in improving emission
  events new opportunities with earth observation
• Modelling should help in assessing emission
  inventories
• Inverse modelling considering reduced
  uncertainties of observations to constrain models
  and to improve emission inventories
  next operational step?

Impact of high resolution emission inventory
MACC/TNO) on NO2 daily peak simulated by CHIMERE
(RMS)