NERAM 2006

1
NERAM 2006
Matching the metric to need modelling exposures
to traffic-related air pollution for policy
support
David Briggs, Kees de Hoogh and John
Gulliver Department of Epidemiology and Public
Health Imperial College London
Vancouver, October 16-18th 2006
2
Some time of life questions

• GIS and exposure modelling
• LUR, focal sum techniques
• Which methods work best how can they be
  compared?
• Time of life - what does it all mean?
• Acute versus chronic
• Long-range versus traffic-related
• Spatial/temporal resolution
• The GEMS study
• Locally-driven versus long-range episodes versus
  normal pollution periods
• Linkage of local and long-range models and air
  pollution data

3
Methods of exposure assessment
Type Methods Examples
Monitor-based Nearest site Thiessen polygon
Average of neighbouring sites City average Buffering
Weighted average IDW Kriging
Indicator-based Source proximity Distance to road
Source intensity Road density Traffic/truck volume
Model-based GIS models Land use regression Co-kriging
Dispersion modelling ADMS-Urban AERMOD
4
ADMS-modelled PM10 concentration London
5
Methods and metrics

• Indicators
• Distance to nearest main road (metres)
• Trafnear traffic flow (vehicles) on nearest
  main road
• HGVnear heavy goods vehicles on nearest main
  road
• Trafdist Trafnear/Distance
• Roads150 road density (length/area) within 150
  metres
• Traf150 vehicle km travelled (flowlength) in
  150 metres
• Models
• LURNO2 NO2 concentration based on land use
  regression model
• ADMSNO2 and ADMSPM NO2 and PM modelled with
  ADMS-Urban
• Monitoring
• Fixed site PM10 and NO2 concentrations - annual
  averages based on hourly

6
Scatterplots Indicators
7
Indicators correlations (bottom left) and in
same quintile (top right)
Distance HGVnear Trafnear Trafdist Roads150 Traf150
Distance 28 22 11 9 9
HGVnear 0.03 48 37 21 29
Trafnear 0.06 0.80 38 22 35
Trafdist -0.45 0.40 0.52 34 46
Roads150 -0.62 0.07 0.06 0.45 81
Traf150 -0.42 0.33 0.43 0.60 0.80
8
Indicators correlations with modelled
traffic-related air pollution
ADMS PM ADMS NO2
Distance -0.47 (-0.70) -0.51 (0.69)
HGVnear 0.34 0.37
Trafnear 0.30 0.32
Trafdist 0.72 0.73
Roads150 0.74 0.75
Traf150 0.71 0.73
Power transformation (D-x)
9
Correlations with mean PM10 concentration
(2001-2004) N71
Distance
Trafnear
HGVnear
Roads150
ADMSPM
Traf150
10
Land use regression
R0.88
R0.61
11
Performance of exposure metrics London
Metric PM10 (N71) PM10 (N14) NO2 (N8)
Distance -0.40 (-0.47) -0.44 (-0.74) -0.68 (-0.62)
HGVnear 0.31 0.25 0.18
Trafdist 0.30 0.61 0.56
Roads150 0.40 0.46 0.70
Traf150 0.37 0.36 0.53
ADMS 0.51 0.81 0.72
LUR N/A 0.88 0.61
Power transformation (D-x)
12
Conclusions so far.

• Indicators only weakly to moderately correlated
• Reasonably strong correlations between some
  indicators Distance (power transformed),
  Trafdist, Roads150 and Trafdist and modelled TRP
• Variable capability to reflect geographic
  variations in PM10 concentration
• HGV counts on nearest road poor predictor
  (despite widespread use)
• Distance (power transformed) moderately
  predictive (R20.2-0.5)
• Dispersion and LUR seem to give best results
  (R20.3-0.6)

BUT is monitored PM the gold standard?
13
(No Transcript)
14
Relationships between rural and urban monitoring
sites (n365 days)
Urban site Rural site Species Constant Slope R2 Ratio (urban/rural)
Rochester PM2.5 2.45 0.97 0.82 1.17
Bloomsbury (urban centre) Rochester PM10 8.46 1.06 0.62 1.25
Harwell PM10 6.26 0.86 0.61 1.37
Kensington (kerbside) Rochester PM10 5.36 0.91 0.63 1.16
Marylebone (urban background) Rochester PM10 20.9 1.06 0.31 2.05
15
Conclusions 1

• Monitored PM dominated by long-range particles
• 100 in urban background
• lt80 in urban centre
• gt50 in kerbside
• Little within-city/regional variation in
  long-range component, but drives temporal
  variation
• Time-series studies therefore valid in assigning
  constant exposure across city
• But mainly detect effects of long-range component

16
Conclusions 2.

• Traffic-related particles represent a small
  add-on
• Accounts for majority of spatial variation
• Modelled by dispersion/LUR models
• But need for more standardisation
• Emissions data are the weak element
• Very localised
• Exposures therefore mainly in streets/transport
  environments
• Short duration high concentration

17
Conclusions 3

• What are implications for health?
• Spatial clustering (e.g. near-road studies)
• Are toxicologies of local and long-range
  components different?
• What should policy focus on?
• Local policy small, local effects
• More emphasis on transport environments
• Is hotspot policy appropriate

18
Thank you
Time for bed..