Talk title: The PCM Model and Modelling Uncertainty
Routes to Clean Air is a two-day conference from the IAQM where academics, professionals and policy makers share their experiences of improving traffic emissions.
This event highlights the importance of public communication and behavioural change surrounding road transport and air quality issues.
Presentation: Farmer-led climate adaptation - Project launch and overview by ...
Routes to Clean Air 2016 - Dr Christine McHugh & Marilena Karyampa, Arup
1. The PCM Model and Modelling
Uncertainty
Dr Christine McHugh
Marilena Karyampa
12 October 2016
2. 2
• What is the PCM model?
• Background concentrations
• Model trends and uncertainty
• Performance at roadside sites in London
• Quality of previous forecasts
• Conclusions
Contents
3. 3
• PCM: Pollution Climate Mapping
• Collection of models used by Defra to report on UK’s
compliance with EU Directive 2008/50/EC
• Base year model and projections model for each pollutant (incl.
NOx and NO2)
• Outputs on a 1x1km2 grid of background conditions and approx.
9,000 road links
PCM model
4. 4
• https://uk-air.defra.gov.uk/data/pcm-data
• Produced and updated on an annual basis
• “These maps should not be mistaken for the Local Authority-
specific maps which are alternative maps based on the same model
results as presented here but which provide source-sector splits and
projections to future years by Local Authority to aid the Local Air
Quality Management process.”
• Values are the same as the LAQM background maps.
PCM model – background concentrations
5. 5
PCM road links
• road link within 1x1km2 grid
• background value and
roadside value
• background concentration
varies by road link
https://uk-air.defra.gov.uk/library/no2ten/2015-
no2-projections-from-2013-data
LAQM background maps
• 1x1km2 grid
• source-sector splits
• available for each local
authority
• annual projections to 2030
• use within LAQM process
• base year 2013 (published
July 2016)
https://uk-air.defra.gov.uk/data/laqm-
background-maps?year=2013
Background concentrations
6. 6
PCM model resultsLAQM background maps
Background concentrations
– – 1 km – –
––1km––
– – 1 km – –
––1km––
7. 7
• Comparison originally undertaken in Autumn 2015
• Years of assessment: 2013 and 2025
• Background concentrations from LAQM maps (2011 based) and
PCM road links (2013 based)
• Conclusions:
- Magnitude of difference greater for 2013 than 2025
- 2013: PCM > LAQM by 0-6μg/m3
- 2025: PCM < LAQM by 0-4μg/m3
Background concentrations
10. 10
Background concentrations
• Difference in NO2 concentrations between LAQM backgrounds
and PCM road link background values for Zone 1 (Greater
London)
• 20% of values with a difference in concentrations
-15
-10
-5
0
5
10
15
PCM2013-LAQM2013(NO2)
-15
-10
-5
0
5
10
15
PCM2025-LAQM2025(NO2)
2013 2025
11. 11
Background concentrations
• Distribution of differences in NO2 concentrations using “bins” of
1μg/m3 or 2μg/m3
< -6 -6 to -4 -4 to -2 -2 to -1 -1 to 0 0 to +1 +1 to +2 +2 to +4 +4 to +6 > +6
< -6 -6 to -4 -4 to -2 -2 to -1 -1 to 0 0 to +1 +1 to +2 +2 to +4 +4 to +6 > +6
2013
2025
PCM NO2 – LAQM NO2
PCM NO2 – LAQM NO2
12. 12
• Comparison of NO2 concentrations between LAQM backgrounds
2013 and 2011 based.
• For the year 2013, the new maps have higher NO2 concentrations
typically by 0-6μg/m3
Expected since the new base year is 2013 and the model has
been adjusted based on measurements for that year.
• For the year 2025, the new maps have lower NO2 concentrations
typically by 0-4μg/m3
The new maps predict a greater improvement in background
concentrations for the future years.
Background concentrations
14. 14
• AQC report on future background concentrations1
• They predict that concentrations will reduce with time, but not in
the same extent as predicted in the Defra maps.
• Method for uplifting future backgrounds
Chiswick example: suggests that
future NO2 backgrounds should
be 15% higher than the LAQM
values
Background concentrations
1 Air Quality Consultants (2016) Deriving background concentrations
of NOx and NO2 for use with ‘CURED V2A’
16. 16
• The figure shows
the outcome of model
verification for
roadside sites
• Most modelled
values are within 30%
of the monitored
value
• ….need to delve
deeper than a scatter
plot which masks the
overall picture,
especially around
40μg/m3
Model trends and uncertainty
40
40
from Defra AQ Plan Technical Report (Dec 2015)
17. 17
• Analysis of the
frequency of the
difference between
modelled and monitored
concentrations
- as % and as μg/m3
- negative values indicate
model under-prediction
• The model mostly
under-predicts
Model trends and uncertainty
Frequency of different percentage errors
Frequency of different absolute errors
18. 18
• Crucially, the model has a greater tendency to under-predict for
monitored concentrations close to 40µg/m3 (35-45µg/m3) and over
40µg/m3, than it does overall
Model trends and uncertainty
Difference in concentration (modelled – monitored)
as function of monitored concentration
19. 19
- If the limit value is used as
the assessment level, i.e.
40µg/m3, then there is a
48% likelihood that the
monitored concentration
would exceed 40µg/m3;
- Even if an assessment level
is selected that is 10% lower
than the limit value (i.e.
36µg/m3) then there is still a
37% likelihood that the
monitored concentration
would exceed 40µg/m3.
Model trends and uncertainty
Use of a lower, surrogate
threshold increases
confidence that the actual
threshold of 40µg/m3 will
be achieved.
7%
16%
32%
37%
48%
cf M. Bull (2010), H13-244, harmo.org
20. 20
• Model predictions are on a
road link basis: the same
concentration is predicted for all
roadside locations
• No account is taken of local
differences: junctions, bus stops
or particular traffic conditions,
such as queuing traffic
• Predictions are calculated at a
distance of 4m from kerbside
Performance at roadside sites in London
Monitored vs Modelled annual mean concentrations
at 46 automatic roadside monitoring stations (2013)
40
40
Model over-estimates at the lower
concentrations and under-estimate
at the higher concentrations.
There is substantial spread in the
difference between monitored and
predicted concentrations. 10 out of
the 46 model forecasts (22%), lie
outside the ±30% envelope
21. 21
• As the national model, PCM
has been used for over 10 years to
predict future pollutant
concentrations and hence to
predict UK compliance with the
EU limit values
• Consider predictions (for
London) published in 2007 to
support the 2006 review of the Air
Quality Strategy
• Top graph shows model
predictions for 2001 (a past year).
- Max under-estimate 8µg/m3
- Max over-estimate 14µg/m3
• Bottom graph shows model
predictions for 2010.
- Max under-estimate 26µg/m3
- Max over-estimate 4µg/m3
C. Quality of past predictions
2001 modelled and monitored
2010 modelled and monitored
Data from AEAT/ENV/R/2456/Issue 1 report to Defra, January 2007
22. 22
• Forecasts are sensitive to input data and to the
‘base year’
- For 2001 (a past year), max difference: 14µg/m3
(average 5µg/m3)
- For 2010, max difference: 11µg/m3 (avg 4µg/m3)
- For 2020, max difference: 12µg/m3 (avg 4µg/m3)
Quality of past predictions
Different base years: 2003, 2004
Sensitivity to base year is striking
No sensitivity of the model
predictions to base year or similar
inputs has been presented within
Defra’s 2015 Air Quality Plan
From CERC FM642/TR04/R1/06 report to Defra & devolved authorities, 2006
23. 23
• The background concentrations of the PCM road links are from the count point, even
though the links may cross through multiple grid squares.
• The latest background maps predict an even greater improvement in future background
concentrations than the previous ones.
• To increase confidence in forecasts of compliance with the 40mg/m3 limit value,
forecasts of achieving a lower, surrogate level of concentration should be used. Using
40mg/m3 there is a 48% likelihood of exceeding the limit value
• At roadside sites in London the model over-estimates at the lower concentrations
and under-estimates at the higher concentrations
• Previous forecasts
- Forecasts of concentrations at the highest recording sites are over-optimistic with respect to
predicting compliance with EU limit value;
- Model forecasts are sensitive to the input data and default parameters. Choice of a different
base year caused a difference of up to 4.5mg/m3 in model predictions (over 10% of the EU
limit value)
Conclusions