Building a Conceptual Model for PM over Hong Kong: A Weight-of-Evidence Approach to Evaluating Source Apportionment Results Jay Turner, Varun Yadav Washington University in St. Louis Zibing Yuan, Alexis Lau Hong Kong University of Science and Technology Peter Louie Hong Kong Environmental Protection Department Air Quality Management December 6-7, 2012 Mumbai, India
37
Embed
Building a Conceptual Model for PM over Hong Kong: A Weight-of-Evidence Approach to Evaluating Source Apportionment Results Jay Turner, Varun Yadav Washington.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Building a Conceptual Model for PM over Hong Kong: A Weight-of-Evidence Approach to Evaluating
Source Apportionment Results
Jay Turner, Varun Yadav
Washington University in St. Louis
Zibing Yuan, Alexis LauHong Kong University of Science and Technology
Peter LouieHong Kong Environmental Protection Department
Air Quality ManagementDecember 6-7, 2012Mumbai, India
Integrated Data Analysis and Characterization of Particulate Matter in Hong Kong
• Project funded by Hong Kong Environmental Protection Department (HKEPD)
• Update PM10 and PM2.5 source apportionments for Hong Kong
• Develop a conceptual model for ambient PM over Hong Kong
Conceptual Model Framework
underlined = this study
Data Analysis Approach
PM10, PM2.5
(24 hour, speciated)TEOM
(hourly, PM10)Air Mass Back
Trajectories
Source Apportionment
Air MassClustering
Baseline-Excess
Source Contribution
Estimates
ExcessConcentration
Total Mass
Local/RegionalContributions
Datasets
Approach
TemporalTrends
Air Mass basedTemporalTrends
Today’s Presentation
• A series of snapshots from the conceptual model development (process-focused)
• Examples from the weight-of-evidence used to support the source apportionment
Hong Kong and Haze
January 2012
Hong Kong Air Quality: Two Metrics for Long-Term Trends
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
PM
10 S
peci
es A
nnua
l Ave
rage
( g
/m3)
0
5
10
15
20
25
% o
f ho
urs
with
vis
ibili
ty <
8 k
m (
RH
< 8
0%)
0
5
10
15
20
25
TE
OM
PM
10 A
nnua
l Ave
rage
( g
/m3)
0
15
30
45
60
75
Sulfate (SO42-)
Total Carbon (TC) VisibilityTEOM PM10
particulate matter mass
hours with poor visibility
Hong Kong Air Quality – Long Term Trends
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
PM
10 S
peci
es A
nnua
l Ave
rage
( g
/m3)
0
5
10
15
20
25
% o
f ho
urs
with
vis
ibili
ty <
8 k
m (
RH
< 8
0%)
0
5
10
15
20
25
TE
OM
PM
10 A
nnua
l Ave
rage
( g
/m3)
0
15
30
45
60
75
Sulfate (SO42-)
Total Carbon (TC) VisibilityTEOM PM10
dramatic change in particulate matter composition, 1998-2003
Yuan, Lau, Yadav, Turner, and Louie (submitted)
Hong Kong – A Complex Setting
Pearl River Delta (PRD)
Hong Kong PM10 Speciation Network
• Long time series (12+ years)• High volume sampler
– Measurement artifacts for nitrate and organic carbon
• 1-in-6 day 24-hour integrated samples– Seven sites routinely operated since 1998
(additional sites for portions of this time period)• Sampling not synchronized across sites
- Can’t assess day-specific spatial variability- Can construct a daily time series for Hong Kong
region, at least for regional scale components
Hong Kong PM10 Speciation Network
Yuen Long
Tung Chung Central / Western
Tsuen Wan
Sham Shui Po
Kwun Tong
Mong Kok
HONG KONG SAR
Shenzhen City, Guangdong Province
Roadside Station
General Station
Container Terminal
HKIA
Shekou Port
Yuen Long
Tung Chung Central / Western
Tsuen Wan
Sham Shui Po
Kwun Tong
Mong Kok
HONG KONG SAR
Shenzhen City, Guangdong Province
Roadside Station
General Station
Container Terminal
HKIA
Shekou Port
Modeling Tools – This Study
Multivariate Factor Analysis Models
– Positive Matrix Factorization (PMF)
– Principal Components Analysis (PCA) with Absolute Principle Components Scores (APCS)
• Absolute Principle Components Analysis (APCA)
– Unmix
Other Multivariate Models
– Chemical Mass Balance (CMB)
Source Apportionment - Steps
Collection Validation Exploration
Data Conditioning
Model Execution
Evaluation Interpretation
data
modeling
assessment
13
Source Apportionment – This Study
Collection Validation Exploration
Data Conditioning
Model Execution
Evaluation Interpretation
data
modeling
assessment
Data Quality Assessment
14
• Secondary sulfate factor gives the largest contribution, accounting for 22% of the ambient PM10 in HK.
• Contributions from vehicle exhaust, aged sea salt, secondary nitrate, and coal combustion / biomass burning factors are comparable, with each around 15%.
• Contributions from residual oil, fresh sea salt, crustal soil, and zinc smelting (trace metals) factors are generally <5%.
15
PM10 Source Contribution Percentages
(marine source)
(maybe)
Annual Variation of PM10 Source Contributions (μg/m3)Roadside and General Stations
Different Models,SometimesDifferent Results
• We largely understand why, but beyond the scope of this discussion
Some aspects make us a bit nervous…
• Is soil apportioned correctly?• If so, do we understand its source?• Are the profiles and data even comparable (analytical )
Vehicle Exhaust Contributions : Source Apportionment Modeling versus Emission Inventory
Hong Kong air pollutant emission inventory, available at: http://www.epd.gov.hk/epd/english/environmentinhk/air/data/emission_inve_rsp_C.html
PM Emissions from Road Transport in Hong Kong, tonnes/year
0 1000 2000 3000 4000 5000
gen
eral stations veh
icle exhau
st sou
rce contrib
ution
from P
MF
, g/m
3
0
2
4
6
8
10
reduced major axis (RMA) regression:
94.0
0.16.00004.00020.02
R
EISCE
our modeling agrees with the (independently developed)
emissions estimates!
Hong Kong PM2.5 Speciation Network
• 1-in-3 day 24-hour integrated sampling with three PM2.5 FRM samplers
• Three-to-four sites operated for three one-year periods over the past decade
• Different samplers and analyses compared to PM10
• Lab-reported error structures strong function of analysis batch
• Example… Pb for 2008-2009 sampling campaign
• Dashed lines is the error structure from the collocated data
PM2.5 Error Structures
Pb concentration, ug/m3
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
Pb
un
cert
ain
ty,
ug
/m3
0.000
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
Dec 08 - Feb 09Mar 09 - May 09Jun 09 - Aug 09Sep 09 - Dec 09from collocated data
PM2.5 versus PM10 Source Contributions
Relating Pollution to Synoptic Air Mass Transport Patterns
• Relate observed concentration at receptor to air mass transport history– Air mass back trajectories, e.g., HYSPLIT
• Approach #1… Assign concentration at receptor to points along back trajectory and estimate ensemble relationships for impacts, e.g.,– Potential source contribution function (PSCF)– Quantitative transport bias analysis (QTBA)
• Approach #2… Classify air mass trajectories based on transport pattern similarities, independent of air pollutant data– Use these “clusters” in air pollutant data analyses– Not the same as synoptic typing (e.g., with PCA)
PSCF Example – Sulfate in St. Louis
24
Sulfate Potential Source Contribution Function (PSCF) analysis, Lee and Hopke (2006)Chemical Speciation Network Data (1-in-3 day)
1Dorling, S.R., Davies, T.D., and Pierce C.E. (1992): Cluster Analysis: A technique for estimating the synoptic meteorological controls on air and precipitation chemistry – methods and applications. Atmospheric Environment 26A, 2575-2581.
2Moody, J.L., Munger, J.W., Goldstein, A.H., Jacob, D.J. and Wofsy, S.C. (1998): Harvard Forest regional-scale air mass composition by Patterns in Atmospheric Transport History (PATH). Journal of Geophysical Research 103, 13181-13194.
Clustering the Trajectories• Many approaches
– In this study we used Dorling’s algorithm1 with refinements
Clustering of Seven-Day Air Mass Back Trajectories
• Generate air mass back trajectories (4/day for 11 years… more than 16,000 trajectories!)
• Cluster into five categories
• Examine relationships between particulate matter composition, concentration, and air mass transport pattern
Air Mass Clusters – This Study
• East Coast of China, relatively fast moving (ECC Fast)• East Coast of China, relatively slow moving (ECC Slow)• East• South/Southwest (S/SW)• Stagnant
27
Cluster-Censored Trends in Source Contributions
yr)-3
g/(m 06.014.0slope
motor vehicle
cluster class
Slow ECC Fast ECC Stagnant S/SW East
sca
led
co
nce
ntr
atio
n
-1
0
1
2
3
4
5
Vehicular Exhaust
Cluster-Censored Trends in Source Contributions
yr)-3
g/(m 06.012.0slope yr)-3
g/(m 06.014.0slope
secondary sulfate factor
cluster class
Slow ECC Fast ECC Stagnant S/SW East
sca
led
co
nce
ntr
atio
n
-1
0
1
2
3
4
5
Secondary Sulfate
motor vehicle
cluster class
Slow ECC Fast ECC Stagnant S/SW East
sca
led
co
nce
ntr
atio
n
-1
0
1
2
3
4
5
Vehicular Exhaust
Thank you, mainland China!
Hong Kong PM10 Mass Network
• Compared to the Air Quality Objectives• Hourly data set for more than ten years