Pm modeling and source apportionment l.jpg
This presentation is the property of its rightful owner.
Sponsored Links
1 / 55

PM Modeling and Source Apportionment PowerPoint PPT Presentation


  • 62 Views
  • Uploaded on
  • Presentation posted in: General

PM Modeling and Source Apportionment. Amit Marmur, Dan Cohan, Helena Park, Jeameen Baek, Sangil Lee, Mei Zhang, Jim Boylan, Katie Wade,Jim Mulholland, …, and Armistead (Ted) Russell Georgia Institute of Technology. With Special Thanks to:. Eric Edgerton, Ben Hartsell and John Jansen

Download Presentation

PM Modeling and Source Apportionment

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Pm modeling and source apportionment l.jpg

PM Modeling and Source Apportionment

Amit Marmur, Dan Cohan, Helena Park, Jeameen Baek, Sangil Lee, Mei Zhang, Jim Boylan, Katie Wade,Jim Mulholland, …, and

Armistead (Ted) Russell

Georgia Institute of Technology

Georgia Institute of Technology


With special thanks to l.jpg

With Special Thanks to:

  • Eric Edgerton, Ben Hartsell and John Jansen

    • for making the required observations possible as part of SEARCH

      • Southeastern Aerosol Research and Characterization study

    • Discussions and additional analyses

  • Mike Kleeman

    • Additional source apportionment calculations (see also, 1PE11)

  • Phil Hopke

  • Paige Tolbert and the Emory crew

    • As part of ARIES, SOPHIA, and follow on studies

  • NIEHS, US EPA, FHWA, Southern Company, SAMI

    • Financial assistance

  • And more…

Georgia Institute of Technology


Genesis l.jpg

Genesis

  • (How) Can we use “air quality models” to help identify associations between PM sources and health impacts?

    • Species vs. sources

      • E.g., Laden et al., 2000

Georgia Institute of Technology


Epidemiology l.jpg

Epidemiology

  • Identify associations between air quality metrics and health endpoints:

Health endpoints

Statistical

Analysis

(e.g. time series)

Sulfate

Association

Georgia Institute of Technology


Association between cvd visits and air quality l.jpg

Association between CVD Visits and Air Quality

Georgia Institute of Technology

(See Tolbert et al., 9C2)


Issues l.jpg

Issues

  • May not be measuring the species primarily impacting health

    • Observations limited to subset of compounds present

  • Many species are correlated

    • Inhibits correctly isolating impacts of a species/primary actors

      • Inhibits identifying the important source(s)

  • Observations have errors

    • Traditional: Measurement is not perfect

    • Representativeness (is this an error? Yes, in an epi-sense)

  • Observations are sparse

    • Limited spatially and temporally

  • Multiple pollutants may combine to impact health

    • Statistical models can have trouble identifying such phenomena

  • Ultimately want how a source impacts health

    • We control sources

Georgia Institute of Technology


Use aq models to address issues link sources to impacts l.jpg

Health Endpoints

Statistical

Analysis

Use AQ Models to Address Issues: Link Sources to Impacts

Data

Air Quality

Model

Source

Impacts

S(x,t)

Association between

Source Impact

and Health Endpoints

Georgia Institute of Technology


Use aq models to address issues assess error s provide increased coverage l.jpg

Use AQ Models to Address Issues: Assess Errors, Provide Increased Coverage

Data

Air Quality

Model

Air Quality

C(x,t)

Health Endpoints

Site

Representative?

Association between

Concentrations

and Health Endpoints

Monitored

Air Quality

Ci(x,t)

Georgia Institute of Technology


Slide9 l.jpg

But!

  • Model errors are largely unknown

    • Can assess performance (?), but that is but part of the concern

      • Perfect performance not expected

        • Spatial variability

        • Errors

  • Trading one set of problems for another?

    • Are the results any more useful?

Georgia Institute of Technology


Pm modeling and source apportionment10 l.jpg

PM Modeling and Source Apportionment*

  • What types of models are out there?

  • How well do these models work?

    • Reproducing species concentrations

    • Quantifying source impacts

  • For what can we use them?

  • What are the issues to address?

  • How can we reconcile results?

    • Between simulations and observations

    • Between models

Georgia Institute of Technology

*On slide 10, the talk starts…


Slide11 l.jpg

Emissions-

Based

Hybrid

Receptor

CMB

FA

Lag.

Eulerian (grid)

PMF

Molec. Mark.

Norm.

Source

Specific*

“Mixed PM”

UNMIX

PM (Source Apportionment) Models(those capable of providing some type of information as to how specific sources impact air quality)

PM Models

Georgia Institute of Technology

*Kleeman et al. See 1E1.


Source based models l.jpg

Chemistry

Source-based Models

Air Quality Model

Emissions

Meteorology

Georgia Institute of Technology


Source based models13 l.jpg

Source-based Models

  • Strengths

    • Direct link between sources and air quality

    • Provides spatial, temporal and chemical coverage

  • Weaknesses

    • Result accuracy limited by input data accuracy (meteorology, emissions…)

    • Resource intensive

Georgia Institute of Technology


Receptor models l.jpg

Receptor Models

Obsserved

Air Quality

Ci(t)

Source

Impacts

Sj(t)

Ci - ambient concentration of specie i (g/m3)

fi,j - fraction of specie i in emissions from source j

Sj - contribution (source-strength) of source j (g/m3)

Georgia Institute of Technology


Receptor models15 l.jpg

Receptor Models

  • Strengths

    • Results tied to observed air quality

      • Reproduce observations reasonably well, but…

    • Less resource intensive (provided data is available)

  • Weaknesses

    • Data dependent (accuracy, availability, quantity, etc.)

      • Monitor

      • Source characteristics

    • Not apparent how to calculate uncertainties

    • Do not add “coverage” directly

Georgia Institute of Technology


Hybrid inverse model approach l.jpg

Hybrid: Inverse Model Approach*

INPUTS

Emissions (Eij(x,t))

Ci(x,t), Fij(x,t), & Sj(x,t)

Air Quality

Model +

DDM-3D

Other Inputs

Receptor Model

New emissions:

Eij(x,t)

Observations taken

from routine measurement

networks or special

field studies

Main assumption in the formulation:

A major source for the discrepancy between predictions and observations are the emission estimates

*Other, probably better, hybrid approaches exist

Georgia Institute of Technology


Source apportionment application l.jpg

Source Apportionment Application

  • So, we have these tools… how well do they work?

  • Approach

    • Apply to similar data sets

      • Compare results

      • Try to understand differences

    • Primary data set:

      • SEARCH1 + ASACA2

        • Southeast… Atlanta focus

        • Daily, speciated, PM2.5 since 1999

Georgia Institute of Technology

1. Edgerton et al., 4C1; 2. Butler et al., 2001


Search asaca l.jpg

rural

suburban

urban

Yorkville (YRK)

North Birmingham (BHM)

Jefferson Street (JST)

Centreville (CTR)

Oak Grove (OAK)

Outlying Landing Field #8 (OLF)

Gulfport (GFP)

Pensacola (PNS)

SEARCH & ASACA

SEARCH

Funding from EPRI, Southern Company

ASACA

Georgia Institute of Technology


Questions l.jpg

Questions

  • How consistent are the source apportionment results from various models?

  • How well do the emissions-based models perform?

  • How representative is a site?

  • What are the issues related to applying source apportionment models in health assessment research?

  • How can we reconcile results?

Georgia Institute of Technology

*On slide 10, the talk starts…


Source apportionment results l.jpg

Source Apportionment Results

  • Hopke and co-workers (Kim et al., 2003; 2004) for Jefferson Street SEARCH site (see, also 1PE4…)

Average Source Contribution

}22

  • Notes:

  • CMB-MM from Zheng et al., 2002 for different periods, given for comparison

  • Averaged results do not reflect day-to-day variations

Georgia Institute of Technology


Slide21 l.jpg

Daily Variation

LGO-CMB: see

Marmur et al.,

6C1

PMF: See Liu et al., 5PC7

Georgia Institute of Technology


Receptor models22 l.jpg

Receptor Models

  • Approaches do not give “same” source apportionment results… yet

    • Relative daily contributions vary

      • Important for associations with health studies

        • Introduces additional uncertainty

    • Long term averages more similar

      • More robust for attainment planning

  • Using receptor-model results directly in epidemiological analysis has problem(s)

    • Results often driven by one species (e.g., EC for DPM), so might as well use EC, and not introduce additional uncertainty

    • No good way to quantify uncertainty

Georgia Institute of Technology


Emissions based model ebm source apportionment l.jpg

Emissions-based Model (EBM)Source Apportionment

  • Southeast: Models 3

    • DDM-3D sensitivity/source apportionment tool

    • Modeled 3 years

      • Application to health studies

        • Provides additional chemical, spatial and temporal information

        • Allows receptor model testing

      • Concentrate on July 01/Jan 02 ESP periods

        • Compare CMAQ with molecular marker CMB

  • California: CIT (Kleeman)

  • But first… model performance comments

    • CAMX-PM (Pandis), URM (SAMI), CMAQ (VISTAS)

Georgia Institute of Technology


Slide24 l.jpg

Species of PM 2.5 in JST

MODEL(CMAQ) OBS

29.42 (mg/m3) 22.53 (mg/m3)

January 2002 July 2001

28.07 (mg/m3) 13.28 (mg/m3)

Winter problem largely nitrate + ammonium

Georgia Institute of Technology


Sami urm l.jpg

SAMI: URM

Georgia Institute of Technology


Performance l.jpg

Performance

Sulfate

EPI OC

FAQS*

Simulated a bit low:

Analyses suggests

SOA low

VISTAS

*Fall Line Air Quality Study,

Epi: 3-year modeling,

VISTAS: UCR/ENVIRON

Georgia Institute of Technology


Slide27 l.jpg

Mean Fractional Error: Combined Studies

Georgia Institute of Technology

Plot by J. Boylan


Slide28 l.jpg

VISTAS PM Modeling Performance

Georgia Institute of Technology

Modeling conducted by ENVIRON, UC-Riverside. Plot by J. Boylan


Slide29 l.jpg

Species of PM 2.5

(OBS:Left column, MODEL(CMAQ): right column)

OBS

MODEL (CMAQ)

January 2002 July 2001

Too much simulated nitrate and soil dust in winter

Georgia Institute of Technology


Performance30 l.jpg

Performance

  • PM Performance (Seignuer et al., 2003; see also 6C2)

    • Errors from recent studies using CMAQ, REMSAD

      • Organic carbon: 50-140% error

      • Nitrate: 50-2000% error

    • Understand the reason for much of the error in nitrate

      • Deposition, heterogeneous reaction

      • Ammonia emissions still rather uncertain

    • OC more difficult

      • Understand part

        • Heteorgenous paths not included

      • More complex mixture

      • Primary/precursor emissions more uncertain

Nitrate

Georgia Institute of Technology


Slide31 l.jpg

Predicted vs. Estimated in Organic Aerosol in Pittsburgh (Pandis and co-workers)Primary and Secondary OA

Predicted [g/m3]

Estimated [g/m3]

• EC Tracer Method (Cabada et al., 2003)

See also 4D4, 5D2…

Georgia Institute of Technology


Limitations on model performance l.jpg

Limitations on Model Performance

  • The are (should be) real limits on model performance expectations

    • Spatial variability in concentrations

    • Spatial, temporal and compositional “diffusion” of emissions

    • Met model removal of fine scale (temporal and spatial) fluctuations (Rao and co-workers)

    • Stochastic, poorly captured, events (wildfires, traffic jams, upsets, etc.)  

    • Uncertainty in process descriptions and other inputs

      • Heterogeneous formation routes

Georgia Institute of Technology


Spatial variability l.jpg

Spatial Variability

  • Spatial correlation vs. temporal correlation (Wade et al., 2004)

    • Power to distinguish health associations in temporal health studies

    • Sulfate uniform, EC loses correlation rapidly

  • Data withholding using ASACA data:

    • Interpolate from three other stations, compare to obs.

    • EC: Norm. Error=0.6

      • TC: 0.2!

    • Sulfate: NE = 0.12

EC

Sulfate

Georgia Institute of Technology


Emissions diffusion l.jpg

Emissions “Diffusion”

On-road OC Emissions

Dial Variation of ATL emissions

Default profile (black) vs. plane/engine dependent operations (red)

Chemical dilution: assume source X has same emissions composition, independent of location, etc. (~)

Nonroad OC Emissions

Georgia Institute of Technology


Slide35 l.jpg

Wildfire and Prescribed burn

Capturing stochastic events using satellites:

32

3.4

56

19

19

51

Black: estimates based on fire records

Red: estimates based on satellite images (Ito and Penner, 2004)

Georgia Institute of Technology


Slide36 l.jpg

Sulfate Mean Fractional Error

X

Spatial variability limit?

Georgia Institute of Technology


Slide37 l.jpg

EC Mean Fractional Error

X

Georgia Institute of Technology


How good are they l.jpg

How Good Are They?

  • All evidence suggests that they describe the processes most affecting the evolution of ozone and (if equipped) particulate matter (o.k., many components of PM) after pollutant emission

Now getting sufficient data

Evaluation

Application

Computational

implementation

Mathematics

Science (chemistry/physics)

Georgia Institute of Technology


How good are they39 l.jpg

How Good Are They?

  • All evidence suggests that they describe the processes most affecting the evolution of ozone and (if equipped) particulate matter (o.k., many components of PM) after pollutant emission

Now getting sufficient data:

Holes will get filled

Evaluation

Application

Computational

implementation

Mathematics

Science (chemistry/physics)

Georgia Institute of Technology


Emissions based model performance l.jpg

Emissions-based Model Performance

  • Some species well captured

    • Sulfate, ammonium, EC(?)

  • “Routine” modeling has performance issues

    • Multiple causes

      • Species dependent

    • OC tends to be a little low

      • Heterogeneous formation? (or emissions or meteorology)

  • Some “research-detail” modeling appears to capture observed levels relatively well

    • Finer temporal variation captured as well

  • Real limits on performance

    • Data with-holding and statistical analysis suggests model performance may be limited due to spatial variability (5PC5)

  • Longer term averages look reasonable for most species

    • Nitrate high

  • This is not an evaluation of source-apportionment accuracy

    • But it is an indication of how well one might do

Georgia Institute of Technology


Slide41 l.jpg

Source apportionment of PM 2.5 in JST

CMAQ CMB

24.42 (mg/m3) 22.53 (mg/m3)

January 2002 July 2001

28.07 (mg/m3) 13.28 (mg/m3)

Georgia Institute of Technology


Slide42 l.jpg

Source apportionment of PM 2.5

(CMB:Left column, CMAQ: right column)

CMB

CMAQ

January 2002 July 2001

Georgia Institute of Technology


Slide43 l.jpg

Source apportionment of PM 2.5 in JST (July 2001)

(CMB: 1st column, CMAQ (12km): 2nd column, CMAQ (36km): 3rd column)

CMB with MM

CMAQ (12 km)

CMAQ (36 km)

Reasonable

agreement…

Georgia Institute of Technology

Note. CMB data are missing on July 1, 2, 5, 11, 22, 24, and 28.


Slide44 l.jpg

Source apportionment of PM 2.5 in JST (Jan 2001)

(CMB: 1st column, CMAQ (12km): 2nd column, CMAQ (36km): 3rd column)

CMB with MM

CMAQ (12 km)

CMAQ (36 km)

Remarkable agreement…

most

others not

Georgia Institute of Technology


Cmaq vs cmb primary pm source fractions l.jpg

CMAQ vs. CMB* Primary PM Source Fractions

More variation than I would expect in emissions and large volume average

Georgia Institute of Technology

*Not using molecular markers


California kleeman see 1pe11 l.jpg

California (Kleeman: see 1PE11)

Georgia Institute of Technology


Ebm application site representativeness l.jpg

EBM Application: Site Representativeness

  • Compare observations to each other and to model results to help assess site representativeness

    • Grid model provides volume-averaged concentrations

      • Desired for health study

  • Assessed representativeness of Jefferson Street site used in epidemiological studies

    • Found it better correlated with simulations for most species than other Atlanta sites

Georgia Institute of Technology


Results so 4 2 l.jpg

Results: SO4-2

Georgia Institute of Technology


Emissions based models l.jpg

Emissions-Based Models

  • EBM’s can provide additional information

    • Coverage (chemical, spatial and temporal)

      • Intelligent interpolator

    • Source contributions

  • Relatively little day-to-day variation in source fractions from EBM

    • Reflects inventory

    • May not be capturing sub-grid(?... Not really grid) scale effects

      • Inventory is spatially and temporally averaged

      • May inhibit use for health studies

  • Agreement between EBM and CBM good, at times, less so at others

    • Longer term averages look reasonable:

      • Applicable for control strategy guidance, with care

        • understand limitations

    • Not apparent which is best

Georgia Institute of Technology


Getting back to health association application what s best l.jpg

Getting back to Health Association Application: What’s Best?

Air Quality

Model SA

Data

Health

Endpoints

Source-Health

Associations

Air Qual.

Data

Species-

Health

Associations

Air Quality

Model SA

Georgia Institute of Technology


Slide51 l.jpg

Or?

Observd

Air Quality

C(x,t)

Air Quality

Model

C(x,t), S(x,t)

Understanding

Of AQM & Obs.

Limitations

Data

C(x,t), S(x,t)

Health

Endpoints

Source/Species

Health Associations

Georgia Institute of Technology


Summary l.jpg

Summary

  • Application of PM Source apportionment models in health studies more demanding than traditional “attainment-type” modeling

    • New (and relatively unexplored) set of issues

  • Receptor models do not, yet, give same results

    • Nor do they agree with emissions-based model results (that’s o.k. for now)

    • Need a way to better quantify uncertainty

    • If results driven by a single species, little is gained, for epi application

  • Receptor models (probably) lead to excess variability for application in health studies

    • Representativeness error

    • Not yet clear if model application, itself, decreases or increases representativeness error over directly using observations

  • Emissions-based models

    • Likely underestimate variability (too tied to minimally varying inventory)

    • Performance is spotty

  • Groups actively trying to reconcile differences

    • Focus on emissions, range of observations, applying different models

    • Hybrid approaches?

Georgia Institute of Technology


Acknowledgements l.jpg

Acknowledgements

  • Staff and students in the Air Resources Engineering Center of Georgia Tech

  • SEARCH, Emory, Clarkson, UC Davis teams.

  • SAMI

  • GA DNR

  • Georgia Power

  • US EPA

  • NIEHS

  • Georgia Tech

Georgia Institute of Technology


Effect of grid resolution l.jpg

Effect of Grid Resolution

(4x too big)

Georgia Institute of Technology


Slide55 l.jpg

Georgia Institute of Technology


  • Login