Uncertainties in PM2.5 Gravimetric and Speciation Measurements and

What Can We Learn from Them

• William Malm• Cooperative Institute for Research in the Atmosphere, Colorado State University,

Fort Collins, CO 80523-1375

• Bret A. Schichtel • National Park Service-Air Resources Division, CIRA, Colorado State University,

Fort Collins, CO 80523-1375

• Marc L. Pitchford• National Oceanic and Atmospheric Administration-Air Resources Laboratory, c/o

Desert Research Institute, 755 E Flamingo Rd, Las Vegas, NV 89119-7363

Non-urban and urban PM2.5

networks

Non-urban and urban PM2.5

networksInteragency Monitoring of

PROtected Visual Environments

(IMPROVE) Network

Interagency Monitoring of

PROtected Visual Environments

(IMPROVE) Network

Chemical Speciation Network (CSN)

Chemical Speciation Network (CSN)

Virgin IslandsVirgin Islands

Neighborhood- and Urban-Scale

Monitoring Objectives

Neighborhood- and Urban-Scale

Monitoring Objectives

• Determine compliance with air quality

standards

• Validation of regional scale models

• Assess source/receptor relationships

• Evaluate health, radiative, and

ecological effects

• Determine compliance with air quality

standards

• Validation of regional scale models

• Assess source/receptor relationships

• Evaluate health, radiative, and

ecological effects

IMPROVE Monitoring ObjectivesIMPROVE Monitoring Objectives

• Tracking long term temporal changes in

visibility (extinction)

• Assess source/receptor relationships

• Serve as a regional backdrop for

special studies and understanding non

urban background

• Used for regional modeling validation

studies

• Tracking long term temporal changes in

visibility (extinction)

• Assess source/receptor relationships

• Serve as a regional backdrop for

special studies and understanding non

urban background

• Used for regional modeling validation

studies

OBJECTIVES

• Estimate average bias in gravimetric PM2.5 measurement

• Estimate average bias in reconstructed PM2.5 from measured aerosol species

• Estimate average bias in reconstructed extinction using standard IMPROVE algorithms

PM2.5 Federal Reference

Methods (FRMs)

PM2.5 Federal Reference

Methods (FRMs)

Andersen RAASThermo Fisher Scientific, formerly Andersen Instruments, Smyrna, GA

URG MASSURG Corp., Raleigh, NC

Partisol SamplerThermo Fisher Scientific, formerly Rupprecht & Patashnick, Albany, NY

BGI PQ-200BGI, Inc., Waltham, MA

Speciation Monitors (EPA Speciation Network)Speciation Monitors (EPA Speciation Network)

Mass Aerosol Sampling System (MASS)URG Corporation, Raleigh, NC

Reference Ambient Air Sampler (RAAS)Andersen Instruments, Smyrna, GA

Spiral Aerosol Speciation Sampler (SASS)Met One Instruments, Grants Pass, OR

Interagency Monitoring of Protected Visual Environments (IMPROVE) SamplerAir Resource Specialists, Ft. Collins, CO

Other Speciation MonitorsOther Speciation Monitors

Dual Channel Sequential Filter Sampler

and Sequential Gas SamplerDesert Research Institute, Reno, NV

Dichotomous Virtual ImpactorAndersen Instruments, Smyrna, GA

Paired MinivolsAirmetrics, Inc., Springfield, OR

Partisol 2300 Speciation SamplerRupprecht & Patashnick, Albany, NY

Dichotomous Partisol-Plus SamplerRupprecht and Patashnick, Albany, NY

URG 3000NSpeciation Sampler URG Corporation, Raleigh, NC

Sampler Design Characteristics

47mmWhatma47mmWhatma47mmWhatma47mmWhatma47mmWhatma25mmPallQuartz Filter Type

QFQFQFQFQFQ For

QBQQuartz Filter Pack

Configuration

3rd day/6th day3rd day/6th day3rd day3rd day3rd day3rd daySampling Frequency

23.6cm/sec14.2cm/sec23.7cm/sec9.5cm/sec10.3cm/sec107.2cm/s

ecFilter Face Velocity

16.7l/min10.0l/min16.7l/min6.7l/min7.3l/min22.7l/minFlow Rate

242544Number of Channels

2214617918181Number of Sites (2006)

R&P2025

sequential FRM

R&P2300URG MASSMet One SASSAndersenIMPROVESampler type

CSNCSNCSNCSNCSNIMPROVENetwork

Speciation

SamplerVariable

Sampler Intercomparison and

Artifact Correction

• 1) Comparison of carbon across samplers

– Warren H. White, 16 April 2008 EPA

workshop

• 2) Converting the CSN carbon concentrations to match IMPROVE on the average

– Accounting for positive additive and

multiplicative negative artifacts

– Accounting for different thermal optical

methods.

11

Comparison of IMPROVE and CSN Carbon

at Collocated Sites

Anderson Met One R&P URG

urban collocations of CSN and IMPROVE carbon measurements

12

• The EC difference between CSN and IMPROVE shows little dependence on the

CSN sampler, suggesting that it is mainly analytical.

•IMPROVE:TOR (DRI) – CSN: TOT(NIOSH)

13

0

5

10

15

20

0 5 10 15 20

CSN TC, ug/m3

IMP

RO

VE

TC

+, ug/m

3MetOne 6.7 lpm

Anderson 7.3 lpm

R&P 16.7 lpm

URG 16.7 lpm

2005

For TC, unlike EC,

different CSN

samplers show

different biases

relative to

IMPROVE.

Note that the CSN offset is not determined simply by flow rate.

14

0

5

10

15

20

0 5 10 15 20

MetOne TC, ug/m3

IMP

RO

VE

TC

+, ug/m

3

Dec, Jan, Feb

Mar, Apr, May

Jun, Jul, Aug

Sep, Oct, Nov

2005

The CSN offset

shows no obvious

seasonality. In that

respect it behaves

more like IMPROVE

field blanks than

IMPROVE backup

filters.

15

0

5

10

15

20

0 5 10 15 20

MetOne TC, ug/m3

IMP

RO

VE

TC

+, ug/m

3

Birmingham

Detroit

Fresno

Phoenix

Pittsburgh

Rubidoux

2005

Relating IMPROVE and CSN TCMaking CSN look like IMPROVE

• Recall - Measured TC has a positive OC artifact.

- IMPROVE corrects for the artifact but CSN does not

- IMPROVE TC has a negative multiplicative artifact due loss of OCfrom high face velocities (FV)

- IMPROVE FV = 107.2 cm/s

- CSM MetOne FV = 9.5 cm/s

- Assume CSN negative artifact is 0

• Then[TC]IMP = [TC] – BIMP[OC] B - multiplicative artifact

[TC]CSN = [TC] + ACSN/VMetOne A/V – additive artifact

• Combine[TC]CSN = [TC]IMP + BIMP[OC] + ACSN/VMetOne

The multiplicative artifact (1+bOC) and the monthly positive additive organic artifact (a) used to relate CSN and IMPROVE carbon

concentrations. The units for the positive artifacts are µg/m3 and 1+bOC

is unitless.

1.1aDec

1.0aNov

1.2aOct

1.5aSep

1.9aAug

1.8aJul

1.7aJun

1.6aMay

1.4aApr

1.2aMar

1.3aFeb

1.1aJan

1.21+bOC

MetOne

Correction Factors

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

1+bO

C

Jan

Feb Mar

Apr

ilM

ayJu

ne

July

Aug Sep Oct

Nov Dec

Fac

tor

OFFSET

Comparison of CSN and adjusted

CSN to IMPROVE Carbon

Governing equations[ ]

))))lablablablab(RH(RH(RH(RHssssssss''''SSfSSfSSfSSfSoilSoilSoilSoil LACLACLACLACstuff)stuff)stuff)stuff) ofofofof (lots(lots(lots(lotsococococOCROCROCROCR))))lablablablab(RH(RH(RH(RHffff 3333NONONONO4444NHNHNHNH4444xSOxSOxSOxSOPMPMPMPM ''''inorginorginorginorg2.52.52.52.5

+

++++=

• x=(NH4)2SO4/H2SO4<=1.35 – lowest in summer – acidic aerosol

• highest in summer (D/Do)3≈1.3

• FM biased low because of volatilization (nitrate, SVOC, etc?)

• Roc(lots of stuff) probably highest in summer months

• f‘ss(RHlab) ≈ 1.3

• LAC may have a multiplier

• Soil may be have some seasonal and spatial dependence

3333))))lablablablab(RH(RH(RH(RH

ooooDDDDDDDD

3333

speciesspeciesspeciesspeciesρρρρspeciesspeciesspeciesspeciesmix,mix,mix,mix,ρρρρ

))))lablablablab(RH(RH(RH(RH''''ffff

=

Bias Between Estimated Fine Mass

and Gravimetric Fine Mass and

Bias as function of massPM2.5 = 1.375*SO4 +1.29*NO3+1.8*OC+Other

Temporal plot of PM2.5-PM2.5avg and the percent difference between reconstructed

and gravimetric mass (Brigantine WR) %DM=[(PM2.5-RPM2.5)/PM2.5]*100 = b1+b2(cos(f(T))

Average percent seasonal variability between reconstructed and gravimetric

mass for the IMPROVE monitoring network

%DM=[(PM2.5-RPM2.5)/PM2.5]*100 = b1+b2(cos(f(T))

Average percent bias between reconstructed and

gravimetric mass for the IMPROVE monitoring

network. The red or green color indicates that

reconstructed mass is an over or underestimate of

gravimetric mass respectively.

Summary table of the percent seasonal variability and average bias of

reconstructed versus gravimetric mass

16822.7-17.77.003.4AVG BIAS (CSN)

1589.4-17.94.44-3.9AVG BIAS (IMPROVE)

16819.00.04.517.6% Seasonal

Variability (CSN)

15826.70.06.1414.9% Seasonal

Variability (IMPROVE)

NMaxMinStd DevMeanVariable

AssumptionsAssumptions

• Do not account for volatilization of SVOC’s

• SVOC loss on Teflon and quartz substrates are

the same

• Gravimetric mass of soil dust, seasalt, and non

volatilized POM are measured without bias

• Nitrate and sulfates are measured without loss

on nylon substrate and nitrate and sulfates are

fully ammoniated

IMPROVE

0

5

10

15

20

All Winter Spring Summer Fall

Co

nc

en

tra

tio

n (

ug

/m3

)

STN Suburban

0

5

10

15

20

All Winter Spring Summer Fall

Co

nc

en

tra

tio

n (

ug

/m3

)

STN Urban

0

5

10

15

20

All Winter Spring Summer Fall

Co

nc

en

tra

tio

n (

ug

/m3

)

LAC

POM

SEAS

SOIL

NO3

SO4

Stacked bar charts showing average concentrations of each species for all and seasonal data for IMPROVE, CSN suburban, and CSN urban.

Center City

Inorganic water retention, nitrate loss

and Roc factor

Water Fraction for Sulfate

1.00

1.05

1.10

1.15

1.20

1.25

1.30

Average Winter Spring Summer Fall

Fra

ction

Nitrate Loss (corrected for water)

0.00

0.10

0.20

0.30

0.40

0.50

0.60

Average Winter Spring Summer Fall

Fra

ction N

itra

te L

oss

Roc Factor

1.00

1.10

1.20

1.30

1.40

1.50

1.60

1.70

1.80

1.90

Average Winter Spring Summer Fall

Ro

c F

acto

r

PM2.5 = a1*1.375*SO4 +a2*1.29*NO3+a3*OC+a4*Other

Center City

Suburban

IMPROVE

Comparison of regression and ratio

method for estimating Roc

Comparison of Regression and Ratio Roc Derivation

0.00

0.50

1.00

1.50

2.00

2.50

3.00

Average W inter Spring Summer Fall

Ro

c

Center City (Reg)

Suburban (Reg)

IMPROVE (Reg)

Center City (Ratio)

Suburban (Ratio)

IMPROVE (Ratio)

• Other= (NH4)2SO4+NH4NO3+Soil+LAC+SS

• POM = FM-Other

• Roc=(FM-Other)/OC

SUM of Bias IMPROVE

-0.40

-0.20

0.00

0.20

0.40

0.60

Average Winter Spring Summer FallConcen

tration

(ug/m

3)

Bias (Center City)

-2.00

-1.50

-1.00

-0.50

0.00

0.50

1.00

1.50

2.00

Average Winter Spring Summer Fall

Co

nc

en

tra

tio

n (

ug

/m3

)

Bias (Suburban)

-2.00

-1.50

-1.00

-0.50

0.00

0.50

1.00

1.50

2.00

Average Winter Spring Summer FallCo

nc

en

tra

tio

n (

ug

/m3

)PM2.5-RPM2.5= ))))RFMRFMRFMRFM(FM(FM(FM(FM iiiiiiii iiii −∑

= (PM2.5SO4-1.375*SO4)+( PM2.5NO3-1.29*NO3)+

(PM2.5POM-1.8*OC)+(PM2.5other-Other)

FMNO3-1.29*NO3

FMPOM-1.8*OC

FMSO4-1.375*SO4

FMother-Other

SUM

Assumptions for estimating gravimetric

and reconstructed mass bias sans

accounting for SVOC loses

Assumptions for estimating gravimetric

and reconstructed mass bias sans

accounting for SVOC loses• Weighing of Teflon filter includes retained water on sulfate and nitrate

aerosols

• Nitrate and sulfates are measured without loss on nylon substrate

• Organic volatilization from Teflon and quartz substrates are the same

• Gravimetric mass of total carbon, soil dust, and sea salt determined

without bias

Gravimetric:

PM2.5 Bias = (a1 -1)*1.375 *SO4

+(a2-1)1.29*NO3Reconstructed:

RPM2.5 Bias=(1.8-a3)*OC+(1-a4)*Other

Average bias in gravimetric and reconstructed

PM2.5 mass concentration for the IMPROVE

and CSN datasets

FM Bias on Teflon filter

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

Winter Spring Summer Fall AVG

Concentration (ug/m

3)

Reconstructed Mass Bias

-0.50

0.00

0.50

1.00

1.50

2.00

Winter Spring Summer Fall AverageConcentration (ug/m

3)

IMPROVE

Suburban

City Center

Seasonal and spatial variability in fine gravimetric mass bias for the CSN monitoring network

PM2.5measured-PM2.5ambient

Seasonal and spatial variability in fine gravimetric mass bias for the IMPROVE monitoring network

PM2.5measured-PM2.5ambient

Seasonal and spatial variability in reconstructed mass bias for the CSN monitoring network.

PM2.5measured-PM2.5reconstructed

Seasonal and spatial variability in reconstructed mass bias for the IMPROVE monitoring network

PM2.5measured-PM2.5reconstructed

SUMMARY

• Approximately 20% of OC is volatilized in the IMPROVE sampling system

• The Roc factor (POM/OC) has a seasonal dependence that varies from about 1.4 in the winter to 1.6-1.8 in the summer

• Urban/suburban Roc factors may be marginally higher than rural factors

• Nitrate volatilization from the Teflon substrates vary from about 10% in the winter to 35-40% in the summer

• There doesn’t appear to be a systematic difference in nitrate volatilization between CSN and IMPROVE monitoring sites

• About 20% of inorganics species mass is retained water

SUMMARY (Cont)

• Bias between PM2.5 gravimetric and reconstructed mass is lowest (negative) in winter and highest (positive) in summer

• This bias is primarily driven by assuming a constant Roc factor and retained water by inorganic species

• Gravimetric PM2.5 in CSN network is biased high by one to two ug/m3 in most the east and during the summer biased low by about two ug/m3 in southern California

• IMPROVE reconstructed mass is biased high during all seasons but highest in winter. During summer it is biased low in the desert southwest

END

0

5

10

15

20

25

30

35

0 5 10 15 20 25 30 35

IMPROVE TC, ug/m3

CS

N-S

TN

TC

, ug/m

3

MetOne

Anderson

R&P

URG0

5

10

15

20

0 5 10 15 20

CSN TC, ug/m3

IMP

RO

VE

TC

+, ug/m

3

MetOne 6.7 lpm

Anderson 7.3 lpm

R&P 16.7 lpm

URG 16.7 lpm

2005

42

0.1

1

10

0.1 1 10

routine MetOne EC, ug/m3

collo

cate

d M

etO

ne E

C, ug/m

3 2003-4

2005

2006

Considerably

more scatter is

observed in

routine CSN

measurements by

collocated

MetOne

samplers,

particularly in the

earlier years.

Data are from Bakersfield,* Boston,* Cleveland,* New Brunswick* and

Rubidoux. * Not collocated with IMPROVE

43

EXPECTATION:

[TC]CSN = [EC]IMP + (1+BIMP*)[OC]IMP + ACSN/VMetOne

OLS REGRESSION:

[TC]CSN = (1+bEC)[EC]IMP + (1+bOC)[OC]IMP + a1 +…+ a12 + e

2005-6 observations at 7 MetOne sites (excluding Phoenix):

bEC = 0.008 (+/-0.05)

no sampling artifact for IMPROVE EC

bOC = 0.22 (+/-0.03)

~ 20% sampling loss for IMPROVE OC

rms(e) = 0.9 ug/m3 (r2 = 0.94, n = 728)

amm � next slide

44

EC – the short story:

-

-

-

- EC has little to no positive artifact for both

IMPROVE and CSN

-Multiplicative bias between CSN and IMPROVE

that is sampler independent

0,IMPε ≅ _IMP artifact adjε =

,newIMP CSNα≅ 1α >

,CSN CSNφ ϕ≅ φ ϕ≠ samplers

45

TC – the short story:

_IMP artifact adjε =

( ) ,newIMP CSNλ θ≅ − 1,λ <

,CSN CSNφ ϕ≠ φ ϕ≠ samplers

0θ >

Relating IMPROVE and CSN ECMaking CSN look like IMPROVE

• Recall - EC has little to no positive artifact for both IMPROVE

and CSN

- CSN and IMPROVE have a multiplicative bias that is sampler independent

• ECIMP = α * ECCSN

• α ~ 1.3

Non-urban and urban PM2.5

networks

Non-urban and urban PM2.5

networks

Interagency Monitoring of

PROtected Visual Environments

(IMPROVE) Network

Interagency Monitoring of

PROtected Visual Environments

(IMPROVE) Network

Chemical Speciation Network (CSN)

Chemical Speciation Network (CSN)

Virgin IslandsVirgin Islands

Ammonium Sulfate

Ammonium Nitrate

Organic Carbon

Fine Soil