Bruno Debus Andy Weakley Satoshi Takahama Ann Dillner Oct 22 th 2019 Petaluma California 1 Acknowledgements Funding for this project EPA and IMPROVE NPS Cooperative Agreement P11AC91045 ID: 784754
Download The PPT/PDF document "FT-IR OC & EC predictions in IMPROVE..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
FT-IR OC & EC predictions in IMPROVE & CSN networks across multiple years
Bruno Debus, Andy Weakley, Satoshi Takahama, Ann Dillner Oct 22th, 2019Petaluma - California
1
Slide2AcknowledgementsFunding for this project:
EPA and IMPROVE (NPS Cooperative Agreement P11AC91045) EPRI (Agreement 10003745 and 10005355)Swiss Polytechnic University-Lausanne (EPFL) Collaborators, post-docs and undergraduate / graduate students: CSN, FRM, IMPROVE, SEARCH programs and site/state personnelJoann Rice, Mike Hays, Emily Li, EPABret Schichtel and Scott Copeland, IMPROVEStephanie Shaw and Eric Edgerton, EPRI/ARARandall Martin and the SPARTAN personnel, SPARTAN and Washington UniversityDave Diner and MAIA team members, MAIA and JPL2Alexandra BorisKelsey SeibertTravis Ruthenburg
Mohammed Kamruzzaman
Charlotte Burki
Amir Yazdani
Brian Trout
Jenny Hand
Katie George
Charity
Coury
Sean Raffuse
Tony Wexler
Slide3Non-destructiveFast and low-costAnalyzing all IMPROVE and CSN PTFE filters
5 min/sample, 3 instruments, 400-700 filters/wk6 undergrads and 1 lab supervisorPM2.5 PTFE/Teflon filtersRoutinely collected No gas phase adsorptionFT-IR spectra are information richTOR OC and ECOrganic functional groups, OMSources Inorganics including SO4, NO3, SiO3FT-IR: Strengths & Limitations for network applicationsCalibration methods are complex PTFE
filter manufacturer (Pall, MTL) dependent
No
directly comparable methods for functional groups and OM to validate data (
no gold standard
)
Slide4FT-IR in routine network measurements
FT-IR spectroscopy
Extract quantitative information about IR active substances
Mass
Carbon
Ions
Elements
Functional groups
IMPROVE
75,505 Teflon filters (2015 – 2018)
CSN
26,936 Teflon filters (2017 – 2018)
4
Teflon filters
Slide5Quantitative analysis of ambient samples using FT-IR
Basic considerations5
Slide6Global model Single calibration using a sample subset
from every siteLow prediction quality for samples collected during wildfires eventsSmoke impacted sample detection?
Sites with unusual composition
Atypical / Typical partitioning?
IMPROVE
From
“Global”
to
“
Mutli
-level”
modeling
CSN
A.T
.
Weakley
et al(2018
) Ambient aerosol composition by infrared spectroscopy and partial least squares in
the chemical
speciation network: Multilevel modeling for elemental carbon, Aerosol Science
and Technology
, 52:6,
642-654
Requires collocated Teflon & Quartz modules at
every location
around the network
Drawbacks & limitations
:
1
2
3
Slide7Wildfire
detection – IMPROVE (2015)Based on a simple OC/EC criterionSeasonality & TOR OC concentrations are consistent with fire season / emissions 7≈ 341 samples
1
Slide8Typical
/ Atypical sites – CSN (2017)Cluster 1: 124 Typical sitesCluster 2: 15 Atypical sitesSite name
State
Adams County
CO
Platteville
CO
Criscuolo Park
CT
Rome - Elementary School
GA
Indianapolis - Washington Park
IN
Elizabeth Lab
NJ
New York - Division Street
NY
Harvard Yard (Cleveland)
OH
Southerly WTP
OH
Akron - 5 Points
OH
Marcus Hook
PA
NE Wastewater Treatment Plant
PA
Jail at Bayamon
PR
Seattle 10th Ave
WA
Charleston NCore
WV
Atypical sites ( n = 15)
Each site is represented as function of it mean OC/EC ratio and prediction error (Global model)
Clustering is used to partition Atypical sites from Typical sites
8
2
Slide9Site selection - Flowchart
9TypicalAtypicalNon FireSite number
Site combination
Optimization
3
Slide102015
2016 2017 2018Site number - Optimization (EC)IMPROVECSN
Slide11Site combination - Optimization (EC)
11Identify site combination with optimal predictions:Examine 3,000 potential site list candidates using a Monte Carlo method
High R
2
& near zero bias
Reliable predictions for
both
OC & EC
Consistent predictions across multiple years
Slide12Optimal IMPROVE site list
1214 % of the network
Slide13Optimal
CSN site lists13≈ 14 % of the network
Slide14Calibration / FT-IR predictions Results
14
Slide15Results – OC & EC prediction (IMPROVE)
R
2
Bias
(%)
Error
(%)
<
MDL
(%)
OC
0.98
-0.3
12.9
4.3
EC
0.92
0.2
25.7
32.7
2015 – 2016 – 2017
–
2018
Satisfactory prediction metrics across a 4 year period
Predictions from both Fire & non fire impacted samples are reported together
15
Slide16Results – OC & EC prediction (CSN)
R2Bias (%)Error (%)< MDL (%)
OC
0.93
-0.8
13.6
0.6
EC
0.75
0.5
25.5
9.9
2017
–
2018
S
atisfactory prediction metrics across a 2 year period
Predictions from both Typical & Atypical sites are reported together
16
Slide17Ions & Elements FT-IR predictions
IMPROVECSNR2 Bias
(%)
Error
(%)
< MDL
(%)
S
0.98
0.2
5.6
1.4
PM
2.5
0.98
0.5
5.7
0.2
SO
4
0.98
0.1
5.8
1.1
NH
4
0.95
1.2
8.8
1.2
Soil
0.98
1.5
9.5
8.9
Si
0.98
2.0
11.2
14.7
Ca
0.97
1.0
11.3
6.9
Al
0.98
0.6
12.3
8.8
OC
0.98
0.8
12.7
0.4
HIPS
0.88
-3.0
22.1
17.7
Fe
0.93
3.1
23.9
16.8
Ti
0.92
0.6
24.8
19.4
EC
0.91
1.6
25.9
15.8
NO
3
(winter north)
0.93
10.7
48.6
25.1
R
2
Bias
(%)
Error
(%)
< MDL
(%)
S
0.94
0.1
9.6
2.7
OC
0.94
-0.7
13.3
0.7
SO
4
0.83
-0.3
14.8
4.2
EC
0.79
0.2
25.2
12.6
Ca
0.82
-0.9
31.1
16.7
Si
0.86
-2.9
41.3
34.6
NO
3
0.88
13.3
45.7
15.2
Ti
0.68
-9.6
59.9
27.1
NH
4
0.84
3.0
66.8
47.7
Al
0.73
-41.7
102.6
68.6
Similar (%) error compared to OC
Similar (%) error compared to EC
Besides carbon, additional IR active materials can be predicted from Teflon filters
(
XRF
, IC
)
This data can be used for QC and the calibrations developed for CSN could be extended to FRM (as previously shown for OC
& EC
)
17
Slide18Conclusions
18Multi-levels models accommodate unique variations in aerosols composition across the networks and improve predictionsIMPROVE Fire / Non fire modelsCSN Atypical / Typical sitesThe number of sites retained in the calibration to maintain accurate predictions and the corresponding site selection was optimized via a Monte Carlo methodIMPROVE 22 sites retained (14 % of the network)CSN 20 sites retained (14 % of the network)The multi-level modeling provides reliable TOR-equivalent OC & EC concentrations across a 4 years of IMPROVE and 2 years for CSN.In addition to carbon, IR active materials sulfate and silicate can be predicted from IR spectra of Teflon filters
Useful for QC for IMPROVE and CSN
CSN calibrations can be used for the FRM network
Slide19Thank you for your attentionPlease send request for
additional plots / analysis to bdebus@ucdavis.edu
Slide2020
Slide21Supporting Materials
21
Slide22FT-IR Lab – UC Davis
Automatic LN2 refilling systemPurge systemPurge system
IR1
IR2
IR3
Slide23Wildfire detection – IMPROVE (2016)
≈ 180 samples
Slide24Wildfire detection – IMPROVE (2017)
≈ 620 samples
Slide25Wildfire detection – IMPROVE (2018)
≈ 560 samples
Slide26Initial site selection strategy
Each site is summarized by it median TOR EC & NH4 concentrationsSite close to the bin center is selected for calibration (representative)The optimal number of site is assessed by varying the number of binsExample of bin segmentation (IMPROVE 2015)26
Slide27Inter-year comparison of the top 30 sites list candidates (
IMPROVE – OC)
Optimum
Slide28Inter-year comparison of the top 30 sites list candidates (
IMPROVE
– EC)
Optimum
Slide29Inter-year comparison of the top 30
Atypical
sites list candidates (
CSN
– OC)
Optimum
Slide30Inter-year comparison of the top 30
Atypical
sites list candidates (
CSN
– EC)
Optimum
Slide31Inter-year comparison of the top 30
T
ypical
sites list candidates (
CSN
– OC)
Optimum
Slide32Inter-year comparison of the top 30
T
ypical
sites list candidates (
CSN
– EC)
Optimum
Slide33Site index
Site nameState02-090-0034Alaska NCoreAK06-067-0006Sacramento - Del Paso ManorCA11-001-0043Washington DC - McMillan ReservoirDC18-163-0021Evansville - Buena Vista RoadIN29-510-0085St. Louis - Blair Street
MO
34-013-0003
Newark Firehouse
NJ
36-031-0003
Whiteface
NY
37-067-0022
Winston-Salem - Hattie Ave
NC
37-119-0041
Garinger
High School
NC
39-113-0038
Sinclair Community College
OH
40-109-1037
OCUSA Campus
OK
42-071-0012
Lancaster Downwind
PA
42-125-5001
East of Pittsburgh- Florence
PA
48-201-1039
Deer Park
TX
53-033-0080
Seattle - Beacon Hill
WA
55-119-8001
Perkinstown CASTNET
WI
Typical sites
Atypical sites
Site index
Site name
State
08-123-0008
Platteville
CO
09-009-0027
Criscuolo
Park
CT
18-097-0078
Indianapolis - Washington Park
IN
34-039-0004
Elizabeth Lab
NJ
IMPROVE
CSN
Optimal site lists – Details
Site index
Site name
State
Affiliation
BAND1
Bandelier
NM
NPS
BIBE1
Big Bend National Park
TX
NPS
CABA1
Casco Bay
ME
STATE
CORI1
Columbia River Gorge
OR
FS
FLTO1
Flat Tops Wilderness
CO
FS
GLAC1
Glacier
MT
NPS
HAVO1
Hawaii Volcanoes
HI
NPS
JARB1
Jarbidge
NV
FS
LASU2
Lake Sugema
IA
STATE
LIGO1
Linville Gorge
NC
FS
LTCC1
Lake Tahoe Community College
CA
STATE
MAVI1
Martha's Vineyard
MA
TRIBE
MONT1
Monture
MT
FS
MOOS1
Moosehorn
ME
FWS
MORA1
Mount Rainier
WA
NPS
OLYM1
Olympic
WA
NPS
RAFA1
San Rafael
CA
FS
SHRO1
Shining Rock
NC
FS
TALL1
Tallgrass
KS
STATE
THSI1
Three Sisters
OR
FS
VIIS1
Virgin Islands
VI
NPS
WHIT1
White Mountain
NM
FS
Slide34IMPROVE 2015 – OC & EC prediction
R2Bias (%)Error
(%)
<
MDL
(%)
OC
0.98
0.8
12.7
0.4
EC
0.91
1.6
25.9
15.8
Slide35IMPROVE 2016 – OC & EC prediction
R2Bias (%)Error (%)< MDL (%)
OC
0.97
-1.7
14.5
1.5
EC
0.91
0.4
29.0
18.8
Slide36IMPROVE 2017 – OC & EC prediction
R2Bias (%)Error (%)< MDL (%)
OC
0.98
-0.8
12.1
1.5
EC
0.91
-0.2
23.8
4.8
Slide37IMPROVE 2018 – OC & EC prediction
R2Bias (%)Error (%)< MDL (%)
OC
0.99
0.6
12.4
0.4
EC
0.92
-0.9
24.0
7.1
Slide38CSN
2017 – OC & EC predictionR2Bias (%)
Error
(%)
<
MDL
(%)
OC
0.94
-0.7
13.3
0.7
EC
0.79
0.2
25.2
12.9
Slide39CSN
2018 – OC & EC predictionR2Bias (%)Error (%)< MDL (%)
OC
0.92
-1.0
14.1
0.4
EC
0.69
1.3
26.0
7.1
Slide40OC bias distribution –
IMPROVE (2015 – 2018)Percentile bias located within TOR uncertainty boundaries
Slide41E
C bias distribution – IMPROVE (2015 – 2018)Percentile bias located within TOR uncertainty boundaries
Slide42OC bias distribution –
CSN (2017 – 2018)Percentile bias located within TOR uncertainty boundaries
Slide43E
C bias distribution – CSN (2017 – 2018)Percentile bias located within TOR uncertainty boundaries
Slide44IMPROVE – Prediction of IR active ions & elements from Teflon filter
Winter North sample only
Slide45Spatial distribution of the 79 sites considered for developing a Winter North Nitrate calibration (IMPROVE)
Slide46CSN – Prediction of IR active ions & elements from Teflon filter