A new approach to eliminate the broadband absorption in DOAS spectra M. Ródenas 1, E. Soria 2, J.D....
-
Upload
abigayle-crawford -
Category
Documents
-
view
214 -
download
0
Transcript of A new approach to eliminate the broadband absorption in DOAS spectra M. Ródenas 1, E. Soria 2, J.D....
A new approach to eliminate the A new approach to eliminate the broadband absorption inbroadband absorption in
DOAS spectraDOAS spectra
M. Ródenas1, E. Soria2, J.D. Martín2
1Fundación CEAM, EUPHORE Labs, 46980 Paterna, Valencia, Spain 2Dpto. Ingeniería Electrónica. Univ. de Valencia, Burjassot-Valencia, Spain
Speaker: Mila Ródenas
Fundación CEAM: EUPHOREFundación CEAM: EUPHOREFundación CEAM: EUPHOREFundación CEAM: EUPHORE
EUPHORE is one of the major research platforms in Europe with outstanding analytical infrastructure to investigate chemical
processes related to tropospheric chemistry.
Measurement of contaminants Provide information for models Air quality policies
improvement
Special focus on the Mediterranean Basin
Doas scheme
Doas scheme
DOAS Setup White-multireflection cell: 8m base-path
(EUPHORE) Optical Pathlengths: 128-386-640-896-
1154m Resolution: 0.04, 0.08, 0.16 nm/pixel
(FWHM=0.18, 0.35, 0.76 nm, resp.) Several sets of mirrors for detection of
different compounds Range: UV and Visible (200 – 1000 nm) Automatic adjustment of mirrors in the
chamber Automatic positioning of optical fibers Spectrometer (Acton-500 Pro) and
detector (PDA-3904 Hamamatsu) are
thermostated
DOAS setup at EUPHOREDOAS setup at EUPHOREDOAS setup at EUPHOREDOAS setup at EUPHORE
Lambert-Beer Law:Lambert-Beer Law:Lambert-Beer Law:Lambert-Beer Law:
iii LcII ·exp0
Broadband structures in DOAS spectraBroadband structures in DOAS spectraBroadband structures in DOAS spectraBroadband structures in DOAS spectra
Longitud de onda (nm)
Inte
nsid
ad (u
.a.)
Scattering de Mie
Scattering de Rayleigh
I0 ()
OD'
'0
Longitud de onda (nm)
Inte
nsid
ad (u
.a.)
Scattering de Mie
Scattering de Rayleigh
I0 ()
OD'
'0
BroadbandNarrow bands
Origin of broadbandOrigin of broadbandOrigin of broadbandOrigin of broadband
• Mie scattering (aerosols, particles, λ-
1.3)
• Raileigh scattering (molecules, λ-4)
• Low frequency contribution of the compound of interest
• Change of beam position on dielectrical mirrors, etc
• Change of lamp spectrum
• Compounds with broad structure, etc
''ln0
NLccI
IOD MR
jjj
a
jjj
ChoicesChoices
Binomial smoothing
Polynomials
Regression
Moving average
Savitzky-Golay
Fourier, etc
FILTERING of broadband structuresFILTERING of broadband structuresFILTERING of broadband structuresFILTERING of broadband structures
When to apply it?When to apply it?
During the fitting routine
Prior to the fitting routine
Applied methodApplied method
High-pass filtering
Subtraction of low-pass filter
DrawbacksDrawbacksDrawbacksDrawbacks
ANALYST MUST CHOOSE:
• polynomial degree, r
• region of analysis
• use or not pre-filters?
Classic DOAS approachClassic DOAS approachClassic DOAS approachClassic DOAS approach
Lambert-Beer Law j
jjjjr iddSaiPiF ),...)(,('·)()( 1,0,
Linear fitting (Pr, aj)
Non-linear fitting (dj,k)
ANALYSIS ALGORITHM: Classic MethodANALYSIS ALGORITHM: Classic MethodANALYSIS ALGORITHM: Classic MethodANALYSIS ALGORITHM: Classic Method
Reduction of diff. cross section if a pre-filter is used, i.e. INCREASE OF DETECTION LIMIT
Degrees of freedom
No exact rules (Need to check up residuals)
Does the broadband match exactly a polynomial? (structures interfering or higher residuals)
Dependence with the compounds absorption features
ANALYSIS ALGORITHM: Proposed methodANALYSIS ALGORITHM: Proposed methodANALYSIS ALGORITHM: Proposed methodANALYSIS ALGORITHM: Proposed method
Proposed approachProposed approachProposed approachProposed approach
Linear fitting (Pri, aji=aj)
Non-linear fitting (dj,k)
j
jjjjr iddSaiPiF ),...)(,('·)()( 1,0,
Fitting of moving windows of n channels centered in i (Pri, aji) using low r
AdvantagesAdvantagesAdvantagesAdvantages
• Less intervention of analyst
• Low polynomial orders used - width of analysis spectral range - shape of
compounds/broadband
• Better match between modelled and real broadband
• No need of pre-filter, i.e. NO DECREASE OF DETECTION LIMIT
Analysis conditionsAnalysis conditionsAnalysis conditionsAnalysis conditions
Combination of concentrations [0-100ppb] Spectral dispersion not allowed (i.e., linear fitting) Proposed Method: width of windows, n=80 channels
polynomial degree, r=3 Classic implementation with Polynomial degree: [1-9] & binomial
pre-filters Proposed implementation with MATLAB
RESULTS: Test on NORESULTS: Test on NO22 and Glyoxal and GlyoxalRESULTS: Test on NORESULTS: Test on NO22 and Glyoxal and Glyoxal
+ noise
0.0E+00
2.0E-04
4.0E-04
6.0E-04
8.0E-04
1.0E-03
1.2E-03
1.4E-03
1.6E-03
400 420 440 460 480
Wavelength (nm)
O D
(C
o m
po
u n
d s
)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
O D
(B
r o
a d
b a
n d
)
NO2GlyoxalBroadband
c
OD’ of 1 ppb of compound at L=1154m
ANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: Results
NONO22: Concentration error (%): Concentration error (%)NONO22: Concentration error (%): Concentration error (%)
Concentration error %
-40
-20
0
20
40
60
80
100
10 20 30 40 50 60 70 80 90 100 110
Theoretical C [ppb]
G1
G2
G3
G4
G5
G6
G7
G8
G9
G1_500i
G2_500i
G3_500i
G4_500i
G5_500i
G6_500i
G7_500i
G8_500i
G9_500i
No Pre-Filter
Pre-Filter
Concentration error %
0
2
4
6
8
10
12
14
16
18
20
10 20 30 40 50 60 70 80 90 100 110
Theoretical C [ppb]
G1_500i
G2_500i
G3_500i
G4_500i
G5_500i
G6_500i
G7_500i
G8_500i
G9_500i
c
Concentration error %
0
2
4
6
8
10
12
14
16
18
20
10 20 30 40 50 60 70 80 90 100 110
Theoretical C [ppb]
G1_500i
G2_500i
G3_500i
G4_500i
G5_500i
G6_500i
G7_500i
G8_500i
G9_500i
This work
Pre-Filter Proposed Method
Cerror=(CTheo-Ccalc)/CTheo*100
ANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: ResultsANALYSIS ALGORITHM: Results
GLYOXAL: Concentration error (%)GLYOXAL: Concentration error (%)GLYOXAL: Concentration error (%)GLYOXAL: Concentration error (%)
Glyoxal Concentration error (%)
0.0
0.1
1.0
10.0
100.0
1000.0
10000.0
10 20 30 40 50 60 70 80 90 100 110
Theoretical C [ppb]
Glyoxal Concentration error (%)
0.0
0.1
1.0
10.0
100.0
1000.0
10000.0
10 20 30 40 50 60 70 80 90 100 110
Theoretical C [ppb]
Pre-Filter
Proposed Method
No Pre-Filter
Residual Root Mean Square (RMS)
7.58
E-0
4
8.03
E-0
4
0.0E+00
2.0E-03
4.0E-03
6.0E-03
8.0E-03
1.0E-02
1.2E-02
1.4E-02
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filterPre-filter (500 iter)Pre-filter (2000 iter)
Results: Test on NOResults: Test on NO2 2 (355 – 408 nm)(355 – 408 nm)Results: Test on NOResults: Test on NO2 2 (355 – 408 nm)(355 – 408 nm)
NO2
355-408 nm
disp=0.16 nm/ch
Binomial filter:Classic
n=80 channels, r=3:Proposed
NO2
1.1
0
1.0
6
0
20
40
60
80
100
120
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filter
Pre-filter (500 iter)
Pre-filter (2000 iter)
(ppb)
N
CCE calcTheor
2
2/1 )(
NO2
1.10
1.06
0
5
10
15
20
25
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filter
Pre-filter (500 iter)
Pre-filter (2000 iter)
(ppb)
N
CCE calcTheor
2
2/1 )(
NO2
8.64
1.492.
965.
00
0
5
10
15
20
25
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filter
Pre-filter (500 iter)
Pre-filter (2000 iter)
(ppb)
306 – 408 nm
Results: Test on NOResults: Test on NO2 2 (320 – 400 nm)(320 – 400 nm)Results: Test on NOResults: Test on NO2 2 (320 – 400 nm)(320 – 400 nm)
NO2
320-400 nm
disp=0.16 nm/ch
Binomial filter:Classic
n=100 channels, r=3:Proposed
Similar results using the best fitting parameters of the classical method and using the proposed method
Strong dependence on the analysis range used in classic method
DEGREES OF FREEDOM REDUCED IN THE PROPOSED
METHOD
NO2
1.10
1.06
0
5
10
15
20
25
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filter
Pre-filter (500 iter)
Pre-filter (2000 iter)
(ppb)
N
CCE calcTheor
2
2/1 )(
Residual Root Mean Square (RMS)
7.58
E-04
8.03
E-04
0.0E+00
2.0E-03
4.0E-03
6.0E-03
8.0E-03
1.0E-02
1.2E-02
1.4E-02
G1 G2 G3 G4 G5 G6 G7 G8 G9 ProposedMethod
ANALYSIS METHOD
No pre-filterPre-filter (500 iter)Pre-filter (2000 iter)
ANALYSIS ALGORITHM: ConclusionsANALYSIS ALGORITHM: ConclusionsANALYSIS ALGORITHM: ConclusionsANALYSIS ALGORITHM: Conclusions
COMPARATIVE CLASSIC vs COMPARATIVE CLASSIC vs PROPOSEDPROPOSED ALGORITHM ALGORITHMCOMPARATIVE CLASSIC vs COMPARATIVE CLASSIC vs PROPOSEDPROPOSED ALGORITHM ALGORITHM
Features Classic Proposed
Correct results
Highly dependent on: Polynomial degree Pre-fi lter used? Spectral structure of
compounds
Yes
Concentration off set I nitial parameters
dependent Negligible (good
results)
Pre-fi lter Depends on spectra No
Sensitivity with initial parameters
Strong Weak
Wider ranges possible?
? Yes
User intervention High Low
*
MUCH HIGHER USER INTERVENTION NEEDED IN CLASSIC THAN IN PROPOSED METHOD
ANALYSIS ALGORITHM: Future worksANALYSIS ALGORITHM: Future worksANALYSIS ALGORITHM: Future worksANALYSIS ALGORITHM: Future works
Test of the algorithm to analyse other compounds
Use of different polynomials or curves
Pre – processing of spectra to enhance features(e.g.. derivatives)
Thanks for your
attention
ACKNOWLEDGEMENTS: GENERALITAT VALENCIANA, BANCAJA,
EUROCHAMP, AND ACCENT FOR SUPPORT.