Supplement of Atmos. Meas. Tech., 9, 441–454, 2016 http://www.atmos-meas-tech.net/9/441/2016/ doi:10.5194/amt-9-441-2016-supplement © Author(s) 2016. CC Attribution 3.0 License.
Supplement of Predicting ambient aerosol thermal–optical reflectance (TOR) measurements from infrared spectra: extending the predictions to different years and different sites M. Reggente et al. Correspondence to: S. Takahama (
[email protected]) and A. M. Dillner (
[email protected])
The copyright of individual parts of the supplement might differ from the CC-BY 3.0 licence.
S1.
FT-IR OC
300
Bias = 0.28 µg m3 Error = 0.43 µg m3 Norm. Error = 25 % R2 = 0.79
0
100
FT−IR OC (µg)
500
Test 2013 Addl (only Korea & Fresno)
0
100
300 500 TOR OC (µg)
Figure S1. Scatterplot and performance metrics between FT-IR OC and TOR OC of the Korea and Fresno sites (Site ID 10 and 11 respectively) calibrated with the Calibration 2011 set. Concentration units of µg m-3 for bias and error are based on the IMPROVE nominal volume of 32.8 µg m-3.
FT-IR OC, Calibration 2011 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Sac and Fox
St. Marks
Trapper Creek
6
Absolute Error (μgm−3)
4
2
0 6
4
2
0 0
2500 5000 7500 100000
2500 5000 7500 100000 2 DM
2500 5000 7500 100000
2500 5000 7500 10000
Figure S2. FT-IR OC, Calibration 2011. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4.
FT-IR OC, Test 2011 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Sac and Fox
St. Marks
Trapper Creek
6
Absolute Error (μgm−3)
4
2
0 6
4
2
0 0
2500 5000 7500 100000
2500 5000 7500 100000 2 DM
2500 5000 7500 100000
2500 5000 7500 10000
Figure S3. FT-IR OC, Test 2011. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4. The sample from Sac and Fox site that lies in the TP sector is the sample that we have considered as outlier with respect to the other sample in the same site, and therefore not used to calculate the values of the classification boundaries. FT-IR OC, Test 2013 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Proctor Maple R. F. B
St. Marks
Trapper Creek
6
Absolute Error (μgm−3)
4
2
0 6
4
2
0 0
2500 5000 7500 100000
2500 5000 7500 100000 2 DM
2500 5000 7500 100000
2500 5000 7500 10000
Figure S4. FT-IR OC, Test 2013. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset)
against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4. FT-IR OC, Test 2013 Addl Birmingham
Bliss SP
Cape Romain NWR
Fresno
Gr. Smoky Mt. NP
Hoover
Okefenokee NWR
Puget Sound
South Korea
Tallgrass
Yosemite
Yosemite B
6
4
2
Absolute Error (μgm−3)
0 6
4
2
0 6
4
2
0 0
2500 5000 7500 100000
2500 5000 7500 100000 2 DM
2500 5000 7500 100000
2500 5000 7500 10000
Figure S5. FT-IR OC, Test 2013 Addl. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4.
Calibration 2011 −1
0
1
2
3 20
−2
3
−5
0
5
10
15
Comp.1
1.0
−2
−1
0
1
2
Comp.2
−2.5
−1.5
−0.5
0.5
Comp.3
−5
0
5
10
15
20
−2.5
−1.5
−0.5 0.0
0.5
1.0
Figure S6. OC, Scatterplot matrix between the first three scores of the Calibration 2011 dataset. The green ellipses are the contours (one to three standard deviations) of the bivariate normal distribution between the scores’ combinations.
S2.
FT-IR EC Bias = 0.05 μg m3 Error = 0.10 μg m3 Norm. Error = 22 % R2 = 0.85
Bias = 0.13 μg m3 Error = 0.17 μg m3 Norm. Error = 33 % R2 = 0.60
40
60
80
Test 2013 Addl (only Korea)
0
20
FT−IR EC (μg)
100
Test 2013 Addl (only Fresno)
0
20
40
60
80
100 0 20 TOR EC (μg)
40
60
80
100
Figure S7. Scatterplot and performance metrics between FT-IR EC and TOR EC of the Korea and Fresno sites (Site ID 10 and 11 respectively) calibrated with the Calibration 2011 set. Concentration units of µg m-3 for bias and error are based on the IMPROVE nominal volume of 32.8 µg m-3. FT-IR EC, Calibration 2011 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Sac and Fox
St. Marks
Trapper Creek
1.2
Absolute Error (μgm−3)
0.8
0.4
0.0 1.2
0.8
0.4
0.0 0
1000
2000
3000
0
1000
2000
3000
2
0
1000
2000
3000
0
1000
2000
3000
DM
Figure S8. FT-IR OC, Calibration 2011. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4.
FT-IR EC, Test 2011 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Sac and Fox
St. Marks
Trapper Creek
1.2
Absolute Error (μgm−3)
0.8
0.4
0.0 1.2
0.8
0.4
0.0 0
1000
2000
3000
0
1000
2000
3000
2
0
1000
2000
3000
0
1000
2000
3000
DM
Figure S9. FT-IR OC, Test 2011. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4. FT-IR EC, Test 2013 Mesa Verde
Olympic
Phoenix
Phoenix B
Proctor Maple R. F
Proctor Maple R. F. B
St. Marks
Trapper Creek
1.2
Absolute Error (μgm−3)
0.8
0.4
0.0 1.2
0.8
0.4
0.0 0
1000
2000
3000
0
1000
2000
3000
0 2
1000
2000
3000
0
1000
2000
3000
DM
Figure S10. FT-IR OC, Test 2013. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4.
FT-IR EC, Test 2013 Extra Birmingham
Bliss SP
Cape Romain NWR
Fresno
Gr. Smoky Mt. NP
Hoover
Okefenokee NWR
Puget Sound
South Korea
Tallgrass
Yosemite
Yosemite B
1.2
0.8
0.4
Absolute Error (μgm−3)
0.0 1.2
0.8
0.4
0.0 1.2
0.8
0.4
0.0 0
1000
2000
3000
0
1000
2000
3000
2
0
1000
2000
3000
0
1000
2000
3000
DM
Figure S11. FT-IR OC, Test 2013 Addl. Anticipation of the prediction error. Mahalanobis distances per sample and site (between the scores of each dataset described in Section 2.1 and the centroid of the calibration dataset) against absolute errors (between TOR OC and FT-IR OC). Blue dashed lines represents the boundaries used for classification as explained in Section 2.4.