Summary of Reprocessing 2016 IMPROVE Data with New Integration Threshold

Transcription

1 Summary of Reprocessing 216 IMPROVE Data with New Integration Threshold Prepared by Xiaoliang Wang Steven B. Gronstal Dana L. Trimble Judith C. Chow John G. Watson Desert Research Institute Reno, NV Prepared for IMPROVE Steering Committee January 3, 218

2 1. Background UC Davis (217) reported that the 1% percentile and median elemental carbon (EC) values for January-June 216 were lower than those of (Figure 1). The mass absorption efficiency (fabs/ec), defined as the ratio of HIPS-derived filter optical absorption coefficients (fabs in Mm -1 ) to IMPROVE_A EC concentration, appears to have increased during the first half of 216 (Figure 2). Differences between the first half of 216 and prior full years were larger for the 1% and % percentiles than for the 9% percentiles. Organic carbon (OC) percentile values for 216 were similar to those of prior years (Figure 3). OC, EC, TC, PM2., and fabs over measured in the IMPROVE network were downloaded from the Federal Land Manager Environmental Database (CIRA, 217). Figures 4 and show monthly 1%, %, 9%, and average of PM2. concentrations and fabs, respectively. Both 216 PM2. and fabs had the lowest values among , particularly from January to September. Lower PM2. concentrations would partially contribute to lower EC and OC in Figures 1 and 3. Since January 216, carbon analysis of the IMPROVE and Chemical Speciation Network (CSN) were completed with the DRI Model 21 multiwavelength carbon analyzers, while the DRI Model 21 carbon analyzers were used in earlier years. There are two main design differences between Model 21 and Model 21: 1) Model 21 uses a single laser (633 nm) while Model 21 uses seven lasers (i.e., 4, 44, 32, 63, 78, 88, and 98 nm) for optical monitoring; and 2) Model 21 uses a flame ionization detector (FID) for carbon detection, while the Model 21 uses a nondispersive infrared (NDIR) CO2 detector. The carbon integration signal for the Model 21 is calculated as the product of the CO2 concentration (ppm) and NDIR flow rate (2 ml/min). This report: Re-examines the IMPROVE comparison data between Model 21 and Model 21; Evaluates the Model 21 signal integration threshold in relation to the Model 21 threshold; Compares 216 IMPROVE data with different Model 21 signal integration thresholds. 2. Re-examine the IMPROVE Comparison Data between Model 21 and Model 21 Prior to replacement of the aging Model 21 with the Model 21, ~1, IMPROVE samples from the IMPROVE network (October 214 to December 21) were analyzed by both models to verify their comparability. Figure 6 compares the OC, EC, TC, and carbon fraction concentrations by the two instruments with Model 21 NDIR integration threshold set to 2. During initial examination of equivalence, the linear regressions were forced through zero. The slopes of linear regressions forced through the origin for OCR, OCT, ECR, ECT, and TC were within.9 1. with coefficients of determination (R 2 ).94. All carbon fractions except OC1 and EC3 had slopes within ±1% from unity. Based on these results, it was concluded that the Model 21 carbon data were equivalent to the Model 21 and the IMPROVE Steering Committee approved replacement of the aging units. 1

3 3. Examine the appropriateness of Model 21 signal integration threshold During the Model 21 analyzer development, different NDIR integration thresholds were compared to reduce effects of NDIR noise and drift. After analysis of blanks, calibration standards, as well as IMPROVE and Fresno samples, a threshold value of 2 (ppm ml/min) was chosen. However, this integration threshold was not compared to the carbon-equivalent threshold of the FID on the Model 21 as the difference was believed to be negligible. Figure 7 compares Model 21 thermograms of sucrose calibration solutions containing 1 and 18 µg carbon (per punch), corresponding to.22 and 4. µgc/m 3 for IMPROVE samples. For the 1 µg carbon case (Figure 7a), with a threshold of 2, all carbon fractions except OC3 were set to zero, truncating the low-level carbon signals of OC1, OC2, and OC4. For the 18 µg carbon case (Figure 7b), most carbon signals were above 2. Therefore, a threshold of 2 did not result in a significant truncation of the carbon integration signal. The nonzero carbon signal at the EC3 stage, is likely an artifact since sucrose should not produce EC3. The Model 21 used an FID threshold value of 1 to remove noise. This threshold translates to.72 (ppm ml/min) for an equivalent amount of carbon for the Model 21. Figure 7 also shows that a threshold value of.72 reduces truncation while keeping noise level low. To maintain threshold consistency between the Models 21 and 21, the IMPROVE data for year 216 was reprocessed with a threshold value of.72. For signals measured at sample oven temperatures above 8 C, a higher threshold value of 2. was used because more noise occurs at higher temperatures. 4. Comparison of Model 21 Signal Integration Threshold.72 vs. 2. Figure 8 compares all 216 IMPROVE samples for carbon using thresholds of.72 (new) and 2. (original). Linear regressions with intercept set to zero and nonzero are both shown. While the slopes differ <±% from unity and coefficients of determination (R 2 ) are close to 1 for most carbon fractions, there are positive intercepts when the regressions are not forced through the origin. To better illustrate the effect of the lower threshold at lower concentrations, Figure 9 compares concentrations from a subset of samples with TC <~1 µg/filter (9% of total samples). The higher carbon values with the new threshold of.72 are evident from the >1 slopes and positive intercept for most carbon fractions. Most of these values are below the standard deviation of the field blanks, however, so the absolute change is insignificant. Table 1 compares the 1%, %, and 9% percentile carbon concentrations (after subtraction with monthly median blank levels) as well as annual averages of the 216 IMPROVE data with thresholds of.72 and 2.. Lowering threshold from 2. to.72 causes TC concentrations to increase by.36 µg/m 3 (32%) for 1% percentile,.94 µg/m 3 (16%) for % percentile,.117 µg/m 3 (6%) for 9% percentile, and.8 µg/m 3 (8%) for the annual average. Lowering the threshold causes EC concentrations to increase by.1 µg/m 3 (1%) for 1% percentile,.31 µg/m 3 (9%) for % percentile,.38 µg/m 3 (13%) for 9% percentile, and.26 µg/m 3 (2%) for the annual average. Table 2 lists the concentrations of blank samples collected in 216, analyzed with a threshold of.72. The sample concentration increases by lowering the threshold (i.e., the y-x column in Table 1) are 24 6% of average blank levels and are lower than the lower quantifiable limits (LQL; three times the field blank standard deviations) for OC and TC. The average EC concentration increase is comparable to the LQL for EC. 2

4 . Evaluation of the IMPROVE Data OC, EC, and TC over measured in the IMPROVE network were downloaded from the Federal Land Manager Environmental Database (CIRA, 217). The monthly average EC, OC, TC, and mass absorption efficiency (fabs/ec) are plotted in Figures 1-13, respectively. The 216 percentiles from the original submission with threshold value of 2. (216-T2) and reprocessed data with threshold value of.72 (216-T.72) are also plotted. For the 216 data, a constant sampling volume of m 3 was assumed, and the monthly median blank concentrations were subtracted. Figure 1 shows that EC percentiles from 216 with threshold 2. are lower than earlier years, and lowering threshold to.72 increases the EC concentration to be within the range of earlier years. Lowering threshold also increases OC and TC, but the relative changes are less than those of EC due to their higher concentrations. The mass absorption efficiency fabs/ec percentiles also have better agreement with the threshold of.72 than 2. (see also Figure 2). The effect of changing the integration threshold makes frequency distributions of values more consistent with those from prior years. It has little effect on the absolute concentration, however, as the EC concentration differences before and after re-processing are within the reported uncertainty intervals. 6. References: CIRA, (217). Federal Land Manager Environmental Database (FED). Colorado State University, Fort Collins, CO. UCDavis, (217). Interagency Monitoring of Protected Visual Environments (IMPROVE): Semiannual Quality Assurance Report. Air Quality Group, University of California, Davis. 3

5 Figure 1. Comparison of EC concentrations in the IMPROVE network from 211 to 216 at 9%, % (median), and 1% percentiles (UCDavis, 217). Figure 2. Comparison of the ratio of optical absorption coefficients (f Abs measured by the hybrid integrating plate/sphere, in Mm -1 ) and EC concentrations by thermal optical reflectance (in µg/m 3 ) in the IMPROVE network from 211 to 216 (UCDavis, 217). 4

6 Figure 3. Comparison of OC concentrations in the IMPROVE network from 211 to 216 at 9%, % (median), and 1% percentiles (UCDavis, 217).

7 a) b) PM 2. (ng/m 3 ) PM 2., 1% Percentile PM 2. (ng/m 3 ) PM 2., % Percentile c) d) PM 2. (ng/m 3 ) PM 2., 9% Percentile Figure 4. Comparison of PM 2. concentrations in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year (CIRA, 217). PM 2. (ng/m 3 ) PM 2., Average

8 a) b) 7 f Abs (Mm -1 ) f Abs, 1% Percentile c) d) f Abs (Mm -1 ) f Abs, % Percentile f Abs, 9% Percentile f Abs, Average f Abs (Mm -1 ) Figure. Comparison of optical absorption coefficients (f Abs ) in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year (CIRA, 217). f Abs (Mm -1 ) 2.

9 8 a) OCR b) OCT c) ECR d) ECT OCR Model 21 (ng/m 3 ) y =.98x 6.81 R² =.98 y =.9x R² = OCT Model 21 (ng/m 3 ) y =.99x R² =.98 y =.96x R² = OC4 Model 21 (ng/m 3 ) EC1 Model 21 (ng/m 3 ) EC2 Model 21 (ng/m 3 ) EC3 Model 21 (ng/m 3 ) Figure 6. Comparison of Model 21 vs. 21 analysis for 17 IMPROVE samples collected from October 214 to December 21. The Model 21 NDIR signal integration threshold was set to 2 (ppm ml/min). ECR Model 21 (ng/m 3 ) y = 1.11x R² =.9 y = 1.x R² = ECT Model 21 (ng/m 3 ) y = 1.6x 1.29 R² =.9 y = 1.x R² = OCR Model 21 (ng/m 3 ) OCT Model 21 (ng/m 3 ) ECR Model 21 (ng/m 3 ) ECT Model 21 (ng/m 3 ) e) TC f) OC1 g) OC2 h) OC3 TC Model 21 (ng/m 3 ) y =.99x 88.6 R² =.98 y =.96x R² = OC1 Model 21 (ng/m 3 ) y =.6x 1.3 R² =.93 y =.64x R² = OC2 Model 21 (ng/m 3 ) y = 1.4x R² =.9 y =.99x R² = OC3 Model 21 (ng/m 3 ) y =.99x R² =.9 y =.9x R² = TC Model 21 (ng/m 3 ) OC1 Model 21 (ng/m 3 ) OC2 Model 21 (ng/m 3 ) OC3 Model 21 (ng/m 3 ) i) OC4 j) EC1 k) EC2 l) EC3 OC4 Model 21 (ng/m 3 ) y = 1.12x R² =.87 y = 1.6x R² = EC1 Model 21 (ng/m 3 ) y =.9x R² =.97 y =.91x R² = EC2 Model 21 (ng/m 3 ) y =.96x R² =.7 y = 1.2x R² = EC3 Model 21 (ng/m 3 ) y = 1.9x +.49 R² =.2 y = 1.72x R² =

10 m) OPR n) OPT OPR Model 21 (ng/m 3 ) y =.79x R² =.86 y =.8x R² = Figure 6 continued. OPR Model 21 (ng/m 3 ) OPT Model 21 (ng/m 3 ) y =.86x R² =.9 y =.91x R² = OPT Model 21 (ng/m 3 ) 9

11 a) Thermogram of 1 µg carbon (1 µl of 1 ppm sucrose calibration solution) Carbon Signal (ppm x ml/min) OC1 Carbon OC2 OC3 OC4 Integraton Threshold=2 EC1 EC2 Temperature Analysis Time (s) EC3 Integraton Threshold= Oven Temperature ( C) b) Thermogram of 18 µg carbon (1 µl of 18 ppm sucrose calibration solution) Carbon Signal (ppm x ml/min) OC1 OC2 OC3 Carbon Integraton Threshold= Analysis Time (s) Figure 7. Model 21 thermograms of sucrose solutions containing a) 1 and b) 18 µg carbon with left y-axis (carbon signal) zoomed in to 1 ppm ml/min. The horizontal dash lines indicate carbon integration threshold of 2 and.72 ppm ml/min. EC2 EC3 OC4 EC1 Temperature Integraton Threshold= Oven Temperature ( C) 1

12 OCR New (µg/filter) y =.99x R² = 1. y =.992x R² = 1. OCT New (µg/filter) y =.989x R² = 1. y =.993x R² = 1. ECR New (µg/filter) y =.99x R² =.99 y = 1.33x R² =.987 ECT New (µg/filter) y =.99x R² =.996 y = 1.4x R² = OCR Original (µg/filter) OCT Original (µg/filter) ECR Original (µg/filter) ECT Original (µg/filter) 11 TC New (µg/filter) y =.99x R² = 1. y =.994x R² = OPR New (µg/filter) y =.989x +.3 R² = 1. y =.991x R² = OPT New (µg/filter) y =.987x +.63 R² = 1. y =.992x R² = OC1 New (µg/filter) y =.987x R² =.999 y =.99x R² = TC Original (µg/filter) OPR Original (µg/filter) OPT Original (µg/filter) OC1 Original (µg/filter) OC2 New (µg/filter) 1 1 y =.99x R² = 1. y =.991x R² = 1. OC3 New (µg/filter) 1 1 y =.984x +.96 R² =.998 y = 1.9x R² =.996 OC4 New (µg/filter) y =.968x R² =.99 y = 1.31x R² =.977 EC1 New (µg/filter) y =.989x R² = 1. y =.99x R² = OC2 Original (µg/filter) OC3 Original (µg/filter) OC4 Original (µg/filter) EC1 Original (µg/filter) EC2 New (µg/filter) y = 1.2x R² =.98 y = 1.21x R² =.89 EC3 New (µg/filter) y =.98x. R² = 1. y =.98x R² = 1. Figure 8. Comparison of all 216 IMPROVE samples with threshold.72 vs. 2. for OC, EC, TC, and carbon fractions. Linear regressions with intercept set to zero (red) or nonzero (black) are both shown EC2 Original (µg/filter) EC3 Original (µg/filter)

13 y = 1.11x R² =.982 y =.971x R² = y = 1.31x R² = y = 1.2x R² = y = 1.6x R² = y = 1.29x R² = y =.976x +. R² = 1..6 y =.976x R² = EC2 Original (µg/filter) y = 1.87x +.71 R² = y = 1.38x R² = y = 1.2x +.91 R² = y = 1.133x R² = OC1 Original (µg/filter) 1 y = 1.1x +.71 R² = y = 1.7x R² = EC1 Original (µg/filter) Figure 9. Comparison of 216 IMPROVE samples with TC<1 µg/filter using threshold.72 vs. 2. for OC, EC, TC, and carbon fractions. Linear regressions with intercept set to zero (red) or nonzero (black) are both shown.. y = 1.42x R² = ECT Original (µg/filter) 2 OC4 Original (µg/filter) 1. EC3 New (µg/filter) 4 2 OC3 Original (µg/filter) y = 1.17x R² =.9 6 OC2 Original (µg/filter) 3 4 y = 1.26x +.87 R² = OPT Original (µg/filter) y =.961x R² = OC3 New (µg/filter) OPR Original (µg/filter) y = 1.23x R² =.99 2 OC4 New (µg/filter) 12 1 y = 1.19x R² =.92 4 ECR Original (µg/filter) y = 1.94x +.8 R² =.93 OPT New (µg/filter) OPR New (µg/filter) TC New (µg/filter) 6 2 TC Original (µg/filter) OC2 New (µg/filter) 4 3 y = 1.39x R² =.99 EC2 New (µg/filter) 2 OCT Original (µg/filter) 1 2 OCR Original (µg/filter) 1 y = 1.139x R² =.93 4 OC1 New (µg/filter) 4 y = 1.9x R² = y = 1.4x +.78 R² =.977 EC1 New (µg/filter) 6 ECR New (µg/filter) 8 6 y = 1.26x R² =.99 1 ECT New (µg/filter) 12 y = 1.3x R² =.99 y = 1.98x R² =.98 1 OCT New (µg/filter) OCR New (µg/filter) EC3 Original (µg/filter) 1

14 13 Table 1. Comparison of 216 IMPROVE sample carbon concentrations (µg/m 3 ) using threshold 2. (x) and.72 (y). Fractions 1% percentile % percentile 9% percentile average x y y-x y/x x y y-x y/x x y y-x y/x x y y-x y/x OC1... NA..1.1 NA OC OC OC EC EC NA EC3... NA... NA... NA OPR OC EC..1.1 NA TC Table IMPROVE field blank carbon concentrations (µg/m 3 ) using threshold.72. 1% % 9% Fraction Average LQL Percentile Percentile Percentile * OC OC OC OC EC EC EC OPR OC EC TC * LQL is defined as three times the field blank standard deviation.

15 a) b) EC, 1% Percentile EC (ng/m 3 ) T T EC (ng/m 3 ) Series6 216-T.72 EC, % Percentile c) d) EC, 9% Percentile 3 EC, Average 2 EC (ng/m 3 ) T2 216-T T2 216-T Figure 1. Comparison of EC concentrations in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year. Both threshold values of 2. (216-T2) and.72 (216-T.72) are plotted for the 216 data. EC (ng/m 3 ) 2 1

16 a) b) OC, 1% Percentile OC (ng/m 3 ) T T.72 2 OC (ng/m 3 ) T2 216-T.72 OC, % Percentile c) d) OC, 9% Percentile OC, Average OC (ng/m 3 ) T T OC (ng/m 3 ) T T Figure 11. Comparison of OC concentrations in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year. Both threshold values of 2. (216-T2) and.72 (216-T.72) are plotted for the 216 data.

17 a) b) TC, 1% Percentile TC (ng/m 3 ) T T.72 2 TC (ng/m 3 ) T2 216-T.72 TC, % Percentile c) d) TC, 9% Percentile TC (ng/m 3 ) T2 216-T TC (ng/m 3 ) T2 216-T.72 TC, Average Figure 12. Comparison of TC concentrations in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year. Both threshold values of 2. (216-T2) and.72 (216-T.72) are plotted for the 216 data.

18 17 a) b) f Abs /EC (m 2 /g) f Abs /EC, 1% Percentile T2 216-T c) d) f Abs /EC (m 2 /g) f Abs /EC, 9% Percentile T2 216-T T2 216-T Figure 13. Comparison of mass absorption efficiency (fabs/ec) in the IMPROVE network from 211 to 216 at 1%, % (median), and 9% percentiles as well as monthly average in each year. Both threshold values of 2. (216-T2) and.72 (216-T.72) are plotted for the 216 data. f Abs /EC (m 2 /g) f Abs /EC (m 2 /g) f Abs /EC, % Percentile f Abs /EC, Average T2 216-T