The Interagency Monitoring of Protected Visual Environments (IMPROVE) network and the Chemical Speciation Network (CSN) are speciation particulate matter monitoring networks in the U.S.  IMPROVE is located primarily in Class 1 visibility areas and CSN is located in urban and suburban areas.  Carbonaceous aerosol, which can be 50% or more of the particulate matter mass, is measured in both networks using thermal optical reflectance (TOR) on quartz filter samples.  We have developed a less expensive, non-destructive method using Fourier transform-infrared spectroscopy (FT-IR) and partial least squares regression to predict TOR organic (OC) and elemental carbon (EC). 

Infrared spectra are obtained from PTFE filters that are routinely collected in parallel with the quartz filters in both monitoring networks.  Using two-thirds of samples collected at seven IMPROVE sites in 2011, FT-IR can reproduce TOR OC and EC with good accuracy and precision on par with collocated TOR measurements.  Using the 2011 calibrations and samples collected at 17 sites in 2013, we show that FT-IR can measure OC and EC in different years and at different sites than the samples used in the calibration.  For two sites (Fresno, CA and South Korea IMPROVE sites) that likely have different aerosol composition than the primarily rural US sites used in the calibration, making separate calibrations with these sites or adding these sites to the 2011 calibration improved the prediction of TOR OC and EC.  Using a sparse calibration method, we show that TOR OC is predicted using a suite of organic functional groups and that EC is predicted using a few organic functional groups and C-C stretch in ring-structured compounds. 

We developed TOR OC and EC calibrations for the CSN network using samples collected at 10 sites during 2013.  Although the aerosol mass is generally higher in urban areas than in rural areas, samplers used in CSN have lower flowrates and larger filters causing the mass interrogated by the FT-IR beam to be on average ~5 times less in CSN than in IMPROVE.  For OC, the predictions results are very similar to IMPROVE and for the EC the relative error is similar to IMPROVE but the R² drops from 0.96 to 0.85.  Using processed FT-IR spectra, we determine that organic carbon functional groups are used to predict TOR OC and that organic and elemental carbon functional groups and a factor related to pyrolyzed carbon are used to predict EC.