Estimating mean molecular weight, carbon number, and OM∕OC with mid-infrared spectroscopy in organic particulate matter samples from a monitoring network

Organic matter (OM) is a major constituent of fine particulate matter, which contributes significantly to degradation of visibility and radiative forcing, and causes adverse health effects. However, due to its sheer compositional complexity, OM is difficult to characterize in its entirety. Mid-infrared spectroscopy has previously proven useful in the study of OM by providing extensive information about functional group composition with high mass recovery. Herein, we introduce a new method for obtaining additional characteristics such as mean carbon number and molecular weight of these complex organic mixtures using the aliphatic C−H absorbance profile in the mid-infrared spectrum. We apply this technique to spectra acquired non-destructively from Teflon filters used for fine particulate matter quantification at selected sites of the Inter-agency Monitoring of PROtected Visual Environments (IMPROVE) network. Since carbon number and molecular weight are important characteristics used by recent conceptual models to describe evolution in OM composition, this technique can provide semi-quantitative, observational constraints of these variables at the scale of the network. For this task, multivariate statistical models are trained on calibration spectra prepared from atmospherically relevant laboratory standards and are applied to ambient samples. Then, the physical basis linking the absorbance profile of this relatively narrow region in the mid-infrared spectrum to the molecular structure is investigated using a classification approach. The multivariate statistical models predict mean carbon number and molecular weight that are consistent with previous values of organic-mass-to-organic-carbon (OM/OC) ratios estimated for the network using different approaches. The results are also consistent with temporal and spatial variations in these quantities associated with aging processes and different source classes (anthropogenic, biogenic, and burning sources). For instance, the statistical models estimate higher mean carbon number for urban samples and smaller, more fragmented molecules for samples in which substantial aging is anticipated