Most air monitoring programs are designed to measure particulate mass loading to provide data for PM10 and PM2.5 NAAQS and visibility. Because these sizes of particles can come from many sources, they are not useful for apportioning to one source or another. While the Interagency Monitoring of Protected Visual Environments (IMPROVE) program provides speciated aerosol data that are helpful in source attribution analysis, the averaging periods of samples and sparse location of sites make IMPROVE measurements difficult to use for source attribution without supplemental measurements or modeling tools.

Wotawa and Trainer (2000) found that 74 percent of the variance in the average afternoon carbon monoxide levels could not be attributed to anthropogenic sources during the 1995 Southern Oxidant Study (Chameides and Cowling 1995). Analysis of weather patterns indicated that transport of wildland fire smoke from Canada could explain the elevated carbon monoxide levels. Also, they discovered a statistically significant relationship between the elevated carbon monoxide and ground-level ozone concentrations.

Characterization of organic carbon compounds found within the organic carbon fraction of fine particulate matter coupled with inclusion of gaseous volatile organic compounds (VOCs) holds substantial promise in advancing the science of source apportionment (Watson 1997). The key to the use of chemical mass balance methods is the acquisition of accurate data describing the chemical composition of both particulate matter and VOCs in the ambient air and in emissions from specific sources. Several organic compounds unique to wood smoke have been identified including retene, levoglucosan, thermally altered resin, and polycyclic aromatic hydrocarbons (PAH) compounds. These compounds are present in appreciable amounts and can be used as signatures for source apportionment if special precautions are taken during sampling to minimize losses (Standley and Simoneit 1987). Inclusion of these aerosol and VOC components in the speciation analysis appears worthwhile but would increase monitoring and sample analysis costs.

Source Apportionment Methods

Apportionment of particulate matter mass to the respective contributing sources is done through both mechanistic models (dispersion models) and receptor-oriented techniques that are based on the characteristics of the particles collected at the receptor. The best approach is through the use of both techniques, applied independently, to develop a "weight of evidence" assessment of source contributions of smoke from fire. A third approach is through the use of visual and photographic systems that can document visibility conditions over time or track a plume from its source to the point of impact within a Class I area.

Receptor-Oriented Approaches

Receptor-oriented approaches range from simple signature applications to complex data analysis techniques that are based on the spatial, temporal, and chemical constituents ("fingerprint") of various sources.

Simple signature applications for smoke from fire are based on chemically distinct emissions from fire. For example, methyl chloride (CH3Cl) is a gas emitted during wood combustion that has been used in this manner to identify impacts of both residential woodstove smoke and smoke from prescribed fires (Khalil and others 1983).

• Speciated Rollback Model: a simple hybrid model that uses aerosol data collected at the receptor with emission inventories to estimate source impacts.
• Chemical Mass Balance Model CMB7: infers source contributions based on speciated aerosol samples collected at a monitoring site.

Receptor-oriented methods of particle mass source apportionment have proven successful in a large number of urban studies worldwide. A number of these studies have attempted to apportion wildfire smoke on the basis of a set of aerosol and source emission trace elements and compounds. The experimental design of these studies has limited the ability of receptor models to resolve wildfire smoke from other sources. With improvements in speciation of the organic carbon component of the aerosol, and inclusion of carbon monoxide, methyl chloride, and other endemic signatures, the ability of these techniques to resolve sources and minimize uncertainties will increase. Sensitivity studies are needed to determine which additional components beyond the standard array of trace elements, ions, and carbon fractions would be most beneficial to include in future monitoring programs.

Mechanistic Models

Multiple dispersion models have been used to estimate air quality impacts of single or multiple fires at local and regional scales. Eulerian regional-scale models have been principally used for source apportionment application both to estimate contributions to particulate air quality and regional haze. The suitability of such models for apportionment applications largely depends on the completeness and accuracy of the emission inventory inputs used by the model. Unfortunately, few field validations are available.

The speciated roll-back model (NRC 1993) is a simple hybrid model that uses aerosol data collected at the receptor with emission inventories to estimate source impacts. It is a spatially averaged model that disaggregates major particle components into chemically distinct groups that are contributed by different types of sources. A linear rollback model is based on the assumption that ambient concentrations (C) above background (C_b) are directly proportional to total emissions in the region of interest (E):

C ­ C_b=kE (1)

The proportionality constant, k, is determined over a historical time period when both concentrations C and Cb as well as regional emissions E are known. Once k is determined, new concentration estimates can be derived for other emission levels of interest assuming that meteorological conditions are constant over the same averaging time. Because the anthropogenic components in the particle mass consist almost entirely of sulfates, nitrates, organic carbon, elemental carbon, and crustal material, a maximum contribution from fire can be made based on the assumption that all of the organic carbon or elemental carbon is from primary fire emissions. Various complexities can be added to this model; components can be disaggregated by particle-size fraction (coarse versus fine particles) as well as by chemical composition. Additional distinctions can be made between primary and secondary particles, and nonlinear transformation processes can be approximated to account for atmospheric reactions.

Simple proportional speciated rollback models re-quire data on the chemical composition of airborne particles, knowledge or assumptions regarding secondary particle components, an emission inventory for the important source categories for each particle component and each gaseous precursor, and knowledge or assumptions regarding background concentrations for each component of the aerosol and each gaseous precursor.

The speciated rollback model was applied by the NRC Committee on Haze in National Parks and Wilderness Areas to apportion regional haze in the three large regions of the country (East, Southwest, and Pacific Northwest) by including extinction coefficients to the estimated mass concentrations (NRC 1993). The percentage of anthropogenic light extinction apportioned to forest management burning was estimated at 11 percent in the Northwestern United States on an annual basis assuming that about one-third of the measured organic carbon is of natural origin. The 1985 National Acid Precipitation Assessment Program (NAPAP) inventory was used in this analysis, which also assumed that the elemental carbon and organic carbon fractions of the PM2.5emissions for forest management burning were 6 percent and 60 percent, respectively.

• The limitations of the speciated rollback model are several:
• Deviations from the assumption of spatially homogeneous emissions are likely to occur when air quality is most critical at a single receptor where a single emission source can have an inordinate impact.
• Secondary particle formation is assumed to be linear to changes in precursor emissions.
• Meteorological conditions do not change from year to year.
• Emission inventory errors have a direct, proportional effect on the model estimates.

The model can be applied to any temporal concentration such as annual average, worst 20th percentile, or worst daily average scenarios in any region that meets the constraint on the spatial distribution of emission changes. It is straightforward, necessary input data are available, and the model assumptions are easily understood. It makes use of chemical speciation data collected from the IMPROVE network but cannot apportion contributions made from source classes not included in the inventory.

The chemical mass balance model, CMB7 (Watson 1997; Watson and others 1990), infers source contributions based on speciated aerosol samples collected at a monitoring site. Chemical elements and compounds in ambient aerosol are "matched" to speciated source emission profiles "fingerprints" by using least-squares, linear regression techniques to apportion the aerosol mass. CMB7 has been widely used within the regulatory community to identify and quantify the sources of particles emitted directly to the atmosphere. The model is based on the relationship between characteristics of the airborne particle (c_i), the summation of the product of the ambient mass concentration contributed by all sources (S_j), and the fraction of the characteristic component in the sources fingerprint (f_ij).

c_i= Σ_jS_jf_ij                              (2)

Given detailed information about the chemical speciation of the ambient aerosol and similar information about all of the emission sources impacting the receptor, the CMB7 model can apportion the aerosol mass among the sources if certain assumptions are met. To minimize error, there must be more aerosol components than sources to be included in the least squares linear regression fit. If there are more components measured than sources, then the comparison of model-estimated concentrations of these additional components provides a valuable internal check on model consistency.

The chemical components in the source "finger-print" must be conserved and not altered during atmospheric transport -- a rather large limitation.

Model resolution is typically limited to five or six source types, and separation of two sources with similar emission profiles (for example, prescribed burning and residential woodstove smoke) is difficult if both sources are active at the same time.

Systematic error analysis procedures have been developed for the CMB7 model, and the results have been published in model validation studies (NRC 1993). However, the model cannot apportion secondary aerosols (sulfate and nitrate); it is limited in its ability to apportion all of the mass to specific sources.

The ability of the model to apportion smoke from fire depends on several factors:

• The presence or absence of smoke from other forms of vegetative burning (woodstoves, agricultural burning, open burning, and others).
• The magnitude of the smoke impact at the receptor (for example, well-dispersed smoke that contributes small amounts of aerosol mass is more difficult to distinguish).
• The uncertainty in both the ambient aerosol and the source "fingerprint" components that the model most heavily weights in the regression analysis, typically organic carbon, potassium, and elemental carbon. The greater the uncertainty of these measurements, the less "fitting pressure" they have in influencing the regression solution.
• Inclusion of multiple aerosol components that are as nearly unique to smoke from fires (endemic signatures) as possible. These include organic compounds such as retene and levoglucosan, as well as gaseous signature such as carbon monoxide and methyl chloride. The more the source profile distinguishes prescribed or wildland fire smoke from other sources, the more accurate the source apportionment is likely to be.