Outdoor biomass burning – including wildfires, prescribed fires, and agricultural fires – is a major source of trace gases and aerosols that impact health, air quality, and climate. These health- and climate-relevant primary emissions from biomass burning include species deemed as hazardous air pollutants (HAPs), such as benzene, formaldehyde, and acetaldehyde, which themselves can cause acute health effects (Wentworth et al., 2018). In addition to the pollutants directly emitted by fires, chemistry in smoke plumes can produce ozone (O3), which can negatively impact human health (U.S. EPA, 2013) as well as affect vegetation, water quality, soil, and the ecosystems that they support (European Environmental Agency, 2018). O3 formation can occur due to the emission of nitrogen dioxide (NO2), HONO, and volatile organic compounds (VOCs) and the presence of sunlight (Baylon et al., 2018), with enhanced photolysis rates occurring most predominantly during midday, when photolysis rates are fastest. In 2012, the estimated median contribution of fires to maximum daily 8 h average (MDA8) O3 in Texas during the month of June was 2 ppbv, with maximum impacts of over 40 ppbv (McDonald-Buller et al., 2015). The long-range transport of fire emissions has also been found to contribute to elevated peak O3 values in Europe (Ordóñez et al., 2010). Large uncertainties exist, however, in quantifying O3 production, which stems from uncertainties in fire emissions, combustion efficiency, meteorological patterns, chemical and photochemical reactions, and the effects of aerosols on plume chemistry and photolysis rates. Aerosols have been shown to both increase O3 formation (e.g., scattering particles can increase photolysis rates) as well as decrease O3 (absorbing aerosol and black carbon-containing aerosol can reduce photolysis rates) (Baylon et al., 2018). Hence, the aerosol composition and size distribution, which varies within and between plumes (Collier et al., 2016), and the location of the aerosol within the plume (Alvarado et al., 2015) impact O3 production. The presence of clouds also impacts photolysis rates and O3 production (Flynn et al., 2010). All of these factors remain highly variable and uncertain between different plumes.

Biomass burning also emits particulate matter (PM) that impacts air quality, health, and climate. PM impacts the climate directly by scattering or absorbing incoming solar radiation (e.g., Boucher et al., 2013) and indirectly by altering the properties of clouds (e.g., Pierce et al., 2007; Spracklen et al., 2011) with both effects depending on the particle size, mass, and composition (Petters and Kreidenweis, 2007; Seinfeld and Pandis, 2016). Bond et al. (2013) estimated that biomass burning emits about one-third of total global primary carbonaceous aerosol emissions (black carbon (BC) and organic aerosol (OA), with the size, mixing state, and chemical composition of the particles uncertain. A complex evolution of various organic trace gas and aerosol compounds occurs as smoke ages, with all compounds containing a wide variety of volatility and reactivity that determine the ultimate partitioning into the gas or particle state, thus determining the size, mixing state, and ultimately chemical composition of the evolving plume. Hodshire et al. (2019a) reviewed the wide-ranging results from laboratory and field studies of smoke plume aging, which show that measured net OA production or loss is dependent on the fuel and burning conditions, plume dispersion rates, and oxidant species and concentrations; however, no complete theory currently exists that can predict how OA will evolve in different plumes. Further understanding of the magnitude and extent of both the primary and secondary components of biomass burning emissions is thus required to fully understand the global impacts.

Over the last few decades, air quality regulations have resulted in a decrease in PM concentrations in the United States, as PM2.5 is a regulated pollutant under the Clean Air Act National Ambient Air Quality Standards (NAAQS). McClure and Jaffe (2018), however, analyzed PM2.5 measurements made at IMPROVE sites and found a positive trend in the 98th quantile of PM2.5 in the northwest and a negative trend in the rest of the US, attributing the increase to wildfires in the northwest, similar to positive trends in MODIS aerosol optical depth (AOD). They determined that wildfires are causing the increase in PM2.5 at the 98th quantile in the northwest, which could offset anthropogenic reductions in the region. O'Dell et al. (2019) and Knorr et al. (2017) combined surface observations and satellite-based smoke plume estimates and the GEOS-Chem chemical transport model (CTM) to identify trends in summertime smoke, non-smoke, and total PM2.5 across the US. They estimated that future PM emissions from biomass burning may exceed anthropogenic emission levels, including in densely populated areas in the eastern Europe–Russia–central-Asia region. The growing relative importance of biomass burning as a source of pollution increases the need to understand in-plume chemistry and physics.

Three-dimensional (3D) Eulerian CTMs take estimates of the primary emissions from fires and unphysically mix them across large-scale grid boxes, which can lead to incorrect estimates of the ultimate impact of fires on health, air quality, and climate (e.g., Alvarado et al., 2009; Sakamoto et al., 2016; Ramnarine et al., 2019; Hodshire et al., 2019b). Thus, in order to accurately predict biomass burning effects on air quality and climate in regional and global models, a sub-grid-scale representation of aged biomass burning trace gas and aerosol evolution is required. Regarding the impact of coarse-model mixing on O3, Baker (2015) found that the 3D Eulerian Community Multiscale Air Quality Model (CMAQ) tended to overestimate the impact of fires on individual hourly O3 measurements at US Environmental Protection Agency (EPA) Clean Air Status Trends Network (CASTNET) monitoring sites near fires by up to 40 ppbv and underestimate it further downwind by up to 20 ppbv. This behavior is consistent with an incorrect treatment of the sub-grid-scale, near-source O3 and NOy chemistry, where the model underestimates the loss of NOx (NO+NO2) near the source due to the formation of inorganic and organic nitrates, thus overestimating O3 formation near the source. This same error leads to an underestimate of the amount of peroxy nitrates formed near the source, which then leads to an underestimate of O3 formation downwind when the peroxy nitrates decompose, regenerating NOx (Alvarado et al., 2010).

Similarly, the unphysical mixing of biomass burning emissions into large-scale grid boxes can lead to incorrect estimates of OA concentrations and the aerosol size distribution (e.g., Alvarado et al., 2009; Sakamoto et al., 2016; Bian et al., 2017; Hodshire et al., 2019b; Konovalov et al., 2019). The net change in OA mass in a smoke plume as it dilutes and ages is determined from the balance between initial emissions, secondary organic aerosol (SOA) production, and evaporation of both primary organic aerosol (POA) and SOA (Bian et al., 2017; Hodshire et al., 2019b). Unphysically diluting biomass burning emissions leads to unphysical evaporation of the POA and reduces the rates of chemical SOA formation and more of the formed SOA remaining in the gas phase in the 3D Eulerian CTMs. Similarly, the unphysical dilution reduces the aerosol number concentration, reducing coagulation rates, while the more diluted smoke will not reach the high concentrations needed to nucleate new particles. As the evolution of the aerosol size distribution in smoke plumes is primarily controlled by OA mass changes, coagulation, and nucleation, 3D Eulerian CTMs will have difficulty accurately simulation the aerosol size distribution changes without parameterizing these sub-grid-scale processes.

The initial aerosol size, number, and mass in biomass burning smoke plumes can vary with fuel type (Janhäll et al., 2010) (e.g., Boreal versus Savannah) and combustion condition (Hosseini et al., 2010) and are leading uncertainties in the predictions of PM in regional and global models (Lee et al., 2013). These inputs are often based on spatially sparse, point measurements taken at only one stage in the atmospheric lifetime of a biomass burning plume with some measurements representing fresh emissions and some representing aged emissions (Pierce et al., 2007; Janhäll et al., 2010; Akagi et al., 2011; Hodshire et al., 2019a). These sparse inputs do not account for many of the non-linear physical and chemical changes that take place within a smoke plume near the fire, with the coarse grid scales of regional and global models (tens to hundreds of kilometers) too large to resolve near-source smoke plume chemical and physical evolution. By accounting for sub-grid aerosol processes that occur in biomass burning plumes, such as coagulation and condensation or evaporation of organic species, the biomass burning impact on aerosol number concentration and size distribution can be better simulated (Ramnarine et al., 2019). In order to resolve aerosol processes in biomass burning plumes, regional and global models thus require grid-scale-appropriate, aged aerosol emissions size distributions to accurately simulate the health and climate effects of biomass burning aerosols in global and regional atmospheric models. Additionally, in order to better characterize the chemical processes in biomass burning plumes, an improved understanding of the oxidant and radical concentrations, photolysis rates, and parameterizations of reaction rates for different classifications of smoke is needed (Hodshire et al., 2019a).

Several types of models have been used to simulate the dispersion and transport of smoke plumes, including box models, Gaussian plume models, Lagrangian puff and particle dispersion models (e.g., CALPUFF, SCIPUFF, HYSPLIT, FLEXPART), and 3D Eulerian models (e.g., Goodrick et al., 2013, and the references therein). A smaller number of models have included the gas (e.g., Mason et al., 2001) and aerosol (e.g., Trentmann et al., 2003) chemistry of these plumes, and a smaller number still have tried to predict how the aerosol size distribution changes within the smoke plume (e.g., Sakamoto et al., 2016; Hodshire et al., 2019b). As an initial attempt to represent sub-grid plume chemistry and physics in coarse-grid models, Lonsdale et al. (2015) developed a parameterization of trace gas and aerosol formation in biomass burning plumes using the Aerosol Simulation Program (ASP; Alvarado et al., 2015) as a box model. ASP simulates the gas-phase, aerosol-phase, and heterogeneous chemistry of young biomass burning smoke plumes, including the formation of O3 and secondary inorganic and organic aerosol. The ASP box model parameterization included predicted normalized excess mixing ratios (NEMRs; Akagi et al., 2011) of O3, NOx, PAN, and other trace gases and aerosol species in terms of the fuel type, temperature, latitude, day of year, and starting hour of fire emission. Separate parameterizations were built for each fuel type, which included savannah, tropical forest, temperate forest, and boreal forest. McDonald-Buller et al. (2015) used a subset of this ASP-based parameterization to adjust the chemistry of biomass burning in the Comprehensive Air Quality Model with Extensions (CAMx) and found that this approach reduced the median impact of biomass burning on MDA8 O3 in Texas by 0.3 ppbv or 15 %. However, because the parameterization was fit to the ASP box model, it did not include cross-plume gradients in trace gas and aerosol concentrations, which may be important for accurately simulating non-linear chemistry and partitioning (Garofalo et al., 2019; Hodshire et al., 2019b; Bian et al., 2017). To account for non-linear cross-plume dilution effects, Sakamoto et al. (2016) used the large-eddy simulation (LES) cloud-resolving model, the System for Atmospheric Modelling (SAM; Khairoutdinov and Randall, 2003; Stevens et al., 2012), coupled with the TwO Moment Aerosol Sectional (TOMAS) microphysics module to parameterize the coagulation of aerosols in biomass burning plumes (Sakamoto et al., 2015, 2016). This parameterization was used in Ramnarine et al. (2019) to investigate the impact of sub-grid coagulation on radiative forcing. However, while the SAM-TOMAS model used by Sakamoto et al. (2016) resolved plume gradients, their study did not include chemistry and phase partition. There remains a need for a modeling system that resolves plume gradients while simulating the chemical and physical processes relevant for air quality and climate.

To address the need for a dispersion-resolving model with online chemistry, partitioning, and microphysics that can help answer the biomass burning questions described above, we have developed an integrated model of ASP (Sect. 2.1) coupled with the SAM model (Sect. 2.2). We have evaluated the performance of the new model, SAM-ASP v1.0 described in Sect. 2.3, in simulating the measurements of CO, O3, NOy, and aerosols for the Williams fire in California (Sects. 3 and 4). This integrated model is able to simulate both the detailed chemistry and the horizontal and vertical dispersion affecting the near-source evolution of biomass burning gas and aerosol chemistry and physics. Model code and inputs are publicly available as described in Sect. 6.

We evaluated the performance of the newly coupled SAM-ASP v1.0 model by comparing model output to observations of the Williams fire made by Akagi et al. (2012), which was previously simulated using ASP in a Lagrangian box model by Alvarado et al. (2015). Emission ratios for this simulation were based on observed relative background-corrected concentration close to the source from Alvarado et al. (2015) and included observed values for many gas-phase species measured by Akagi et al. (2012). Plume injection height was set to between 1200 and 1400 m, as this was the height at which the plume was observed to level off, where a small amount of vertical mixing can be seen as the plume ages.

The large-scale meteorological forcing in SAM-ASP v1.0 is driven by the 3 h, 32 km resolution North American Regional 30 Reanalysis (NARR; Messinger et al., 2006) meteorology dataset. The fire simulated in this study to evaluate the SAM-ASP v1.0 model was a prescribed fire measured on 17 November 2009 called the Williams fire (Akagi et al., 2012). This fire covered 81 ha north of Buellton, CA, with the fuel type classified as chaparral and the vegetation burned consisting of coastal sage scrub and scrub oak woodland understory. Surface temperatures ranged from 19 ∘C at 09:00 local time to 24 ∘C at 12:20 local time with clear skies throughout the fire duration. The plume built up gradually during the day with most of the smoke rising to ∼1200–1336 m above mean sea level and then drifting in a northeast direction. Two flights were conducted during the day on board a US Forest Services Twin Otter aircraft to sample initial emissions and aged smoke with an airborne Fourier transform infrared spectrometer instrument from the University of Montana taken during both flights including background measurements sampled at similar altitudes to in-plume measurements just outside the plume. Trace gas species emission factors determined as initial emissions (within minutes of the emission source) by Akagi et al. (2012) are used to initialize ASP and included: O3, nitrous acid (HONO), ammonia (NH3), ammonium (NH4), nitric oxide (NO), NO2, NOx (as NO), OA, and peroxy acetyl nitrate (PAN). Additional emission factors not measured, but needed to initialize ASP, were included using emission factors from Table 2 of Akagi et al. (2011), with the full list of ASP species provided in the model code repository described in Sect. 6. Initial aerosol size distribution information is inferred from the smoke study of Grieshop et al. (2009a, b). Full details of the fire and measurements are also further described in Akagi et al. (2012). Model input background concentrations were assumed based on measurements taken outside of the defined plume. Additional static inputs required by ASP include a photolysis rate parameterization based on the time and latitude of the fire and chemical data on aqueous phase ions and inorganics. Details of the static ASP inputs are further described in Alvarado (2008).

The model used here has 10 size bins with the total number of fire-emitted particles derived from multiplying the CO flux (based on measured values) by the ratio of particle number enhancement (number of particles cm-3) to CO enhancement (ppb) (ΔN/ΔCO=23.7 particles cm-3ppb-1). We use a number mean diameter of 0.1 µm and a standard deviation of 1.9 based on the wood smoke study of Grieshop et al. (2009a, b).

The NEMR calculations were determined by calculating the average species of interest (X) and CO concentration across the plume, as was done in the measurements. The NEMR (ΔX/ΔCO or ΔX/ΔCO2) is then calculated relative to CO or CO2 since they have relatively long lifetimes for the fire location, low background variability, and there were no other major nearby sources as described in Akagi et al. (2012): 2ΔX/ΔCO=(Xin-plume-Xbackground)/(COin-plume-CObackground). Note that, in general, NEMRs are imprecise indicators of chemical changes, especially for plumes that have traveled far from their original source and may have mixed with different types of background air; thus, defining a single background concentration to subtract from the plume concentration is not a realistic approach (e.g., Yokelson et al., 2013). However, for the Williams fire, the excess mixing ratios downwind tended to vary slowly in time and space compared to measurement frequency, and the background value was computed from the average of a large number of points at the plume altitude (but outside the plume; Akagi et al., 2012).

Averaged NEMR values over the full horizontal domain of the model for the vertical level with the peak ΔCO (and ΔCO2) can be used to compare with aircraft observations of biomass burning plumes. Figure 2 shows a horizontal slice of the simulated ΔCO concentrations (ppbv) as the plume moves downwind (bird's eye view at 1200 m above ground). Note that the initial plume was distributed across two horizontal gird boxes (initial plume width of 1 km) and four vertical grid boxes (initial height from 1200 to 1360 m) and was rectangular. The emissions were distributed proportional to the density of air in each grid box and initially propagated downward due to wind shear and diffusion. The dilution of the plume can be seen in the ΔCO values as high as 16 000 ppbv between the center of the plume within the first hour after initial emission to 5 h downwind, where the plume was modeled to be approximately 100 km wide, with an average in-plume ΔCO concentration of approximately 1000 ppbv. Figure 3 shows a vertical slice of (looking into) the plume at 1, 2, and 5 h downwind for ΔCO, ΔO3/ΔCO, and ΔOA/ΔCO2, and these results will be discussed in the following sections. Note that ΔCO2 was used to as the NEMR denominator for OA, as in Akagi et al. (2012) and Alvarado et al. (2015), as in the field study OA and CO2 were measured on the same inlet, while CO was measured on a different inlet. The uncertainty in the Lagrangian age (horizontal error bars in Fig. 3) was calculated as in Akagi et al. (2012), where the 1σ uncertainty in the average horizontal wind speeds during the sampling period were propagated through the plume age calculation, assuming the distance calculation was accurate.

For better comparison between the ASP v2.1 box model of Alvarado et al. (2015) and SAM-ASP v1.0, all emission ratios and background concentrations were made identical in box models. The same gas-phase chemical mechanism, aerosol thermodynamics routines and parameters, aerosol size distribution routines and parameters, and other chemical parameters were used. Thus the key difference between the two models is the treatment of plume dilution and mixing (with minor differences due to vertical temperature, pressure, and humidity variations in SAM-ASP v1.0 versus constant parameters used in ASP v2.1). In ASP v2.1, the plume is a single well-mixed box and dilution is parameterized by assuming a one-way addition of background air to the plume. As in Mason et al. (2001) we assume a Lagrangian parcel of fixed height (H) and length but variable width y(t). This assumes the plume is well-mixed vertically and capped at the top and bottom by a strong stable layer (or the surface). The temperature and pressure of the parcel are assumed to be constant. The effect of plume dispersion on concentrations is then (Mason et al., 2001; Alvarado, 2008) 3dCqdtdisp=-1ytdytdtCq-Cqa, where Cq is the concentration of species q within the parcel (molecules cm-3) and Cqa is the concentration of species q in the atmosphere outside of the parcel. The form of y(t) is assumed to be y(t)=yo2+8Kyt2, where yo is the initial plume width (Mason et al., 2001). Ky represents the horizontal diffusivity of the atmosphere. The effect of plume dispersion then becomes 4dCqdtdisp=-4Kyyo2-8Kyt2Cq-Cqa. This equation is used with the observations of the rate of change in excess CO in the Williams fire plume to derive best fit values for Kyusing the observed value of yo.

In SAM-ASP v1.0, horizontal and vertical mixing between the boxes of the Lagrangian wall are calculated as part of the tracer transport routines of SAM 6.10.10 described in Sect. 2.2 (Khairoutdinov and Randall, 2003). In addition, unlike the ASP v2.1 box model of Alvarado et al. (2015), plume gradients are preserved in SAM-ASP v1.0. Thus, the chemistry taking place in the center of the plume may differ from that in the edges of the plume, potentially changing the plume-average NEMRs from those calculated with the well-mixed box assumption in ASP v2.1.

We have described a new coupled model, SAM-ASP v1.0, for simulating the gas and aerosol chemistry within biomass burning smoke plumes. The model adds the Aerosol Simulation Program v2.1 (ASP v2.1) as an embedded subroutine within the System for Atmospheric Modeling v6.10 (SAM v6.10). When configured as a 2D Lagrangian wall, the newly coupled SAM-ASP model allows for a detailed examination of the chemical and physical evolution of fine-scale biomass burning plumes.

SAM-ASP is able to simulate the complex, non-linear production of O3 and changes in PM as plumes age. It is able to resolve the cross-plume chemistry, gas-to-particle partitioning, and microphysics that coarser grid-scale CTMs are not able to. Model results indicate that SAM-ASP is able to accurately simulate the dilution of CO in a California chaparral smoke plume mostly, except for a slight initial underprediction, as well as accurately predict the chemical loss of NOx and HONO and the production of O3 and PAN within the plume. SAM-ASP also resolves the cross-plume concentration of trace gases and aerosol. However, when compared to observations, the simulation with SAM-ASP did not show any significant differences with respect to a much simpler box model simulation, potentially because the photolysis rates within both simulations were identical rather than allowing the photolysis rates to vary with predicted aerosol concentrations.

Future work will involve testing SAM-ASP simulations against observed plume crosswind and vertical gradients as well as size distributions. Future work will also include the development of a biomass burning parameterization of plume-scale chemical and physical trace gas and aerosol evolution for use in coarser grid-scale CTMs (that cannot resolve plumes) as well as the implementation of on-line photolysis calculations to explicitly simulate the effect of in-plume aerosols on photolysis rates.