To anticipate the impacts of smoke, the timing and location of smoke concentrations become important. Data on the site-specific surface concentrations of respirable particles and gases often are needed for estimating impacts on public health and welfare. Data on the cumulative concentrations of elements that scatter and absorb light also are needed to estimate impacts on visibility and haze. Ambient air quality can be measured at a point or as distribution of air quality over any space and time of interest. Ambient air quality is affected by:

• the pollutants emitted to the atmosphere from fires,
• the background air quality that has already been degraded by other sources,
• the transport of the polluted parcels of the atmosphere,
• dispersion due to atmospheric movement and turbulence, secondary reactions, and removal processes.

Plume rise is an important component of transport, because it determines where in the vertical structure of the atmosphere dispersion will begin. Other basic elements that are important in the trajectory and dispersion of smoke are heat release, advection and diffusion, scavenging, and chemical transformations.

Overall, transport and dispersion has proven extremely difficult to model accurately, especially in complex terrain. For example, detailed, gridded, three-dimensional meteorological data are required to model transport and dispersion, but expert judgment is often required to supplement or substitute for such modeled predictions. Despite the difficulties of modeling, since about 1990 modeling systems used to assess the air quality impact of fires have grown increasingly important to both the fire planning and air quality communities. Today, there is a broad range of transport and dispersion models available. These models fall into four major categories:

• Plume models (VSMOKE and SASEM)
• Puff models (Hysplit, Calpuff, and NFSpuff)
• Particle models (PB-Piedmont)
• Grid models(REMSAD, Models-3/ CMAQ)

Although progress has been made, none of the currently available models fully meet the needs of fire planners and air resource managers. Much of the deficiency in current modeling approaches is caused by inherent uncertainties associated with turbulent motions between the fire, smoke, and the atmosphere that are compounded by the highly variable distribution of fuel elements, composition, and condition. Another source of deficiency is that most available models were originally designed for well-behaved sources such as industrial stacks or automobile emissions, while emissions from fire can be extremely variable in both time and space. Also, outputs from currently available models do not always match the temporal or spatial scale needed for land management application.

Due to these deficiencies, current models to predict trajectory or air quality impacts from fires are inadequate in coverage and are incomplete in scope (Sandberg and others 1999). Application of these models are appropriate mainly for relatively homogeneous fuelbeds and steady state burn conditions, restricting them to fires on a local scale or to those where fuels are scattered or piled uniformly over the landscape.

Because of new interest in modeling emissions on a regional scale, land managers need transport and dispersion models that include all fire and fuel types as well as multiple sources. Such models need to be linked to other systems that track fire activity and behavior as well as provide for variable scaling to fit the area of interest. At the operational level, models that support real-time decision making during fire operations in both wildland fire situation analysis and go/no-go decision making are also needed (Breyfogle and Ferguson 1996).

The consumption of biomass produces thermal energy, and this energy creates buoyancy to lift smoke particles and other pollutants above the fire. Heat release rate is the amount of thermal energy generated per unit of time. Total heat release from a fire or class of fires is a function of the heat content of the biomass, fuel consumed, ignition method and pattern, and area burned.

The early work of Anderson (1969) and Rothermel (1972) created fundamental equations for combustion energy in a variety of fuelbeds. Sandberg and Peterson (1984) adapted the combustion equations to model the temporal change in energy during flaming and smoldering combustion (Emission Production Model, EPMv.1.02). Currently, EPM provides heat release rates for most biomass smoke dispersion models (Harms and others 1997; Harrison 1995; Lavdas 1996; Sestak and Riebau 1988; Scire and others 2000a) and has been used to estimate the change in global biomass emissions patterns due to changes in land use (Ferguson and others 2000). The model, however, requires a constant rate of ignition with constant slope and wind. Such homogeneous conditions may be approximated during prescribed fires that are ignited with a deliberate pattern of drip torches or airborne incendiaries, or during portions of wildfires that experience relatively constant spread rates, both over fuelbed strata that retain a relatively consistent spatial and compositional pattern. To use EPM effectively for modeling source strength, the fire area and ignition duration are broken into space and time segments that meet the steady-state criteria.

Albini and others (1995), Albini and Reinhardt (1995), and Albini and Reinhardt (1997) do not explicitly derive temporal changes in combustion energy in their model, BurnUp, but they do assign source heat in steps of flaming and smoldering that are estimated from total fuel consumption. They have linked their model with the fire spread model, FARSITE (Finney 1998), which allows ignition rates and subsequent heat-release rates to vary over the landscape. The coupled system is computationally expensive and not yet associated with a plume rise component but may offer a reasonable approximation of the temporal and spatial varying emission rates of fires.

Modeling of the transport and dispersion of industrial stack plumes has occurred for decades, prompting a variety of techniques. But application to fires is much more limited (Breyfogle and Ferguson 1996). Part of the reason for this is that source strength from undulating and meandering fires is so difficult to simulate accurately. Therefore, applications have been appropriate mainly for relatively homogeneous fuelbeds and steady state burn conditions. This has restricted most transport and dispersion modeling to fires on a local scale and to those started in harvest residue from land clearing operations where fuels are scattered uniformly over the landscape or collected into piles (Hardy and others 1993; Hummel and Rafsnider 1995; Lavdas 1996; Sestak and Riebau 1988). Global-scale modeling also has taken place where fuelbed and ignition patterns are assumed to be approximately steady state in relation to the grid size (Kasischke and Stocks 2000; Levin 1996).

Gaussian plume models (Harms and others 1997; Lavdas 1996; Sestak and Riebau 1988; Southern Forest Fire Laboratory Personnel 1976) are useful for places with relatively flat terrain, for circumstances when input data are scarce, and for evaluating surface concentrations relatively near the source. These models typically require only an estimate of atmospheric stability, trajectory wind speed and direction, and emission rates. Fires are modeled independently. Therefore, accumulations of smoke from multiple fires are ignored. Some Western States require SASEM modeling of prescribed burns before they can be permitted (Battye and Battye 2002).

Puff models (Draxler and Hess 1998; Harrison 1995; Hummel and Rafsnider 1995; Scire and others 2000a) are needed when simulating long-range transport, or transport that occurs during changeable environmental conditions such as influences from complex terrain or variable weather. NFSpuff has an easy user interface, but because of its internal terrain data files it is restricted to applications in the Western States, excluding Alaska (Harrison 1995). NFSpuff is the most commonly used puff model for prescribed fire planning (Dull and others 1998).  Hysplit (Draxler and Hess 1998) currently is programmed to accept only 16 individual sources and assumes a constant rate of emissions with no plume rise. Hysplit (Draxler and Hess 1998) and Calpuff (Scire and others 2000a) both include simple chemistry. All three models are linked to the MM5 meteorological model (Grell and others 1995). NFSpuff can function with a simple trajectory wind, and Hysplit and Calpuff can accept other gridded weather input data.

Particle models are used in coupled fire-atmosphere modeling (Reisner and others 2000) and for tracking critical signature elements (Achtemeier 1994, 2000; Draxler and Hess 1998). The sophistication of these types of models and their computational requirements, however, has thus far limited their application to research development or individual case studies.

Eulerian photochemical grid models are highly useful in estimating smoke concentrations from many sources over large domains. In addition, their ability to model secondary chemical reactions and transformations is needed for determining ozone concentrations and regional haze conditions. Regional planning organizations such as the Western Regional Air Partnership (WRAP), are evaluating the photochemical models Models-3/CMAQ (Byun and Ching 1999) and REMSAD (Systems Applications International 2002) for use in guiding State implementation plans (SIPs) and Tribal implementation plans (TIPs).

Additional work is needed to fill critical gaps in the modeling systems identified above. As the need for better information on the impact of fire on air quality increase, so too will demands on modeling systems.

Heat, particle, and gas emissions from fires vary in time and space, causing unique patterns of convection and resulting plume rise. This plume rise is a function of free convection in the atmosphere, which is caused by density differences within the fluid. As a fire heats and expands air near the ground, large density differences between the heated volume and the surrounding air mass are created, causing the heated parcel to rise. The potential height of the resulting plume depends on the heat energy of the source and rise velocity, which is affected by the exchange and conservation of mass, radiant heat loss, the buoyancy force, and turbulent mixing with the ambient air.

Hot, flaming fires can develop central convective columns with counter-rotating vortices that involve massive entrainment of the surrounding air mass (Clark and others 1996; Haines and Smith 1987; Haines and Updike 1971). This stage of fire can produce fast-rising plumes and turbulent downdrafts, carrying sparks that ignite new fires (see Fire and Plumes). Cumulonimbus clouds often develop with accompanying lightning and rain. Dynamic plume rise brings gas and particles high into the atmosphere where strong winds can disperse the smoke hundreds to thousands of kilometers. As high intensity fires cool, however, the central column often collapses, creating numerous small convective cells that are less dynamic but equally active in carrying smoke into the atmosphere. Smoldering fires often create plumes that are neutrally buoyant, limiting widespread dispersion but allowing surface winds to dominate smoke trajectories. This can lead to accumulations of smoke in valleys and basins at night.

Because plume rise can eventually result in wide-spread dispersion, plume rise calculations are essential for determining the height above ground from which plume dispersion is initiated. Uncertainties in such calculations can result in inaccurate predictions of plume transport and downwind smoke impacts. Given the pressing need to predict the impact of plumes from fires, the need for improved plume rise calculations is apparent.

Limitations of current plume-rise models

The basic mechanisms and algorithms used to describe plume rise and buoyancy were developed in the mid-1960s by Briggs (1969) for industrial, ducted emissions. These methods are still used today to estimate the plume rise and buoyancy of fires in spite of the significant differences in characteristics between ducted emissions and prescribed and wildland fires:

• Heat released from ducted sources is precisely known and usually emitted at relatively constant rates during a single phase of combustion. Heat released from fires is a function of fuel loading, fuel conditions, and ignition method through several phases of combustion (pre-ignition, flaming, smoldering, and residual), which create highly variable magnitudes and rates of heat release.
• Nearly all of the energy generated at the source of a ducted plume is transmitted to convection energy. In open burning, however, significant amounts of energy are lost by conduction and radiation, reducing the amount of available energy for convection.
• Plumes from ducted sources create single convective columns, but low intensity understory burning that occurs over broad areas does not develop a cohesive plume.

To improve plume rise predictions, emission production models need to do a better job of characterizing the spatial and temporal pattern of heat release from fires, and plume rise models need to be improved to account for the energy lost from the convective system through radiation and turbulent mixing. While models such as EPM and Burnup (described in Heat Release) simulate variable rates of heat release from fires, both models use general estimates of spatial distributions of fuel, including structure, composition, and moisture content. Also, significant elements of fires that influence convective energy -- such as the distribution of naturally piled fuel ("jackpots"), amount and density of rotten fuel and duff, and release of water vapor -- are not adequately captured.

Rough approximations on the proportion of energy available for convection were made more than 40 years ago (Brown and Davis 1959). Despite efforts to improve plume rise calculations by removing the density difference assumption (Scire and others 2000a), they still are in use today.

Low intensity fires that typically do not have a cohesive convective column must be treated, from a modeling perspective, as an area source in Eulerian grid models. In Lagrangian dispersion models, there is currently no valid means of calculating plume rise from unconsolidated convection. Eulerian coordinates (used by box and grid models) are coordinate systems that are fixed in space and time, and there is no attempt to identify individual particles or parcels from one time to the next. Lagrangian models (bell-shape or Gaussian distribution pattern, often applied to plume and puff models) are used to show concentrations crosswind of the plume.

Another complication for modeling is that once plumes from fires enter the atmosphere, their fluctuating convection dynamics make them more susceptible to erratic behavior than well-mannered industrial stacks. For example, different parts of a plume can be carried to different heights in the atmosphere at the same time. This causes unusual splitting patterns if there is a notable wind shear between lofted elevations, causing different portions of the plume to be transported in different directions. Therefore, predictions of the plumes impact on visibility and air quality under these conditions become highly uncertain (Walcek 2002). Even when the behavior of plumes from fires resembles that of stack plumes, the varying and widely distributed locations of wildland sources prevent consistent study. For example, downwash of plumes has been observed from ducted (stack) emissions after an inversion breaks up-  conditions that are common at the end of an onshore breeze if the plume is above the inversion at its source (de Nevers 2000; Venkatram 1988) or if horizontal stratification in the lower atmosphere is disrupted by mountains (de Nevers 2000).

These characteristics of plumes from fire are strikingly different than those of ducted industrial emissions yet little research has been done on this topic in the past several decades.

In most existing models, the horizontal advection of smoke and its diffusion (lateral and vertical spread) are assumed to be controlled mainly by wind, and the formation and dissipation of atmospheric eddies. These elements are greatly simplified by assuming constant wind (at least for an hourly time step) in some cases (such as VSMOKE and SASEM), and a Gaussian dispersion is nearly always imposed. Perhaps the most critical issues are the constantly changing nature of the plume due to scavenging, chemical transformation, and changing convection dynamics that affect plume transport.

Many photochemical and dispersion models depend on gridded meteorological inputs. Unfortunately, numerical formulations of dynamic meteorological models (for example, MM5: Grell and others 1995; RAMS: Pielke and others 1992) do not adequately conserve several important scalar quantities (Byun 1999a, 1999b). Therefore, modelers often introduce mass-conserving interpolations. For example, Models-3/CMAQ (Byun and Ching 1999) uses the MCIP scheme (Byun and others 1999), Calpuff (Scire and others 2000a) employs CALMET (Scire and others 2000b), and TSARS+ (Hummel and Rafsnider 1995) is linked with NUATMOS (Ross and others 1988). Driving a photochemical or dispersion model without these mass-conserving schemes will produce inaccurate results, especially near the ground surface.

Smoke particles by nature of their small size provide efficient cloud condensation nuclei. This allows cloud droplets to condense around fine particles, called nucleation scavenging. Scavenging within a cloud also can occur as particles impinge on cloud droplets through Brownian diffusion, inertial impaction, or collision by electrical, thermal, or pressure-gradient forces (Jennings 1998). Cloud droplets eventually coalesce into sizes large enough to precipitate out, thus removing smoke aerosols from the atmosphere. While interstitial cloud scavenging, especially nucleation scavenging, is thought to dominate the pollution removal process, particles also may be removed by impacting raindrops below a cloud. Jennings (1998) reviews several theories on pollution scavenging but contends that there is little experimental evidence to support such theories.

The size and chemical structure of particles determine their efficiency in nucleation or other scavenging mechanisms. While the chemical composition of smoke is reasonably well known (see Atmospheric and Plume Chemistry), distributions of particle size from fire are not. The few airborne measurements (Hobbs and others 1996; Martins and others 1996; Radke and others 1990) do not distinguish fire characteristics or combustion dynamics, which play important roles in the range of particle sizes emitted from a fire. Therefore, the efficiency of scavenging biomass smoke particles out of the atmosphere by cloud droplets, rain, or other mechanism has not been quantified.

Chemical transformations provide another mechanism for changing particle and gas concentrations within a plume. Chemical transformation in the plume can be important in regional-scale modeling programs where sulfate chemistry and ozone formation are of interest (see Atmospheric and Plume Chemistry). Oxidation within the smoke plume causes a loss of electrons during chemical transformation processes, which increases polarity of a molecule and improves its water solubility (Schroeder and Lane 1988). This improves scavenging mechanisms by cloud and rain droplets. Chemical transformation rates depend on complex interactions between catalysts and environmental conditions such as turbulent mixing rates.

One of the simplest ways of estimating smoke concentrations is to assume that plumes diffuse in a Gaussian pattern along the centerline of a steady wind trajectory. Plume models usually assume steady-state conditions during the life of the plume, which means relatively constant emission rates, wind speed, and wind direction. For this reason, they can be used only to estimate concentrations relatively near the source or for a short duration. Their steady-state approximation also restricts plume models to conditions that do not include the influence of topography or significant changes in land use, such as flow from a forest to grassland or across a land-water boundary.

Gaussian plume models have a great benefit in places and circumstances that restrict the amount of available input data. They can be run fast and have simple but realistic output that can be easily interpreted. Many regulatory guidelines from the EPA are based on Gaussian plume models.

Plume models typically are in Lagrangian coordinates that follow particles or parcels as they move, assigning the positions in space of a particle or parcel at some arbitrarily selected moment. (Lagrangian coordinates are used by plume, puff, and particle models.) Examples adapted for wildland biomass smoke include VSMOKE (Harms and others 1997; Lavdas 1996) and SASEM (Riebau and others 1988; Sestak and Riebau 1988). Both models follow regulatory guidelines in their development and offer a simple screening tool for examining potential concentrations at receptor locations from straight-line trajectories relatively near the source. However, SASEM directly compares downwind concentrations with ambient standards and calculates visibility impairment in a simple manner. It is also used as a State regulatory model in Wyoming, Colorado, New Mexico, and Arizona, and has been recommended for use by the EPA.

Plume rise models developed for other applications might be useful if adapted to fire environments. For example, ALOFT-FT (A Large Outdoor Fire Plume Trajectory Model - Flat Terrain), developed for oil-spill fires (Walton and others 1996), is a computer-based model to predict the downwind distribution of smoke particulate and combustion products from large outdoor fires. It solves the fundamental fluid dynamic equations for the smoke plume and its surroundings with flat terrain. The program contains a graphical user interface for input and output, and a database of fuel and smoke emission parameters that can be modified by the user. The output can be displayed as downwind, crosswind, and vertical smoke concentration contours.

Instead of describing smoke concentrations as a steadily growing plume, puff models characterize the source as individual puffs being released over time. Each puff expands in space in response to the turbulent atmosphere, which usually is approximated as a Gaussian dispersion pattern. Puffs move through the atmosphere according to the trajectory of their center position. Because puffs grow and move independently of each other, tortuous plume patterns in response to changing winds, varying topography, or alternating source strengths can be simulated with some accuracy.

Some models allow puffs to expand, split, compact, and coalesce (Hysplit: Draxler and Hess 1998; Calpuff: Scire and others 2000a) while others retain coherent puffs with constantly expanding volumes (NFSpuff: Harrison 1995). In either case, the variability of puff generation, movement, and dispersion does not restrict the time or distance with which a plume can be modeled. Most puff models are computed in Lagrangian coordinates that allow accurate location of specific concentrations at any time.

In a particle model, the source is simulated by the release of many particles over the duration of the burn. The trajectory of each particle is determined as well as a random component that mimics the effect of atmospheric turbulence. This allows a cluster of particles to expand in space according to the patterns of atmospheric turbulence rather than following a parameterized spatial distribution pattern, such as common Gaussian approximations. Therefore, particle models tend to be the most accurate way of simulating concentrations at any point in time. Because of their numerical complexity, however, particle models usually are restricted to modeling individual point sources with simple chemistry or sources that have critical components such as toxins that must be tracked precisely. Particle models use Lagrangian coordinates for accurate depiction of place of each time of particle movement (for example, Hysplit: Draxler and Hess 1998; PB-Piedmont: Achtemeier 1994, 2000).

Grid models use Eulerian coordinates, disperse pollutants uniformly within a cell, and transport them to adjacent cells. The simplicity of advection and diffusion in a grid model allows these models to more accurately simulate other characteristics of the pollution, such as complex chemical or thermal interactions, and to be used over large domains with multiple sources. This is why grid models commonly are used for estimating regional haze and ozone and are often called Eulerian photochemical models. Much of the future work on fire impact assessment and planning at regional to national scales will be done by using grid models.

Because of their nature, grid models are not used to define accurate timing or locations of pollutant concentrations from individual plumes, only concentrations that fill each cell. This means that sources small relative to the grid size, which create individual plumes, will introduce unrealistic concentrations in places that are outside of the actual plume. Ways of approximating plume position and its related chemical stage include nesting grids to finer and finer spatial resolutions around sources of interest (Chang and others 1993; Odman and Russell 1991), establishing nonuniform grids (Mathur and others 1992), and creating "plume-in-grid" approximations (Byun and Ching 1999; Kumar and Russell 1996; Morris and others 1992; Myer and others 1996; Seigneur and others 1983).

Many regional haze assessments use the Regulatory Modeling System for Aerosols and Acid Deposition (REMSAD) (Systems Applications International 2002). This model was adapted from the urban airshed model­variable grid (UAMV) by removing its plume-in-grid feature and parameterizing explicit chemistry to improve computational efficiency. REMSAD incorporates both atmospheric chemistry and deposition processes to simulate sulfate, nitrate, and organic carbon particle formation and scavenging. As such, it is quite useful for simulations over large regions.

The Models-3/ CMAQ modeling system is designed to integrate the best available modules for simulating the evolution and dispersion of multiple pollutants at a variety of scales (Byun and Ching 1999). It includes chemical transformations of ozone and ozone precursors, transport and concentrations of fine particles and toxics, acid deposition, and visibility degradation.

At the other end of the grid modeling spectra are simple box models that describe pollution characteristics of a small area of interest. Box models instantaneously mix pollutants within a confined area, such as a valley. This type of model usually is restricted to weather conditions that include low wind speeds and a strong temperature inversion that confines the mixing height to within valley walls (Lavdas 1982; Sestak and others 1988). The valley walls, valley bottom, and top of the inversion layer define the box edges. The end segments of each box typically coincide with terrain features of the valley, such as a turn or sudden elevation change. Flow is assumed to be down-valley, and smoke is assumed to instantaneously fill each box segment. Few box models include the complex chemical or particle interactions that are inherent in larger grid models.