Study of Fuel-Smoke Dynamics in a Prescribed Fire of Boreal Black Spruce Forest through Field-Deployable Micro Sensor Systems

: Understanding the combustion dynamics of fuels, and the generation and propagation of smoke in a wildland ﬁre, can inform short-range and long-range pollutant transport models, and help address and mitigate air quality concerns in communities. Smoldering smoke can cause health issues in nearby valley bottoms, and can create hazardous road conditions due to low-visibility. We studied near-ﬁeld smoke dynamics in a prescribed ﬁre of 3.4 hectares of land in a boreal black spruce forest in central Alberta. Smoke generated from the ﬁre was monitored through a network of ﬁve ﬁeld-deployable micro sensor systems. Sensors were placed within 500–1000 m of the ﬁre area at various angles in downwind. Smoke generated from ﬂaming and smoldering combustions showed distinct characteristics. The propagation rates of ﬂaming and smoldering smoke, based on the ﬁne particulate (PM 2.5 ) component, were 0.8 and 0.2 m / s, respectively. The ﬂaming smoke was characterized by sharp rise of PM 2.5 in air with concentrations of up to 940 µ g / m 3 , followed by an exponential decay with a half-life of ~10 min. Smoldering combustion related smoke contributed to PM 2.5 concentrations above 1000 µ g / m 3 with slower decay half-life of ~18 min. PM 2.5 emissions from the burn area during ﬂaming and smoldering phases, integrated over the combustion duration of 2.5 h, were ~15 and ~16 kilograms, respectively, as estimated by our mass balance model.


Introduction
Smoke created from wildland fires causes air quality concerns for communities across North America. Depending on the location, duration, and volume of wildland fires, public health can be subject to moderate to severe risks at nearby and/or distant locations for short to extended periods. Emissions from the burning of biomass in wildland areas primarily generate carbon dioxide (CO 2 ), carbon monoxide (CO), particulate matters (PM), volatile organic compounds (VOC), nitrous oxides (NO x ), ammonia (NH 3 ), small amounts of sulphur dioxide (SO 2 ), and methane (CH 4 ) [1,2]. It has Fire 2020, 3, 30 3 of 17 and immediately after the spread of a firefront in relation to fuel types are not well understood. Better identification and characterization of flaming and smoldering smoke can enhance the modeling of wildland fire smoke propagations in micro level as well as macro level long range transports [25].
Recent developments in low-cost air monitoring sensors have opened up new possibilities of expanding the coverage of air monitoring to remote areas in a cost-effective approach [8,26]. These sensor systems, typically of the size of a shoebox, have energy consumptions of 5-10 watts, and cost significantly less than conventional analyzer systems. In this paper, we describe the deployment of custom-built low-footprint field deployable air monitoring micro systems in a prescribed fire of boreal forest in central Alberta, undertaken in May 2019, to study fuel-smoke dynamics through near field real-time measurements. Our analysis of PM 2.5 measurements by a network of five micro sensor systems identifies and characterizes smoke from both flaming and smoldering phases of the fire. We show that smoke intensity can have strong spatial distributions, and smoke from smoldering origin may have a sustaining presence in the near-field regions of a fire. We demonstrate the practical use of low-cost sensor systems to parameterize fine particulate matter emissions from flaming and smoldering phases of combustion estimated through a mass balance model.

Study Area
The prescribed fire was conducted on 3.4 hectares of land on boreal forest in central Alberta (Pelican Mountain unit 5). The forest area was predominantly covered by black spruce, with canopy closure of over 50%. The ground consisted of thick organic soil covered mostly with feathermosses with a minor presence of Sphagnum mosses. A well-mixed boundary layer with predominant southerly winds at speeds varying in ranges of 5 km/h up to 23 km/h (10 m above ground level open wind speed) with occasional gusts were measured by a sonic anemometer placed in the south of the area. The timing of the fire was late afternoon on 11 May 2019. Fire was ignited through a helicopter-borne torch along the ignition line at the south perimeter of the area. The fire behavior and the spread to the north direction was recorded through in-situ measurements. Details can be found in Thompson et al. [27]. Five micro-stations were deployed at downwind locations of the prescribed fire zone. Four micro-stations were placed approximately at distances of 500 m from unit 5 at northeast, north, northwest, and west-northwest directions, respectively. One micro-station was placed at approximately 1 km distance in northwest direction. Micro-stations were transported to deployment sites through helicopters, and exact locations of deployments were subject to site accessibility. All micro-stations were deployed approximately twenty-four hours before the prescribed fire to monitor the background air quality in the area. Figure 1 shows the micro-station deployment locations. the modeling of wildland fire smoke propagations in micro level as well as macro level long range transports [25]. Recent developments in low-cost air monitoring sensors have opened up new possibilities of expanding the coverage of air monitoring to remote areas in a cost-effective approach [8,26]. These sensor systems, typically of the size of a shoebox, have energy consumptions of 5-10 watts, and cost significantly less than conventional analyzer systems. In this paper, we describe the deployment of custom-built low-footprint field deployable air monitoring micro systems in a prescribed fire of boreal forest in central Alberta, undertaken in May 2019, to study fuel-smoke dynamics through near field real-time measurements. Our analysis of PM2.5 measurements by a network of five micro sensor systems identifies and characterizes smoke from both flaming and smoldering phases of the fire. We show that smoke intensity can have strong spatial distributions, and smoke from smoldering origin may have a sustaining presence in the near-field regions of a fire. We demonstrate the practical use of low-cost sensor systems to parameterize fine particulate matter emissions from flaming and smoldering phases of combustion estimated through a mass balance model.

Study Area
The prescribed fire was conducted on 3.4 hectares of land on boreal forest in central Alberta (Pelican Mountain unit 5). The forest area was predominantly covered by black spruce, with canopy closure of over 50%. The ground consisted of thick organic soil covered mostly with feathermosses with a minor presence of Sphagnum mosses. A well-mixed boundary layer with predominant southerly winds at speeds varying in ranges of 5 km/h up to 23 km/h (10 m above ground level open wind speed) with occasional gusts were measured by a sonic anemometer placed in the south of the area. The timing of the fire was late afternoon on 11 May 2019. Fire was ignited through a helicopterborne torch along the ignition line at the south perimeter of the area. The fire behavior and the spread to the north direction was recorded through in-situ measurements. Details can be found in Thompson et al. [27]. Five micro-stations were deployed at downwind locations of the prescribed fire zone. Four micro-stations were placed approximately at distances of 500 m from unit 5 at northeast, north, northwest, and west-northwest directions, respectively. One micro-station was placed at approximately 1 km distance in northwest direction. Micro-stations were transported to deployment sites through helicopters, and exact locations of deployments were subject to site accessibility. All micro-stations were deployed approximately twenty-four hours before the prescribed fire to monitor the background air quality in the area. Figure 1 shows the micro-station deployment locations.

Micro Sensor Systems
Smoke dynamics during and after the prescribed fire was monitored by a network of five field deployable, low-footprint, sensor equipped micro air monitoring systems. These systems, referred to as micro-stations in this paper, were custom made by Alberta Environment and Parks for emergency deployments and remote area monitoring [8]. The shoebox-sized micro-stations can be conveniently transported and placed on tripods at remote forest locations. They are designed to run on solar power, with a battery back-up time of up to 72 h. The micro-stations were equipped with sensors for detection of airborne particulate matter (PM) with aerodynamic diameters of up to 1, 2.5, and 10 µm (PM 1 , PM 2.5 , and PM 10 , respectively). Some micro-stations were also equipped with additional sensors for monitoring of ozone, carbon dioxide, formaldehyde, and volatile organic compound (VOC). All micro-stations had ambient temperature and humidity sensors. Plantower PMS6003 and QS1005 sensors were used for PM detection. These sensors have a manufacturer specified consistency error of ±10 µg/m 3 with resolutions of 1 µg/m 3 . Aeroqual SM50 ozone sensors were used with a resolution of 12-bit for analog signals. Sensor deployment details are provided in Table 1. Table 1. Micro-station deployment details. Distances are measured from an assumed center location of the fire-area, shown in Figure 1. We used a mass balance model to describe the smoke dynamics and its relations to the types of fuel consumption. Smoke dynamics were monitored through measurements of PM 2.5 at the five locations where micro-stations were deployed. Time-series concentration profiles showed occurrences of three smoke wavefronts when sharp enhancements in PM 2.5 concentrations were recorded simultaneously at two or more stations. Decrease of PM 2.5 concentrations at varying rates of decay followed the three smoke wavefronts at all micro-station locations.

Micro-Station Serial
To understand the time-series PM 2.5 concentration profile in relation to smoke dynamics, we start with equating the influx and outflow of PM 2.5 at an imaginary vertical box in a direction perpendicular to smoke propagation at a measurement location (see Figure 2). Neglecting transverse dispersion, and assuming that excess PM 2.5 mass accumulates with uniform density along an effective length of d, the dynamic balance can be represented as, where, q = flux of PM 2.5 in the incoming smoke (µg/m 2 /s), v = propagation velocity of smoke plume wavefront (m/s), n = effective concentration of PM 2.5 within the vertical three dimensional box (µg/m 3 ), ∆n = n(t 2 ) − n(t 1 ) = increase in PM 2.5 concentration within the box during an interval ∆t (µg/m 3 ), ∆t = t 2 − t 1 = time interval (s) A = area of the imaginary cross section at the measurement location (m 2 ), and d = effective length of virtual box where excess PM 2.5 distribution is considered to be uniform (m).  The left hand side of Equation (1) represents total increase of mass within an imaginary air volume with cross section of A and a depth of d beginning from the measurement location to downwind. The two terms in the right hand side represents the inflow and outflow of PM2.5 mass during a time interval t, respectively. At a steady state condition, the two terms in the right hand side would cancel each other, and the resulting increase in concentration will be zero.
For instances when a smoke wavefront has just past through a micro-station location, The rate constant term v/d in Equation (4) in units of (second) -1 is a measure of propagation speed of smoke wavefronts on a relative scale. The corresponding half-life of smoke decay is given as:

Gaussian Profiling of Smoke Dispersion
Smoke generated from the prescribed fire at Pelican Mountain unit 5 were monitored through four micro-stations deployed at distances of 500 m and a fifth micro-station further downwind at 1 km. The four micro-stations deployed in near-field region cover an arc angle of 128 degrees in the downwind and collectively captured the entire smoke plume. Data collected by these four microstations can thus be used to simulate the plume profile.
For each wavefront, a set of peak PM2.5 concentrations at the four locations were fit into a Gaussian profile to simulate the smoke distribution along the arc length of the propagation wavefront. Arc radius for the Gaussian fit was taken as the average distance of individual microstations from an approximate center location of the burn area (see Figure 1). Distances were calculated from geospatial coordinates. The smoke wavefront analysis thus assumed that the observation points are equidistant from the centre location of the fire area and all of the smoke originated from this location. Peak PM2.5 concentrations measured at the three distinct smoke wavefronts where two or more micro-stations observed elevated levels were then fit into Gaussian function in a polar distribution. Details are given in Appendix A. The left hand side of Equation (1) represents total increase of mass within an imaginary air volume with cross section of A and a depth of d beginning from the measurement location to downwind. The two terms in the right hand side represents the inflow and outflow of PM 2.5 mass during a time interval ∆t, respectively. At a steady state condition, the two terms in the right hand side would cancel each other, and the resulting increase in concentration will be zero.
Equation (1) leads to the differential equation, Equation (2) can be solved as [28], where, n = n 0 at time, t = 0. For instances when a smoke wavefront has just past through a micro-station location, q(t) = 0, and Equation (3) converts to, The rate constant term v/d in Equation (4) in units of (second) −1 is a measure of propagation speed of smoke wavefronts on a relative scale. The corresponding half-life of smoke decay is given as:

Gaussian Profiling of Smoke Dispersion
Smoke generated from the prescribed fire at Pelican Mountain unit 5 were monitored through four micro-stations deployed at distances of 500 m and a fifth micro-station further downwind at 1 km. The four micro-stations deployed in near-field region cover an arc angle of 128 degrees in the downwind and collectively captured the entire smoke plume. Data collected by these four micro-stations can thus be used to simulate the plume profile.
For each wavefront, a set of peak PM 2.5 concentrations at the four locations were fit into a Gaussian profile to simulate the smoke distribution along the arc length of the propagation wavefront. Arc radius for the Gaussian fit was taken as the average distance of individual micro-stations from an approximate center location of the burn area (see Figure 1). Distances were calculated from geospatial coordinates. The smoke wavefront analysis thus assumed that the observation points are equidistant from the centre location of the fire area and all of the smoke originated from this location. Peak PM 2.5 concentrations Fire 2020, 3, 30 6 of 17 measured at the three distinct smoke wavefronts where two or more micro-stations observed elevated levels were then fit into Gaussian function in a polar distribution. Details are given in Appendix A.

PM 2.5 Emission from Combustion of Fuels
Emission of PM 2.5 mass from the fire was estimated through calculation of mass flow at ground level during propagation of the smoke-waves. Ground level flow of PM 2.5 mass at peak intensity of a smoke-wave (wavefront) was calculated as: where, Total PM 2.5 mass in a smoke-wave can then be calculated as: where, M PM2.5 = mass of PM 2.5 in smoke-wave, n(t) = PM 2.5 density as a function of time (µg/m 3 ), n(t) max = peak PM 2.5 intensity at the smoke-wave (µg/m 3 ), t 0 = onset of smoke-wave detection at sensor location, and T = duration of smoke-wave recorded at sensor location (s).
Calculation of PM 2.5 mass in smoke-waves and their relation to overall emissions from combustion of fuels are provided in Appendix B.

Background Ambient Conditions
Background ambient conditions were measured for approximately twenty-four hours before the fire. Concentrations of PM 2.5 were low throughout the period of background measurements. Measured PM 2.5 concentrations for the five micro-stations are shown in Figure 3. Some occasional spikes of PM 2.5 concentrations for up to 35 µg/m 3 were recorded for the micro-station located at the north of unit 5. PM 2.5 levels recorded at other locations were negligible, and below the sensor minimum detection level (MDL) in most of the cases. The overall background PM 2.5 level corresponds to good air quality conditions on the site with no nearby emission sources.  Ambient temperature and relative humidity at all micro-station locations showed typical diurnal cycles. Overnight temperature reduced to around 0 C with gradually heating up to 25 C in the early afternoon. Relative humidity reached above 90% around dawn and went down towards 15% in the early afternoon. At the time of the fire, air temperature and humidity were in the shoulder regions of faster evening period changes. The prescribed fire of 3.4 hectares of land did not result in noticeable variations in ambient temperature and relative humidity at locations of sensor deployments (500-1000 m away). However, it is worth mentioning that the prescribed fire occurred at the shoulder period of typical diurnal cycle with rapid changes in ambient temperature and relative humidity before dusk, and as a result, modest changes in ambient parameters resulting from the fire may have been embedded in a stronger diurnal effect. Temperature and relative humidity variations are shown in Figure 4.

Smoke from Fire
The fire was ignited at 17:49:03 local time. Smoke generated from the fire was monitored by tripod mounted micro-stations deployed at downwind locations. PM2.5 concentrations, recorded on a minute resolution, show time variation and spatial distribution of smoke intensities. Strongest smoke intensities were recorded at micro-station μS 303-100 site located 425 m north of the fire area (see Ambient temperature and relative humidity at all micro-station locations showed typical diurnal cycles. Overnight temperature reduced to around 0 • C with gradually heating up to 25 • C in the early afternoon. Relative humidity reached above 90% around dawn and went down towards 15% in the early afternoon. At the time of the fire, air temperature and humidity were in the shoulder regions of faster evening period changes. The prescribed fire of 3.4 hectares of land did not result in noticeable variations in ambient temperature and relative humidity at locations of sensor deployments (500-1000 m away). However, it is worth mentioning that the prescribed fire occurred at the shoulder period of typical diurnal cycle with rapid changes in ambient temperature and relative humidity before dusk, and as a result, modest changes in ambient parameters resulting from the fire may have been embedded in a stronger diurnal effect. Temperature and relative humidity variations are shown in Figure 4.  Ambient temperature and relative humidity at all micro-station locations showed typical diurnal cycles. Overnight temperature reduced to around 0 C with gradually heating up to 25 C in the early afternoon. Relative humidity reached above 90% around dawn and went down towards 15% in the early afternoon. At the time of the fire, air temperature and humidity were in the shoulder regions of faster evening period changes. The prescribed fire of 3.4 hectares of land did not result in noticeable variations in ambient temperature and relative humidity at locations of sensor deployments (500-1000 m away). However, it is worth mentioning that the prescribed fire occurred at the shoulder period of typical diurnal cycle with rapid changes in ambient temperature and relative humidity before dusk, and as a result, modest changes in ambient parameters resulting from the fire may have been embedded in a stronger diurnal effect. Temperature and relative humidity variations are shown in Figure 4.

Smoke from Fire
The fire was ignited at 17:49:03 local time. Smoke generated from the fire was monitored by tripod mounted micro-stations deployed at downwind locations. PM2.5 concentrations, recorded on a minute resolution, show time variation and spatial distribution of smoke intensities. Strongest smoke intensities were recorded at micro-station μS 303-100 site located 425 m north of the fire area (see May

Smoke from Fire
The fire was ignited at 17:49:03 local time. Smoke generated from the fire was monitored by tripod mounted micro-stations deployed at downwind locations. PM 2.5 concentrations, recorded on a minute resolution, show time variation and spatial distribution of smoke intensities. Strongest smoke intensities were recorded at micro-station µS 303-100 site located 425 m north of the fire area (see Figure 1). Moderate levels of smoke intensities were recorded in the northeast (µS 303-300) and northwest (µS 303-200) locations at 567 and 474 m, respectively. A micro-station located at a further distance of 973 m in the northwest (µS 401-200) recorded a smaller spike of PM 2.5 with additional time delay of 4 min. The micro-station located at the west-northwest direction (µS 401-100) at a distance of 529 m from the fire area did not record any elevated level of PM 2.5 during or after the fire, implying that the smoke propagation was confined within northwest to the east. Plots of PM 2.5 concentrations against time recorded at the five micro-stations are shown in Figure 5. Classification: Protected A Figure 1). Moderate levels of smoke intensities were recorded in the northeast (μS 303-300) and northwest (μS 303-200) locations at 567 and 474 m, respectively. A micro-station located at a further distance of 973 m in the northwest (μS 401-200) recorded a smaller spike of PM2.5 with additional time delay of 4 min. The micro-station located at the west-northwest direction (μS 401-100) at a distance of 529 m from the fire area did not record any elevated level of PM2.5 during or after the fire, implying that the smoke propagation was confined within northwest to the east. Plots of PM2.5 concentrations against time recorded at the five micro-stations are shown in Figure 5. PM2.5 time-series data in Figure 5 reveals important information about fuel-fire behaviour during the prescribed fire. Smoke wavefront reached the north micro-station 8 min after the ignition of the fire, and contributed to a sharp rise in PM2.5 concentration. The concentration level increased from a baseline level of less than 5 μg/m 3 to a peak concentration of 940 μg/m 3 in 6 min, followed by a gradual decay of intensity. A moderate rise in PM2.5 level to 110 μg/m 3 with a delay of 14 min from the time of ignition was observed at the northeast location. Spikes in PM2.5 concentrations in the north and northeast locations within 15 min of the fire-ignition are indicative of smoke generation from flaming combustion of canopy fuels. Combustion of canopy fuels occurred during the time when the fire front swept through unit 5 (the burn area) from the ignition line in the south towards the northern perimeter in about six minutes [27]. The timespan of canopy fuel combustion is in agreement with the time period when continued enhancements in PM2.5 concentrations were recorded at the north and northeast micro-stations. Time duration for the smoke to reach at these two locations are measures of propagation speed of the smoke wavefront, referred as wavefront A in Figure 5. Differences in intensity levels in PM2.5 concentrations from the same smoke wavefront at the two locations is due to the spatial distribution of the smoke over its width. Further analysis on smoke propagation and spatial distribution are given in next sections.
The presence of two more smoke wavefronts, occurring long after the canopy fire had ended, are shown in Figure 5 as wavefronts B and C, respectively. The smoke wavefront B appeared at locations at the north and the two northwest locations (near and further) approximately 30 min after the end of the canopy fire. The smoke wavefront C, observed at the north and northeast locations, occurred 45 min after the canopy fire had ended. At the time of observation of smoke wavefronts B and C, only residual smoke was emitting from the burn area, and firefighters were in operation to extinguish the remaining spot fires. However, it is of interest to note that the peak PM2.5 intensities in  [27]. The timespan of canopy fuel combustion is in agreement with the time period when continued enhancements in PM 2.5 concentrations were recorded at the north and northeast micro-stations. Time duration for the smoke to reach at these two locations are measures of propagation speed of the smoke wavefront, referred as wavefront A in Figure 5. Differences in intensity levels in PM 2.5 concentrations from the same smoke wavefront at the two locations is due to the spatial distribution of the smoke over its width. Further analysis on smoke propagation and spatial distribution are given in next sections.
The presence of two more smoke wavefronts, occurring long after the canopy fire had ended, are shown in Figure 5 as wavefronts B and C, respectively. The smoke wavefront B appeared at locations at the north and the two northwest locations (near and further) approximately 30 min after the end of the canopy fire. The smoke wavefront C, observed at the north and northeast locations, occurred 45 min after the canopy fire had ended. At the time of observation of smoke wavefronts B and C, only residual smoke was emitting from the burn area, and firefighters were in operation to extinguish the remaining spot fires. However, it is of interest to note that the peak PM 2.5 intensities in smokes B and C were comparable or exceeded that occurred immediately after the fire (smoke wavefront A).

Smoke Decay Half-Life
The smoke wavefronts in Figure 5 show similarities in their pattern of slow decay of intensities following a relatively sharp increase of PM 2.5 concentrations to the peak level. The decay rate for the three wavefronts, however, show considerable differences among themselves. We fit Equation (4) to each of the smoke waves to characterize their nature of decay, shown in Figure 6. The curve-fitting parameters, standard deviations of the decaying parameter, and uncertainty estimates are given in Table 2. Decay half-life for smoke A, B, and C were calculated to be 9.7 ± 1.7, 2.7 ± 0.5, and 17.8 ± 0.8 min, respectively. An almost two-fold increase of half-life in smoke C compared to that in A is an indication of smaller propagation rate of the former. The small half-life of only 2.7 min for smoke wavefront B is likely due to a rapid shift in wind direction as discussed in the next section. Classification: Protected A smokes B and C were comparable or exceeded that occurred immediately after the fire (smoke wavefront A).

Smoke Decay Half-Life
The smoke wavefronts in Figure 5 show similarities in their pattern of slow decay of intensities following a relatively sharp increase of PM2.5 concentrations to the peak level. The decay rate for the three wavefronts, however, show considerable differences among themselves. We fit Equation (4) to each of the smoke waves to characterize their nature of decay, shown in Figure 6. The curve-fitting parameters, standard deviations of the decaying parameter, and uncertainty estimates are given in Table 2. Decay half-life for smoke A, B, and C were calculated to be 9.7 ± 1.7, 2.7 ± 0.5, and 17.8 ± 0.8 min, respectively. An almost two-fold increase of half-life in smoke C compared to that in A is an indication of smaller propagation rate of the former. The small half-life of only 2.7 min for smoke wavefront B is likely due to a rapid shift in wind direction as discussed in the next section.

Smoke Wavefronts
Spatial distribution of PM2.5 at the three smoke wavefronts were obtained by fitting Gaussian profiles on the micro-station data points (see Materials and Methods). Plume distributions on polar coordinates are shown in Figure 7. It is apparent that predominant downwind directions was northnortheast during smoke wavefronts A and C, with a brief shift of direction towards north-northwest during the wavefront B. The center of the polar plots represents the center of the burn area (see Figure  1). Symbols in the plots show micro-station locations with corresponding PM2.5 readings. Assumption of equal distance of micro-stations from the fire area was taken for a simplified Gaussian fit. Details of the numerical fit and spatial distribution of the smoke plume over arc-angles are shown in Appendix A.

Smoke Wavefronts
Spatial distribution of PM 2.5 at the three smoke wavefronts were obtained by fitting Gaussian profiles on the micro-station data points (see Materials and Methods). Plume distributions on polar coordinates are shown in Figure 7. It is apparent that predominant downwind directions was north-northeast during smoke wavefronts A and C, with a brief shift of direction towards north-northwest during the wavefront B. The center of the polar plots represents the center of the burn area (see Figure 1). Symbols in the plots show micro-station locations with corresponding PM 2.5 readings. Assumption of equal distance of micro-stations from the fire area was taken for a simplified Gaussian fit. Details of the numerical fit and spatial distribution of the smoke plume over arc-angles are shown in Appendix A.

Discussion
Images of the prescribed fire captured from an observation site on the ground at a distance of approximately 200 m east from the burn area are shown in Figure 8. The image on the left was taken immediately after the ignition occurred and smoke became visible from the observation site. Location of smoke in the image represents the ignition line along the south perimeter of unit 5. The image in the middle was taken three minutes after the ignition of the fire. The spread of fire front in the north and generation of intense smoke from flaming combustion is visible in the image. Flames on the canopy are also visible. Vertical lifting of smoke due to the flaming generated intense heat is indicated by an arrow in the image. The image on the right was captured when canopy fire had completed, and smoke generated from smoldering combustions were emitting from the burn area. The smoldering smoke can be characterized by its dense white appearance and a horizontal propagation [15]. Results and analysis of the sensor data implies that the smoke generated during the flaming combustion and the succeeding smoldering phase traveled downwind predominantly in northnortheast directions with a temporary shift in north-northwest (see Figure 7). Deployment of microstations at different distances and at different downwind angles resulted in non-concurrent occurrences of smoke wavefronts. The appearance of smoke wavefronts at the micro-station locations are summarized in Table 3. The propagation speed of smoke wavefronts at the sensor locations are shown in Figure 9.

Discussion
Images of the prescribed fire captured from an observation site on the ground at a distance of approximately 200 m east from the burn area are shown in Figure 8. The image on the left was taken immediately after the ignition occurred and smoke became visible from the observation site. Location of smoke in the image represents the ignition line along the south perimeter of unit 5. The image in the middle was taken three minutes after the ignition of the fire. The spread of fire front in the north and generation of intense smoke from flaming combustion is visible in the image. Flames on the canopy are also visible. Vertical lifting of smoke due to the flaming generated intense heat is indicated by an arrow in the image. The image on the right was captured when canopy fire had completed, and smoke generated from smoldering combustions were emitting from the burn area. The smoldering smoke can be characterized by its dense white appearance and a horizontal propagation [15].

Discussion
Images of the prescribed fire captured from an observation site on the ground at a distance of approximately 200 m east from the burn area are shown in Figure 8. The image on the left was taken immediately after the ignition occurred and smoke became visible from the observation site. Location of smoke in the image represents the ignition line along the south perimeter of unit 5. The image in the middle was taken three minutes after the ignition of the fire. The spread of fire front in the north and generation of intense smoke from flaming combustion is visible in the image. Flames on the canopy are also visible. Vertical lifting of smoke due to the flaming generated intense heat is indicated by an arrow in the image. The image on the right was captured when canopy fire had completed, and smoke generated from smoldering combustions were emitting from the burn area. The smoldering smoke can be characterized by its dense white appearance and a horizontal propagation [15]. Results and analysis of the sensor data implies that the smoke generated during the flaming combustion and the succeeding smoldering phase traveled downwind predominantly in northnortheast directions with a temporary shift in north-northwest (see Figure 7). Deployment of microstations at different distances and at different downwind angles resulted in non-concurrent occurrences of smoke wavefronts. The appearance of smoke wavefronts at the micro-station locations are summarized in Table 3. The propagation speed of smoke wavefronts at the sensor locations are shown in Figure 9.  Results and analysis of the sensor data implies that the smoke generated during the flaming combustion and the succeeding smoldering phase traveled downwind predominantly in north-northeast directions with a temporary shift in north-northwest (see Figure 7). Deployment of micro-stations at different distances and at different downwind angles resulted in non-concurrent occurrences of smoke wavefronts. The appearance of smoke wavefronts at the micro-station locations are summarized in Table 3. The propagation speed of smoke wavefronts at the sensor locations are shown in Figure 9.     The apparent differences in propagation speeds of smoke wavefronts, irrespective of their distances from the fire area, signify the fact that smoke propagation dynamics for wavefront A is different from those for B and C. Smoke plume A occurred shortly after the fire ignition and propagated to the north and northeast locations at a faster rate, with travel times of 8 and 14 min, respectively. The shorter half-life associated with the smoke as calculated in Section 3.3 is consistent with a faster propagation rate for the smoke wavefront. The rise time of PM2.5 concentrations for smoke A at the location in the north is in agreement with the duration of the fire front sweeping through the burn area. These facts suggest the origin of plume A to be the flaming combustion where canopy fuels were predominantly consumed. The intense heat generated by the flaming combustion phase of fire is likely to add convective effects in the plume propagation.
The smoke plumes referred as B and C show much lower propagation speeds as compared to the initial plume A. These later occurring plumes are attributed to the smoke created in smoldering phase of the fire, where mainly ground fuels were the contributors. Due to the absence of intense heat generated from combustion, smoke generated in this phase are expected to be propagating predominantly on principles of advection-dispersion. This is indeed observed in Figure 9, where lower propagation speeds of plumes B and C are seen irrespective of their travel distances. The calculated value of longer half-life for smoke C in Section 3.3 signifies the fact of advection-dispersion dominated slow propagation. A small value of half-life in plume B in Section 3.3, as mentioned previously, is attributed to the shift of wind direction (see Figure 7), thereby causing a faster decay compared to an advection-dispersion assisted mechanism.
We calculated total PM2.5 emissions from canopy and surface fuels during flaming and smoldering phases of the fire, respectively, using a mass balance model (see Section 2.3.3). Assuming a smoke plume height of 50 m and a uniform PM2.5 distribution vertically, we calculate the peak flow Speed (m/s) The apparent differences in propagation speeds of smoke wavefronts, irrespective of their distances from the fire area, signify the fact that smoke propagation dynamics for wavefront A is different from those for B and C. Smoke plume A occurred shortly after the fire ignition and propagated to the north and northeast locations at a faster rate, with travel times of 8 and 14 min, respectively. The shorter half-life associated with the smoke as calculated in Section 3.3 is consistent with a faster propagation rate for the smoke wavefront. The rise time of PM 2.5 concentrations for smoke A at the location in the north is in agreement with the duration of the fire front sweeping through the burn area. These facts suggest the origin of plume A to be the flaming combustion where canopy fuels were predominantly consumed. The intense heat generated by the flaming combustion phase of fire is likely to add convective effects in the plume propagation.
The smoke plumes referred as B and C show much lower propagation speeds as compared to the initial plume A. These later occurring plumes are attributed to the smoke created in smoldering phase of the fire, where mainly ground fuels were the contributors. Due to the absence of intense heat generated from combustion, smoke generated in this phase are expected to be propagating predominantly on principles of advection-dispersion. This is indeed observed in Figure 9, where lower propagation speeds of plumes B and C are seen irrespective of their travel distances. The calculated value of longer half-life for smoke C in Section 3.3 signifies the fact of advection-dispersion dominated slow propagation. A small value of half-life in plume B in Section 3.3, as mentioned previously, is attributed to the shift of wind direction (see Figure 7), thereby causing a faster decay compared to an advection-dispersion assisted mechanism. We calculated total PM 2.5 emissions from canopy and surface fuels during flaming and smoldering phases of the fire, respectively, using a mass balance model (see Section 2.3.3). Assuming a smoke plume height of 50 m and a uniform PM 2.5 distribution vertically, we calculate the peak flow rate of PM 2.5 at the smoke wavefront of flaming combustion (smoke-wave A) as 2.26 × 10 7 µg/s. Our assumption of height and plume uniformity is based on aerial photographs from a helicopter (see Appendix C). Vertical profiling measurements through drones or tall towers (not available) may result in better precision in estimations in future experiments. The peak flow rate, when integrated for the duration of the plume, provides the total PM 2.5 emission from crown fuels. Using Equation (7), the total PM 2.5 mass from combustion of crown fuels that is present in the ground level plume is 15.2 kg, yielding a mass density of~4.5 kg/ha. The calculated value, however, does not account for the portion of the flaming smoke that is lifted at higher atmospheric levels due to intense heat assisted buoyancy (see Figure 8). Calculated value for smoldering combustion of surface fuels (smoke-waves B and C) represents a total PM 2.5 emission of 16.3 kg from the fire area of 3.4 hectares (~4.8 kg/ha). Details of the PM 2.5 emission calculations from combustion of fuels are provided in Appendix B.
Results in our study show that ground level PM 2.5 concentrations in near-field areas of a wildland fire have strong spatial distributions. Fire originated plume resulted in concentration variations on the order of 1000 µg/m 3 within spatial distances of 500 m. Direction of wind variation may also result in enhancement or depletion of particulates at downwind locations. Similar spatial-temporal variations for other pollutants are expected. Fuel type and the phase of the fire plays important roles in ground level pollutants. Effects from flaming combustion of canopy fuels are expected to have immediate but relatively shorter term effects in near-field areas of fire. Plumes generated during this combustion phase are likely to be aloft at higher atmospheric levels and contributing to long-range transport of pollutants (see Figure 8). Smoke created by the smoldering phase of fires, mostly by the surface fuels, on the other hand, are shown to have a slower ground level propagation and are likely to result in sustained enhancements in particulate levels in ambient air. Although fuel loading from ground fuels contributing to smoldering in our study is estimated to be only 35% of total fuel [27], they contributed to higher particulate concentrations in air in downwind locations and for longer durations of time. This may be a key consideration in cases of wildland fires or preventive prescribed fires that may occur in close vicinity of communities.

Conclusions
We have analyzed the dynamics of smoke propagation in a prescribed wildland fire at Pelican Mountain, central Alberta. A network of five field deployable micro-sensor systems were used to measure near-field real-time smoke intensities. Our analysis identifies differences in propagation and dispersion characteristics of smoke generated from flaming and smoldering phases of combustion.
Smoke created from combustion of canopy fuels showed propagation rate of~0.8 m/s and a shorter presence in the near-field region of the fire area. Smoke decay half-life of 9.7 ± 1.7 min was estimated for the flaming phase of combustion. The smoldering phase of the fire contributed by ground fuels, on the other hand, were characterized by a slower propagation rate of~0.2 m/s, and showed prolonged existence in the nearby region well after the end of the intense canopy fire. Decay half-life of smoke from smoldering phase was estimated to be 17.  Funding: This research was funded in part (development of field deployable micro sensor systems) by Alberta Environment and Parks Innovation Fund, grant number 069A1517. Acknowledgments: Q.H. wants to thank Bob Myrick at Alberta Environment and Parks for his support on this work; Matthew Parsons at Environment and Climate Change Canada for providing information and helpful discussions; and Wendell Pozniak at Alberta Agriculture and Forestry for deployment support. Bigstone Cree First Nation provided crews to do the thinning work, essential firefighting staff, and community support for the project.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Appendix A Smoke Wavefront Profiling
Smoke wavefronts were modeled through a Gaussian fit using the equation: (A1) Fitting parameters are shown in Table A1. Modeled PM 2.5 concentration profiles overlaid on measurement data are shown in Figure A1. First Nation provided crews to do the thinning work, essential firefighting staff, and community support for the project.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Appendix A. Smoke Wavefront Profiling
Smoke wavefronts were modeled through a Gaussian fit using the equation: . .

(A1)
Fitting parameters are shown in Table A1. Modeled PM2.5 concentration profiles overlaid on measurement data are shown in Figure A1.

Appendix B. PM2.5 Emission from Combustion of Fuels
Flow rate of PM2.5 at ground level are calculated using (6) as: where, vmean is the mean propagation velocity of smoke observed at two sensor locations. Profiles in Figure A1 were used to calculate the integrated PM2.5 mass flow at the smoke wavefront. Mean arc radius was taken as 496 m. Mass of PM2.5 at smoke-waves are calculated using Equation (7) (Tables A2-A4). Total emission of PM2.5 based on combustion phases (fuel types) are summarized in Table  A5.

Appendix B PM 2.5 Emission from Combustion of Fuels
Flow rate of PM 2.5 at ground level are calculated using (6) as: where, v mean is the mean propagation velocity of smoke observed at two sensor locations. Profiles in Figure A1 were used to calculate the integrated PM 2.5 mass flow at the smoke wavefront. Mean arc radius was taken as 496 m. Mass of PM 2.5 at smoke-waves are calculated using Equation (7) (Tables A2-A4). Total emission of PM 2.5 based on combustion phases (fuel types) are summarized in Table A5.

Appendix C Estimations of Uncertainties
Smoke propagation rate: Smoke propagation rates for wavefronts A, B, and C at the three sensor locations were calculated from assumption of smoke being originated from a center location of the fire area (see Figure 1). In addition, temporal resolutions of sensors were on the order of 1 min. Uncertainties introduced by these two parameters were estimated as: where, U l and U t are uncertainties introduced due to distance and time of smoke propagation, respectively; and U Total is the overall uncertainty in propagation speed estimations. Velocities v l+∆l and v t+∆t correspond to cases where distance and time of smoke travel are considered as l + ∆l and t + ∆t, respectively. Uncertainties in parameters are shown in Table A6, and the resulting uncertainties in propagation velocity for the three sensor locations are given in Table A7.  The existence of smoke in near-filed locations of the fire area, and directions of their propagation during and after fire occurrence were observed through ground level and aerial photographs. An aerial photograph at 18:31 local time (~40 min after flaming combustion was complete) is shown in Figure A2. Direction of propagation of smoke in the Figure is in agreement with our model analysis for smoke wavefront C. Based on the laminar nature of flow of smoke in the near-field region of fire area in Figure A2 and on-site observations (see Figure 8c), we estimated the plume height from smoldering combustion to be twice the height of the canopy (2 × 25 m= 50 m). A similar estimate was taken for ground propagating component of flaming smoke for consistency. It is worth mentioning that, accurate estimation of emissions from combustion through our model would require a comprehensive vertical profiling of smoke distribution through drone or LiDAR based measurements (out of scope of this experiment due to resource and logistical needs). The emission values calculated from Appendix B thus represent preliminary estimations for future work. To our belief, the emission estimates in our work represent conservative values, and may underestimate actual fuel emissions by up to 50% (plume height of 100 m would represent a two fold increase of present emission estimates). The existence of smoke in near-filed locations of the fire area, and directions of their propagation during and after fire occurrence were observed through ground level and aerial photographs. An aerial photograph at 18:31 local time (~40 min after flaming combustion was complete) is shown in Figure A2. Direction of propagation of smoke in the Figure is in agreement with our model analysis for smoke wavefront C. Based on the laminar nature of flow of smoke in the near-field region of fire area in Figure A2 and on-site observations (see Figure 8c), we estimated the plume height from smoldering combustion to be twice the height of the canopy (2 × 25 m = 50 m). A similar estimate was taken for ground propagating component of flaming smoke for consistency. It is worth mentioning that, accurate estimation of emissions from combustion through our model would require a comprehensive vertical profiling of smoke distribution through drone or LiDAR based measurements (out of scope of this experiment due to resource and logistical needs). The emission values calculated from Appendix B thus represent preliminary estimations for future work. To our belief, the emission estimates in our work represent conservative values, and may underestimate actual fuel emissions by up to 50% (plume height of 100 m would represent a two fold increase of present emission estimates).