Abstract

The coupled fire–atmosphere model consisting of the Weather and Forecasting (WRF) Model coupled with the fire-spread model (SFIRE) module has been used to simulate a bushfire at D’Estrees Bay on Kangaroo Island, South Australia, in December 2007. Initial conditions for the simulations were provided by two global analyses: the GFS operational analysis and ERA-Interim. For each NWP initialization, the simulations were run with and without feedback from the fire to the atmospheric model. The focus of this study was examining how the energy fluxes from the simulated fire modified the local meteorological environment. With feedback enabled, the propagation speed of the sea-breeze frontal line was faster and vertical motion in the frontal zone was enhanced. For one of the initial conditions with feedback on, a vortex developed adjacent to the head fire and remained present for over 5 h of simulation time. The vortex was not present without fire–atmosphere feedback. The results show that the energy fluxes released by a fire can effect significant changes on the surrounding mesoscale atmosphere. This has implications for the appropriate use of weather parameters extracted from NWP and used in prediction for fire operations. These meteorological modifications also have implications for anticipating likely fire behavior.

1. Introduction

The D’Estrees fire was one of several fires ignited by lightning on Kangaroo Island in early December 2007. Figure 1 shows the island’s location, off the coast of southern Australia. Figure 2 shows the fire activity on 8 December. Figure 2 shows four fires burning on the island; however, the D’Estrees fire, at the southeast corner of the island, has a smoke plume that is very different from the other three fires: the plume is much larger and more opaque. This event is of interest because the smoke indicates increased fire activity at the D’Estrees fire relative to the three other fires burning nearby. On the MODIS satellite image of the previous afternoon (not shown), the smoke plumes from each of the four fires were similar in size and opacity.

Fig. 1.

Map of southern South Australia showing location of nested domains used in the WRF simulations over Kangaroo Island. Color shading is topography (m).

Fig. 1.

Map of southern South Australia showing location of nested domains used in the WRF simulations over Kangaroo Island. Color shading is topography (m).

Fig. 2.

NOAA MODIS Aqua satellite image from 1535 LT 8 Dec 2007. Red pixels show active fire area.

Fig. 2.

NOAA MODIS Aqua satellite image from 1535 LT 8 Dec 2007. Red pixels show active fire area.

The heightened fire activity over the D’Estrees fire was explored in the case study by Peace and Mills (2012), who proposed that the increased fire activity at the D’Estrees fire on this day was due to it being located in a sea-breeze convergence zone, and that interactions between the fire and atmosphere played an important role in producing the extensive smoke plume. This study explores the fire and atmosphere interaction processes by examining how the energy released by the D’Estrees fire modified the surrounding atmosphere. To address the subject of fire–atmosphere feedback by examining observational data is challenging because of the difficulties involved in making comprehensive meteorological observations at a fire ground and of making controlled experiments. However, an alternative approach is examining the results of high-resolution coupled fire–atmosphere simulations, as these provide an opportunity for examining how a fire alters the structure of the surrounding atmosphere. The aim of the current study is to use numerical simulations to better understand fire and atmosphere interactions in the context of the D’Estrees fire.

The premise underlying this study is that a fire can modify the dynamical structure of the surrounding atmosphere. This is consistent with observations from fire grounds, which frequently report local weather in the vicinity of the fire as different from prevailing conditions in the broader area. This modification of local meteorology due to the fire’s influence includes elevated temperatures, as well as variation in both speed and direction of wind in the vicinity of the fire, along with pyrocumulus (or pyrocumulonimbus) cloud formations. The modification to local meteorology is driven by heat and water vapor released in the combustion process generating increased buoyancy in the near-fire environment. Potter (2005) describes interactions between a fire’s convection column and the surrounding atmosphere.

An extreme example of dynamical interactions between a fire and the surrounding atmosphere is the fire tornado that developed at the 2003 Canberra bushfires. The event is documented by McRae et al. (2013) and simulated in highly idealized form by Cunningham and Reeder (2009). The D’Estrees fire simulated in this study is a much smaller fire than the Canberra event, burning in more moderate meteorological conditions. These simulations are an example of how a fire can modify the local meteorology in a realistic (as opposed to idealized) atmosphere, and they provide insights into dynamic interaction processes that may occur during a real event.

This study uses the coupled Weather Research and Forecasting (WRF) Model and fire-spread model (SFIRE) module, described in detail by Mandel et al. (2011). WRF and SFIRE couple the WRF Model with an implementation of the Rothermel (1972) fire-spread equations. Coupled fire–atmosphere models have been used in a number of studies to show that dynamical feedback processes, in particular the fire-modified winds, have an important influence on how a fire perimeter evolves. Coupled simulations that examine the influence of the fire-modified winds include Clark et al. [1996; using the Coupled Atmosphere–Wildland Fire–Environment (CAWFE) model, the predecessor to WRF and SFIRE], Coen (2005), Coen et al. (2013), Coen and Riggan (2014), and Kochanski et al. (2013a), all using model releases of WRF and SFIRE or wildland fire module WRF-Fire.

The focus of this study is somewhat different from previous coupled simulations, as the emphasis is on examining how the fire modifies the surrounding mesoscale atmosphere, rather than assessing how the inclusion of fire–atmosphere feedback alters the predicted fire perimeter. The main objective of this study was to explore how the energy fluxes from the fire affected the surrounding atmosphere. Our simulated fire spread was intended to be a reasonable approximation of actual fire spread in order for the energy release to be realistic. However, we did not aim to reproduce observations of fire perimeter or provide a detailed verification, since observations of the event are very limited, thereby restricting verification possibilities. Although the scope of this study allows only limited validation against real data, the WRF and SFIRE model, using Rothermel’s model coupled to a high-resolution NWP model, gave useful insights into dynamical interactions between the fire and the atmosphere.

This paper documents the first of a series of three WRF and SFIRE simulations of Australian fire events. The two companion studies are of the Rocky River and Layman fires. The meteorology of the events is described in the case studies by Peace and Mills (2012) and Peace et al. (2012). These three fires were selected because unusual fire behavior occurred at each one. Coupled simulations have been used to examine the interactions between the fire and surrounding environment and to test the hypothesis developed in the meteorological case studies that fire–atmosphere interactions are a mechanism for producing unexpected fire behavior.

We proceed by describing the WRF and SFIRE model and the configuration used in our simulations, followed by the results. In the discussion we focus on the simulation results of development of a vortex and relocation of a sea-breeze wind change. To conclude, we discuss implications for fire weather forecasts and development of fire simulation models, as well as considering some of the limitations of our approach.

2. WRF and SFIRE

Several models that integrate a fire simulation model with an atmospheric model are in use, including Méso-NH–“ForeFire” (Filippi et al. 2011), “FIRETEC” (Linn et al. 2002), and Wildland-Urban Fire Dynamics Simulator (WFDS; Mell et al. 2010). The coupled model used in this study is WRF and SFIRE, chosen as it is accessible and well supported, and it provides the opportunity to examine in detail the meteorological dynamics of a fire environment. The WRF Model, described by Skamarock et al. (2008), is a numerical weather prediction model with a range of chemistry and physics extensions and user-defined initialization options. WRF is widely used for meteorological operational and research purposes. The SFIRE model is a two-dimensional fire-spread model coupled to WRF, described by Mandel et al. (2011). The simulations shown here were made with the most recent available model release, which at the time was WRF and SFIRE, released by the SFIRE group and sourced in mid-2012 (the link that was used is no longer active; the current version is https://github.com/jbeezley/wrf-fire).

A detailed description of how the fire model and atmospheric model interact is given in Mandel et al. (2011). In summary, at each time step of the atmospheric simulation, fire progression and fuel burnt are calculated from the atmospheric winds and predetermined fuel parameters [default fuel inputs for the United States are from the Anderson (1982) fuel model]. Fire progression is calculated by a level-set implementation of the Rothermel (1972) equations. From the fire progression at each time step, heat and moisture (latent heat) fluxes are calculated from the quantity of fuel consumed at each grid cell and are converted to potential temperature and water-vapor concentration source terms in the atmospheric model. The calculated values are inserted into the lower levels of the atmospheric grids, using an exponential decay with height. Including the heat and moisture fluxes drives the coupling because, at the next time step of the WRF model, the atmospheric (wind) fields respond to the energy released by the fire.

The prototype for coupled simulations such as these is the CAWFE model, described by Clark et al. (1996). Clark et al. (2004) showed how the elliptical shape of a wind-driven fire is formed by the modification to wind strength and direction arising from fire–atmosphere interaction. This key finding established the importance of the interdependence of fire–atmosphere processes, as well as the value of resolving these processes in fire simulation models.

Recent simulations using WRF and SFIRE include the Big Elk fire (Coen 2005), which showed how a fire modifies the local wind flow and the resultant shape of the fire area. Our work follows the approach of Coen by running simulations with and without fire–atmosphere feedback. Simulations of the 2007 Santa Ana fires (Kochanski et al. 2013a) showed that high-resolution coupled modeling is possible at speeds useful for real-time forecasting. Kochanski et al. (2013b) describe a verification of WRF and SFIRE against data from the FireFlux experiment, and propose avenues for future testing with the ultimate objective of developing WRF and SFIRE as an operational tool. Other (less detailed) simulations in the literature include Meadow Creek, Colorado (Beezley et al. 2010), and Harmanli, Bulgaria (Jordanov et al. 2012). Simpson et al. (2013) describe idealized WRF-Fire simulations examining anomalous fire propagation on lee slopes. Coen et al. (2013) describe a series of experiments using WRF-Fire to test how fire perimeter is affected by varying wind and fuel inputs.

Our simulations were initialized with the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) and European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim) data (Dee et al. 2011). Simulations were run in both feedback “on” and feedback “off” mode, with the difference being the inclusion of energy fluxes from the fire to the atmosphere. By comparing simulations with feedback on and off, it is possible to assess how the fire changes the meteorological environment.

The numerical integration ran for 24 h from 1200 UTC 7 December 2007 to 1200 UTC 8 December 2007 (2230 to 2230 local time). The simulation ran with four nested domains. The outer domain covered a 600 km × 600 km area with a grid spacing of 6 km. The inner domain covered a 22 km × 22 km area with a grid spacing of 222 m. The ratio of grid spacing on the four nests was 1:3:9:27. The fire grid was refined on the inner atmospheric grid at a ratio of 1:10, resulting in a surface fire grid of 22 m. Each of the four nests had 100 grid points in the north–south and east–west directions. Fifty-one sigma levels were included in the vertical, with higher resolution near the surface. The outer nest time step was 36 s, with a 1:3:9:18 ratio giving a time step of 2 s on the inner nest, which was sufficient to maintain numerical stability and be computationally efficient. Following Hu et al. (2010) the Yonsei University boundary layer scheme was used, in combination with the Monin–Obukhov surface physics scheme and Dudhia shortwave radiation. Vertical velocity damping was not used, and the third-order turbulence and mixing option was used. No cumulus physics was included, and the Noah land surface model was used.

The simulated fire ignition was at 2200 UTC 7 December 2007 (0830 local time). The ignition time was chosen to allow a 10-h lead time from model start and to be prior to daytime boundary layer activity as well as capture afternoon fire activity. The fuel parameter values for mallee were taken from McCaw (1998), which describes a study of experimental burns in mallee-heath fuel in Western Australia. This experimental analysis evaluated a range of fire-spread prediction models, including the Rothermel model, and the mallee values from that study are the fuel parameters used here as follows: ground fuel moisture 0.10 kg kg−1, initial mass loading of surface fuel 1.15 kg m−2 (18 t ha−1), fuel depth 0.9 m, surface-area-to-volume ratio 3000 m−1, fuel moisture content of extinction 0.16 kg kg−1, and weighting parameter 180.

The Department of Environment, Water and Natural Resources (DEWNR) provided high-resolution topographic grids at 25 m and high-resolution vegetation grids at 10 m as well as available fire-spread data. The topography in the area is relatively low relief, limiting the influence of the slope component of the Rothermel equations. The vegetation grids provided by DEWNR showed homogeneous fuels across the fire area, so a constant fuel was used in the simulations. However, this homogeneity is not reflected in satellite imagery, which shows discontinuities in fuel, including a local lagoon. Despite the discrepancy, a homogenous fuel was used, as a more detailed dataset was not available and the limitation of homogeneity does not strongly impact the purpose of this study.

Fire-perimeter maps of the event were provided by DEWNR; however, they were based on very limited available data at 24-h intervals. In addition, fuel breaks were not mapped (chained fuel breaks were created by dragging a heavy chain between two tractors). Consequently, the simulations are unable to be rigorously verified against observed fire-spread data. The fire-ignition line in the simulation is similarly approximate. In the simulation, fire ignition was set as a line approximately 3.5 km long and 30 m wide, with an initial forward rate of spread of 0.1 m s−1. In preliminary simulations, fire spread was much too fast in comparison to the limited verification data and reasonable estimates of fire spread through the mallee fuels. Therefore, the wind reduction factor used to estimate the midflame height was set at 0.2 (wind-rf) to achieve fire spread that was a reasonable match for the known fire spread, allowing for the noninclusion of suppression activities in the simulation. The wind-rf is highly dependent on fuel structure and has had limited testing for Australian conditions, so precise values are unknown for the fuels in this study. The value of 0.2 lies within the range tabulated by Baughman and Albini (1980) but is lower than the standard U.S. fuel models in SFIRE. A constant value of fuel moisture of 0.10 was used, based on advice given by DEWNR. As the objective of this study was to explore fire–atmosphere interactions, rather than assess the ability of the model to reproduce the observed fire perimeter, the limited verification is not considered to be of major consequence.

In addition to the results shown here, three simulations were run with small changes to test the sensitivity of the results to our configuration. The tests were run with ERA-Interim initialization. These tests were 1) fire wind scaling factor (wind-rf) reduced from 0.20 to 0.18, 2) fire start time earlier by 1 h, and 3) fire-ignition line length reduced by half. These changes modified the eventual fire spread; however, the interactions seen between the fire and the meteorological conditions remained similar. In particular, the major fire-induced circulations that are discussed in the results section were not sensitive to these small changes.

3. Results

a. Meteorology of the WRF simulations

The meteorological conditions over Kangaroo Island on the day of the D’Estrees fire are described in Peace and Mills (2012). Synoptic forcing was weak, with a ridge of high pressure and a weak embedded cold front over the southern ocean. The northern extent of the cold front crossed the island during the afternoon as an enhanced sea-breeze wind shift. Because of the weak synoptic forcing, mesoscale meteorological processes such as sea breezes had a strong influence during the event.

The closest weather station to the fire site is Kingscote, located approximately 25 km to the north (see Fig. 3 for location, note Kingscote is in nest 2). Observed and simulated weather parameters at Kingscote are shown in Fig. 4 for both the GFS and ERA-Interim initializations. Temperature was well forecast for both initializations, although dewpoint temperature was high in both models. Winds were generally well forecast; however, the timing of the southeasterly wind shift at Kingscote was late in both simulations. The earlier timing of the ERA-Interim wind shift was a better match for observations, but wind direction in the wake of the wind shift was better matched by the GFS (see Fig. 4). Figure 4 shows data from nest 2 for the feedback-off simulations. The feedback-on simulations showed near-identical temperature and wind for nests 1 and 2 at Kingscote, as the influence of the fire modification on the local meteorological conditions did not extend sufficiently far from the fire site. In general, the influence of the fire extended through the inner nest 4 and into, but not throughout, nest 3.

Fig. 3.

Wind structure of the sea-breeze convergence zone. The 10-m winds (m s−1) from the feedback-off simulation over nest 2 are shown for 0200 UTC for (top) ERA-Interim and (middle) GFS. (bottom) ERA-Interim winds across the island, with northwesterly winds at the other three fires and convergence over the D’Estrees fire. The red box in the bottom panel shows the geographical reference for the two top panels.

Fig. 3.

Wind structure of the sea-breeze convergence zone. The 10-m winds (m s−1) from the feedback-off simulation over nest 2 are shown for 0200 UTC for (top) ERA-Interim and (middle) GFS. (bottom) ERA-Interim winds across the island, with northwesterly winds at the other three fires and convergence over the D’Estrees fire. The red box in the bottom panel shows the geographical reference for the two top panels.

Fig. 4.

Comparison of Kingscote automatic weather station and WRF (nest 2 includes Kingscote) from 0000 LT 8 Dec to 0000 LT 9 Dec (the WRF run ended at 2230 LT 8 Dec).

Fig. 4.

Comparison of Kingscote automatic weather station and WRF (nest 2 includes Kingscote) from 0000 LT 8 Dec to 0000 LT 9 Dec (the WRF run ended at 2230 LT 8 Dec).

Figure 3 shows the wind structure at 10 m over the island at approximately the same time as the MODIS image in Fig. 2. The bottom panel of Fig. 3 shows northwesterly winds over most of the island, with a southerly sea breeze over southern parts of the island. One likely reason for the enhanced fire plume at the D’Estrees fire is its location in the convergence zone between the northwest winds and the sea breeze from the south. In contrast, the three other fires were in a steady northwesterly airstream. The convergence extended above 850 hPa, with northwesterly flow at 700 hPa, providing a convergent depth of approximately 1800 m. The presence and location of the convergence zone provide very different meteorological conditions at the D’Estrees fire, in comparison with the three other fires on the island.

The structure of the sea-breeze wind shift was different in the ERA-Interim and GFS simulations, as seen in the top two panels of Fig. 3. The ERA-Interim initialization had lighter winds post–wind shift; also the wind shift was more complex, with a distinct southwest and southeast flow direction on either side of Cape Gantheaume. It is likely this difference in wind structure was the reason a vortex developed in the ERA-Interim initialization, but not the GFS.

b. Fire–atmosphere interactions in the WRF and SFIRE simulations

Figure 5 shows the difference in fire spread resulting from two initializations and two feedback settings: ERA-Interim and GFS, feedback on and feedback off. The feedback-off fire spread should be interpreted with caution, since, as Coen et al. (2013) point out, simulating a fire without feedback gives an unrealistic physical representation of fire spread. This is because a physically nonsensical solution to the fire spread is produced when no mechanism for the known fire shape is included. Notwithstanding this, it is useful to consider the difference that feedback produces in output, since it shows how the fire-perimeter contours are affected by the coupling process, both in shape and in final area. Importantly, comparison of runs with feedback on and feedback off allows the fire's effect on the meteorological conditions to be assessed, although in these simulations we are limiting the coupling process to sensible and latent heat and not considering particulate matter such as smoke and other aerosols.

Fig. 5.

Hourly contours of fire-perimeter spread for two initializations and two feedback settings on the inner nest (nest 4) from 2230 to 1130 UTC. (top left) ERA-Interim with feedback on, (bottom left) ERA-Interim with feedback off, (top right) GFS with feedback on, and (bottom right) GFS with feedback off. Shading shows topography contours (m).

Fig. 5.

Hourly contours of fire-perimeter spread for two initializations and two feedback settings on the inner nest (nest 4) from 2230 to 1130 UTC. (top left) ERA-Interim with feedback on, (bottom left) ERA-Interim with feedback off, (top right) GFS with feedback on, and (bottom right) GFS with feedback off. Shading shows topography contours (m).

The difference in fire area in Fig. 5 shows the sensitivity of fire spread to initial conditions and to inclusion of fire–atmosphere feedback. This result, obtained from initial conditions using two NWP datasets, shows the limitation of a deterministic approach to simulating wind-driven fire spread. For the feedback-on runs, the convergent fire-modified winds in the early stages focus the active head fire to a point and accelerate the forward spread. Post–wind shift, this opens up a fire flank of 8–10 km along the northeastern edge of the fire area. Because the northeast flank becomes the new head fire post–wind shift, it results in a longer fire front on the new active flank. A greater overall fire area is produced in the simulation from the GFS initial condition due to it having stronger southwesterly winds than the ERA-Interim (see Fig. 3).

The frontal wind shift seen in the simulations provides another hypothesis explaining the enhanced smoke plume over the D’Estrees fire in comparison to the three other fires burning on the island. Since the southwest wind shift produced a long active flaming front (up to 10 km long) on the northeastern flank of the fire, this would result in a larger overall fire area and therefore increased smoke production (particularly if attributed to the GFS run due to the stronger southwesterly winds).

It is interesting to note that the burnt area ratio (GFS/ERA-Interim) changes from 1.5:1 for feedback on to 1.18:1 for feedback off (with feedback on GFS burns 50% more; without feedback, GFS burns 18% more). As a comparison, Figs. 7 and 8 in Coen (2005) show a very similar total fire area for the feedback-on and feedback-off simulations. The difference between our simulations and those of Coen is the presence of a change in wind direction, which produces a longer fire front and results in an increase in total fire area.

c. Sea-breeze front modification

Figure 6 shows a significant finding from our simulations: the relocation of the sea-breeze frontal line in response to the fire. In both the GFS and ERA-Interim runs, the southeast wind shift line accelerated into the head fire for the feedback-on simulations. Although the times shown in Fig. 6 vary by an hour because of the frontal timing difference between the GFS and ERA-Interim, relocation of the front occurs in both. This is despite the two global models having differing frontal structure, wind strength, and timing. The change in arrival time of the wind shift between the feedback-on and feedback-off simulations produces a distance shift on the order of 2 km, with an arrival-time difference of approximately 30 min.

Fig. 6.

Wind vectors and divergence (shaded; s−1) for ERA-Interim at 0110 UTC and GFS at 0210 UTC. The fire area is outlined in red. The green line provides a stationary reference for the position of the front.

Fig. 6.

Wind vectors and divergence (shaded; s−1) for ERA-Interim at 0110 UTC and GFS at 0210 UTC. The fire area is outlined in red. The green line provides a stationary reference for the position of the front.

Figure 7 shows a cross section of wind structure through the sea-breeze front and fire plume. The top panel shows the fire plume (near 0 m), with a strong ascent plume extending to 1800 m, with upward velocities of approximately 5 m s−1 in the plume. The circulation in the approaching sea-breeze front is evident between −4000 and −8000 m. The bottom panel shows that without feedback, the sea-breeze front is near −6000 m, approximately 2000 m farther south than for feedback on. In addition to the slower arrival time, the feedback-off simulation has weaker vertical wind speeds in the sea-breeze front updraft. Figure 7 also shows the presence of horizontal convective rolls in the boundary layer in the northwesterly flow. The convective rolls are regularly spaced in the bottom panel (feedback off); however, the regular structure is distorted by the fire plume in the top panel (feedback on). In the top panel, the region between the sea-breeze front and fire plume (between −1000 and −3000 m) has a near-zero wind component from north to south.

Fig. 7.

Cross section of υ, w component of wind (vectors) and w component alone (shading; m s−1) for ERA-Interim (top) feedback on and (bottom) feedback off at 0100 UTC. The left-to-right vertical cross section is taken from south to north along the blue line that is shown in the inset. The red line in the inset shows the fire perimeter. The red dot in the top panel shows the position of the fire front.

Fig. 7.

Cross section of υ, w component of wind (vectors) and w component alone (shading; m s−1) for ERA-Interim (top) feedback on and (bottom) feedback off at 0100 UTC. The left-to-right vertical cross section is taken from south to north along the blue line that is shown in the inset. The red line in the inset shows the fire perimeter. The red dot in the top panel shows the position of the fire front.

Figure 8 is similar to Fig. 7, but for potential temperature. The thermal gradient is enhanced for the feedback-on case, with cooler temperatures in the advancing sea-breeze density current and warmer temperatures in the fire plume. Figure 9 shows the temperature difference between the feedback-on and feedback-off cases, with approximately 1°C of warming in the feedback-on simulation between 1000 and 1500 m elevation, 2000–3000 m downstream from the fire plume. The warm tongue extends several kilometers southward from the fire plume at a height near the top of the mixed layer.

Fig. 8.

As in Fig. 7, but with potential temperature (color shading; K) in place of the w wind component.

Fig. 8.

As in Fig. 7, but with potential temperature (color shading; K) in place of the w wind component.

Fig. 9.

Cross section of the difference in potential temperature (shading; K) between the feedback-on and feedback-off ERA-Interim simulations at 0100 UTC. Vectors show υ, w component of wind (m s−1) for feedback off. The left-to-right vertical cross section is taken from south to north along the blue line that is shown in the inset. The red line in the inset shows the fire perimeter.

Fig. 9.

Cross section of the difference in potential temperature (shading; K) between the feedback-on and feedback-off ERA-Interim simulations at 0100 UTC. Vectors show υ, w component of wind (m s−1) for feedback off. The left-to-right vertical cross section is taken from south to north along the blue line that is shown in the inset. The red line in the inset shows the fire perimeter.

Cunningham (2007) describes idealized simulations of a buoyant plume and density current and he shows that the interactions between the two have a significant impact on updraft strength. Although he does not mention a timing shift such as that seen here, his Fig. 5 suggests that the minimum in convergence at the base of the plume was earlier for the density current and buoyant plume simulation, hence an earlier arrival of the density current in his “coupled” case.

Ogawa et al. (2003) present the results of a numerical study of interactions between a sea-breeze front and convective cells and show that fronts are modified by anomalies of wind and temperature ahead of them. They showed that frontogenesis occurs when the sea-breeze front approaches the updraft region of a convective cell, seen as peaks in the vertical velocity. Convective cells and a fire plume have dynamic similarities, and our results are consistent with those of Ogawa et al. (2003).

Another perspective is to consider the dynamics of a gravity current, since sea-breeze fronts are a form of these. The speed of a gravity current is given by Simpson (1997) to be

 
formula

where U is speed of the gravity current perpendicular to the front (m s−1), g is acceleration due to gravity (m s−1), is the difference in temperature pre- and postfront (K), T is the postfront temperature (K), and h is the depth of the cold air (m). Simpson presents a worked example of the passage of a cold front over Melbourne on 8 February 1983, for which the density current was associated with a severe dust storm.

From Eq. (1), an increase in prefront temperature increases the propagation speed of the gravity current. Rearranging Simpson to determine the change in U due to the fire,

 
formula

where subscript 1 denotes no feedback, is the prefront increase in temperature between the feedback-on and feedback-off cases, and is the increase in gravity current speed between the feedback-on and feedback-off cases.

Values of = 1.2 m s−1 and = 1.0 m s−1 were estimated for frontal propagation speed east of Cape Gantheaume and the fire front between 0000 and 0100 UTC for the feedback-on and feedback-off simulations (the values are approximate as the frontal acceleration was not consistent in space or time). This represents an approximately 20% increase in propagation speed of the gravity current due to the energy released by the fire.

The increase in temperature due to the fire occurs over a restricted area. At 0100 UTC, the maximum temperature on the (222 m) atmospheric grid was 28.5°C, located at the southern extent of the head fire. The 2-m temperature fields for the feedback-on and feedback-off simulations both show temperatures around 15.5°C over the sea and 21.5°C over land (land adjacent to the fire, but not strongly affected by it) at 0100 UTC. Temperatures over land behind the front were 17°–18°C. Equation (1) applies for two uniform air masses, but the pre-sea-breeze air mass is not uniform, and therefore it is not clear over what area the prefront temperature should be averaged. Taking 1°C perturbation as a reasonable figure (consistent with the anomalies in the cross sections), Eq. (2) shows this is sufficient to produce a 20% increase in propagation speed of the gravity current.

So, Eq. (1) shows that the change in timing of the front should be expected since, as the prefront temperature increases (e.g., a fire plume is present), the gravity current speed U will increase. Also, the frontogenesis arguments of Ogawa et al. (2003), supported by our Fig. 7, show that as a sea-breeze front approaches a fire plume, the frontal line will have stronger up motion than the same scenario without a fire.

In our simulations, the spatial extent of the modification of the near-surface winds by the fire plume extended several kilometers from the fire front and the strength of the fire-modified winds was of a similar speed to the background flow (5–10 m s−1). Similar wind modification speeds and spatial extent were found by Coen (2005).

The relocation and strengthening of a frontal wind shift in the vicinity of a fire are significant for fire management in southern Australia, since in operational fire weather forecasting considerable emphasis and detail is invested in predicting the strength and arrival time of a change from northwest to southwest winds. The D’Estrees simulations are just one case; however, the result suggests that predicting wind change time and strength using an uncoupled NWP forecast may have limited accuracy in a fire environment, since a fire may cause a wind shift approaching a fire ground to arrive earlier than expected. Also, this simulation shows that a front approaching a fire may have stronger vertical motion on the frontal line relative to a front without a fire. The enhanced updraft has implications for the intensity of spotting activity, since stronger upward motion on the frontal line provides a mechanism for enhanced lofting of fire brands.

d. Fire-induced vortex

A feature of particular interest in the ERA-Interim initialization with feedback on is a long-lived vortex, shown in Fig. 10. The vortex developed on the intersection of the southeast and southwest frontal lines at 0300 UTC (Fig. 11). It first developed just behind the head fire, but by 0320 UTC it had moved to a position just in front of the active fire line. The vortex persisted as a feature located adjacent to the head fire for 5.5 h of simulation time. The size of the vortex varied with time; however, it was generally 1–2 km in diameter and had a vertical extent of 500–600 m. Other smaller, transient vortices also developed in the domain, but were unremarkable in time and space by comparison. A similar vortex did not occur with the feedback-on GFS initialization. Nor did a vortex develop in either of the feedback-off simulations. However, a vortex did develop in each of the three ERA-Interim sensitivity tests described earlier.

Fig. 10.

Wind vectors and vorticity (s−1; shaded, with vorticity about the vertical axis) at 10 m. The fire line (feedback on) is shown by the red line. ERA-Interim initialization is used. (bottom right) The plot region (green box) for the other panels, with the fire-line evolution shown by the red lines.

Fig. 10.

Wind vectors and vorticity (s−1; shaded, with vorticity about the vertical axis) at 10 m. The fire line (feedback on) is shown by the red line. ERA-Interim initialization is used. (bottom right) The plot region (green box) for the other panels, with the fire-line evolution shown by the red lines.

Fig. 11.

As in Fig. 10, but showing the time period of vortex development.

Fig. 11.

As in Fig. 10, but showing the time period of vortex development.

Fire whirls are described by Forthofer and Goodrick (2011) as “vertically-oriented, intensely rotating columns of gas found in or near fires, usually visible because of the presence of flame, smoke, ash and/or other debris.” The vortex in Fig. 10 is not exempt from Forthofer and Goodrick’s definition of a fire whirl as they include whirls with no inner core of flame (recognizing that the physics of WRF and SFIRE are unable to resolve flames or a flaming whirl). Our simulated vortex is consistent with their definition of a fire whirl, as they state the diameter of a fire whirl can range from less than 1 m to possibly 3 km (the simulated vortex is 1–2 km), with wind speeds from < 10 m s−1 to > 50 m s−1 (the simulated vortex has wind speeds around 5 m s−1).

Fire whirls are periodically observed at fire grounds, and frontal wind shift lines are likely to be a favorable location for their development because of the presence of preexisting vorticity that a fire plume can stretch and amplify. Billing and Rawson (1982) describe a large fire tornado that formed near the head of a fire at about the time a slow-moving cold front passed through the area (they described it as a fire tornado, rather than a fire whirl). Similarly, Umscheid et al. (2006) describe a large and long-lived fire whirl that developed on a slow-moving front. Our simulations, in conjunction with the studies of Umscheid et al. (2006) and Billing and Rawson (1982) suggest that in rare cases, slow-moving fronts at a fire ground may produce large fire whirls. Forthofer and Goodrick (2011) list potential fire whirl dangers as increased energy release rate, spread rate, and spotting—all of which threaten the safety of firefighters in the vicinity. Therefore, tools to assist in predicting their likely development would be useful.

Figure 11 shows that the fire whirl developed in a region of enhanced vorticity on the intersection of wind from three directions: southwest and south-southeast winds converging from either side of Cape Gantheaume and northwesterly winds from the northern flank of the fire area (see also Fig. 3). The vortex development is behind, rather than coincident with, the main frontal wind shift and also behind the active fire line. Figure 12 shows vertical wind shear in both speed and direction below 3 km, with speed shear at a maximum around the time of vortex development at 0300 UTC and directional shear enhanced through the period of the vortex being adjacent the fire line. Kochanski et al. (2013c) describe simulations that demonstrate the influence of low-level wind shear on fire propagation.

Fig. 12.

(top) Wind direction and (bottom) wind speed vs height for hourly steps from 0100 to 0600 UTC. The wind profiles are taken at 35.96°S, 137.48°E (near the vortex start that is shown in the bottom-left panel of Fig. 11) for the ERA-Interim feedback-on simulation.

Fig. 12.

(top) Wind direction and (bottom) wind speed vs height for hourly steps from 0100 to 0600 UTC. The wind profiles are taken at 35.96°S, 137.48°E (near the vortex start that is shown in the bottom-left panel of Fig. 11) for the ERA-Interim feedback-on simulation.

Figure 10 shows the fire front takes on a right-angled shape, focused to a point near the fire whirl. Examination of the time series of rate of spread showed that between 0400 and 0630 UTC the fastest rate of spread along the fire front was adjacent to the fire whirl, providing a mechanism for increased fire activity. A dense smoke plume was a feature of the D’Estrees fire (Fig. 2) and the plume extent and opacity suggest a high-intensity fire. A high-intensity fire in response to a fire whirl is consistent with Countryman (1971), in which he states that laboratory experiments and field observations show the burning rate of fuels is increased by the presence of fire whirls. He attributes increased rate of burning to higher wind speeds within the whirl and in the whirl inflow. The vorticity associated with the fire whirl and associated up motion also provides a mechanism for development of an organized plume.

In this study, we focus on documenting the development of the vortex adjacent the fire line. The dynamics of the simulated vortex, and a detailed description of the processes by which it developed, is the intended subject of a further study.

The GFS wind shift and the ERA-Interim vortex from our simulations each present two alternative, and not necessarily mutually exclusive, mechanisms for increased burning at the fire front and consequently the large volume of smoke seen in Fig. 2. The low-level postchange wind speeds in the GFS case present a likely mechanism for a longer active fire line, and hence greater smoke production. The vortex of the ERA-Interim initialized simulation provides another possible explanation: a fire whirl with locally strong winds, hence an increased rate of burning and smoke. Although there were no supporting observations of a fire whirl at the D’Estrees Bay, the observations were very limited and insufficient to prove or disprove its presence.

4. Conclusions

Unusual fire activity was observed at the D’Estrees fire on Kangaroo Island in December 2007. The coupled fire–atmosphere model WRF and SFIRE has been used to simulate the event, with the objective of examining the fire–atmosphere interactions that may have occurred. Two sets of simulations were made: one with feedback from the fire to the atmosphere, and the other without feedback, using two different NWP initialization models—the ERA-Interim and GFS. The simulations show that the energy released by the fire produced two significant modifications to the surrounding atmosphere. The first was faster propagation of a sea-breeze wind shift frontal line and enhanced vertical motion on the sea-breeze frontal boundary. The second was that the simulations initialized with the ERA-Interim analysis developed a long-lived vortex. These atmospheric features were identified as possible mechanisms for enhanced fire behavior: the wind shift produced a longer active fire flank and hence a larger burning area and the vortex provided a mechanism for increased fire intensity. The results of these simulations show that the assumption that evolution of a fire is dependent on the atmosphere, neglecting evolution of the atmosphere dependent on the fire can introduce significant errors.

The difference in fire spread resulting from the two initializations (GFS and ERA-Interim) illustrates another challenge in fire prediction. The fire perimeter is seen to be highly sensitive to small variations in wind speed and direction. Because it is not computationally possible to produce a “perfect” high-resolution wind forecast at midflame height, the difference in fire perimeter for the two NWP initializations reinforces the argument that probabilistic rather than deterministic methods are appropriate for predictive modeling of fire spread to account for the uncertainties in meteorological inputs.

The findings we present here are relevant to topical fire research because they provide new evidence for fire–atmosphere interactions. They are one of the first studies providing numerical evidence of a fire changing the mesoscale atmosphere. These simulations show only one case, so general conclusions cannot be drawn, but the implications are wide reaching for the fire management community as the results raise compelling questions regarding current approaches to providing weather information for fire management decisions and weather inputs to fire behavior simulators. These include the limitations of an uncoupled point forecast derived from an NWP operational run, particularly in a sea-breeze or wind shift environment; the potential for a given fire to respond to and possibly modify the meteorological environment; and the appropriate path forward for research and development of coupled and uncoupled fire behavior models, and the inputs and parameterizations that should be included in predictive models.

The results should be interpreted bearing in mind the capabilities of WRF and SFIRE. Mandel et al. (2011), Coen et al. (2013), and Kochanski et al. (2013a) highlight some of the limitations, including lack of a crown-fire model, and problems associated with the range of scales covered by the model (from synoptic to microscale), to mention a few. One particularly relevant question is whether the treatment of latent and sensible heat fluxes from the fire into the atmospheric grids is a reasonable approximation of processes in a real fire, an assumption that is difficult to verify. In a broader context, many of the limitations of fire-spread models discussed by Finney et al. (2013) similarly apply to coupled as well as uncoupled models, in particular that here we provide a deterministic solution to a stochastic process. Also, the simulated features have been assumed to be an authentic representation for the purposes of this discussion, and this assumption remains untested. However, the WRF model (with and without a fire component) has been well validated in numerous other studies, and, notwithstanding the limitations, our simulations provide useful insights into the feedback process.

Anticipating likely fire behavior is critical to successful mitigation activities during bushfire events and for strategic planning of fuel reduction burns. Because of the expanding Australian wild land–urban interface, bushfires and fuel reduction burns are frequently adjacent to residential areas. These areas are frequently fire-prone mountainous and coastal locations where mesoscale meteorological processes dominate and fuel loads can be high. In addition, climate trends are producing shorter windows of opportunity for fuel reduction burning. This combination of factors is increasing the risks associated with fuel reduction burns and bushfires, increasing the need for comprehensive and accurate information for strategic planning and mitigation decisions.

Since weather forecasts and fire simulation models are key tools for predicting fire behavior, appropriate parameters to include in weather forecasts and appropriate inputs into fire behavior models must be established. Greater understanding of fire–atmosphere interactions will assist in determining appropriate detail in weather forecasts for fires and appropriate inputs to fire behavior models. The evidence from this study shows there are limitations to a fire-simulation model that uses constant meteorological inputs.

Scientific understanding of how a fire interacts with the atmosphere at present is limited, as is understanding of how feedback between the two may manifest as unexpected fire behavior. Since observations are difficult to make at a bushfire, coupled simulations such as this one are an important tool for understanding fire–atmosphere interactions.

Acknowledgments

Thanks are given to the Department of Environment, Water and Natural Resources (Mike Wouters and colleagues) for fire spread and vegetation data. Also, we thank the developers of WRF and SFIRE for their suggestions and assistance. Simulations were run using E-research South Australia IT infrastructure on the supercomputer “Tizard.” ERA-Interim data were obtained from the ECMWF data server. This work has been supported by the Bushfire Cooperative Research Centre and the Bureau of Meteorology.

REFERENCES

REFERENCES
Anderson
,
H.
,
1982
: Aids to determining fuel models for estimating fire behavior. USDA Forest Service, Intermountain Forest and Range Experiment Station General Tech. Rep. INT-122, 28 pp. [Available online at http://www.fs.fed.us/rm/pubs_int/int_gtr122.pdf.]
Baughman
,
R. G.
, and
F. A.
Albini
,
1980
: Estimating midflame windspeeds. Proc. Sixth Conf. on Fire and Forest Meteorology, Seattle, WA, Society of American Foresters,
88
92
.
Beezley
,
J.
,
A.
Kochanski
,
V.
Kondratenko
,
J.
Mandel
, and
B.
Sousedik
, cited
2010
: Simulation of the Meadow Creek fire using WRF-Fire. [Available online at http://www.openwfm.org/w/images/a/ae/Agu10_jb.pdf.]
Billing
,
P.
, and
R.
Rawson
,
1982
: A fire tornado in the sunset country, January 1981. Victoria, Australia, Department of Conservation and Environment, Fire Management Branch Research Rep. 11, 12 pp.
Clark
,
T. L.
,
M. A.
Jenkins
,
J.
Coen
, and
D.
Packham
,
1996
:
A coupled atmosphere–fire model: Convective feedback on fire-line dynamics
.
J. Appl. Meteor.
,
35
,
875
901
, doi:.
Clark
,
T. L.
,
J.
Coen
, and
D.
Latham
,
2004
:
Description of a coupled atmosphere–fire model
.
Int. J. Wildland Fire
,
13
,
49
63
, doi:.
Coen
,
J.
,
2005
:
Simulation of the Big Elk Fire using coupled atmosphere–fire modeling
.
Int. J. Wildland Fire
,
14
,
49
59
, doi:.
Coen
,
J.
, and
P. J.
Riggan
,
2014
:
Simulation and thermal imaging of the 2006 Esperanza Wildfire in southern California: Application of a coupled weather–wildland fire model
.
Int. J. Wildland Fire
,
23
,
755
770
, doi:.
Coen
,
J.
,
M.
Cameron
,
J.
Michalakes
,
E.
Patton
,
P.
Riggan
, and
K.
Yedinak
,
2013
:
WRF-Fire: Coupled weather–wildland fire modeling with the Weather Research and Forecasting Model
.
J. Appl. Meteor. Climatol.
,
52
,
16
38
, doi:.
Countryman
,
C. M.
,
1971
: Fire whirls … why, when and where. USDA Forest Service, Pacific Southwest Forest and Range Experiment Station, 14 pp.
Cunningham
,
P.
,
2007
:
Idealized numerical simulations of the interactions between buoyant plumes and density currents
.
J. Atmos. Sci.
,
64
,
2105
2115
, doi:.
Cunningham
,
P.
, and
M.
Reeder
,
2009
:
Severe convective storms initiated by intense wildfires: Numerical simulations of pyro-convection and pyro-tornadogenesis
.
Geophys. Res. Lett.
, 36, L12812, doi:.
Dee
,
D. P.
, and Coauthors
,
2011
:
The ERA-Interim reanalysis: Configuration and performance of the data assimilation system
.
Quart. J. Roy. Meteor. Soc.
,
137
,
553
597
, doi:.
Filippi
,
J.-B.
,
F.
Bosseur
,
X.
Pialat
,
P.-A.
Santoni
,
S.
Strada
, and
C.
Mari
,
2011
:
Simulation of coupled fire/atmosphere interaction with the MesoNH-ForeFire models
.
J. Combust.
, 540390, doi:.
Finney
,
M.
,
J.
Cohen
,
S.
McAllister
, and
W.
Jolley
,
2013
:
On the need for a theory of wildland fire spread
.
Int. J. Wildland Fire
,
22
,
25
36
, doi:.
Forthofer
,
J.
, and
S.
Goodrick
,
2011
:
Review of vortices in wildland fire
.
J. Combust.
, 984363, doi:.
Hu
,
X.-M.
,
J.
Neilsen-Gammon
, and
F.
Zhang
,
2010
:
Evaluation of three planetary boundary layer schemes in the WRF Model
.
J. Appl. Meteor. Climatol.
,
49
,
1831
1844
, doi:.
Jordanov
,
G.
,
J.
Beezley
,
N.
Dobrinkova
,
A.
Kochanski
,
J.
Mandel
, and
B.
Sousedik
,
2012
: Simulation of the 2009 Harmanli fire (Bulgaria). Eighth Int. Conf., Large-Scale Scientific Computing 2011, Sozopol, Bulgaria, 291–298. [Available online at http://arxiv.org/abs/1106.4736.]
Kochanski
,
A.
,
M.
Jenkins
,
S.
Kruger
,
J.
Mandel
, and
J.
Beezley
,
2013a
:
Real time simulation of the 2007 Santa Ana fires
.
For. Ecol. Manage.
,
294
,
136
149
, doi:.
Kochanski
,
A.
,
M.
Jenkins
,
J.
Mandel
,
J.
Beezley
,
C.
Clements
, and
S.
Krueger
,
2013b
:
Evaluation of WRF-SFIRE performance with field observations from the FireFlux experiment
.
Geosci. Model Dev. Discuss.
,
6
,
121
169
, doi:.
Kochanski
,
A.
,
M.
Jenkins
,
R.
Sun
,
S.
Krueger
,
S.
Abedi
, and
J.
Charney
,
2013c
:
The importance of low-level environmental wind shear to wildfire propagation: Proof of concept
.
J. Geophys. Res. Atmos.
,
118
, 8238–8252, doi:.
Linn
,
R.
,
J.
Riesner
,
J. J.
Colman
, and
J.
Winterkamp
,
2002
:
Studying wildfire behaviour using FIRETEC
.
Int. J. Wildland Fire
,
11
,
233
246
, doi:.
Mandel
,
J.
,
J. D.
Beezley
, and
A. K.
Kochanski
,
2011
:
Coupled atmosphere–wildland fire modeling with WRF 3.3 and SFIRE 2011
.
Geosci. Model Dev.
,
4
,
591
610
, doi:.
McCaw
,
L.
,
1998
: Research as a basis for fire management in mallee heath shrublands of south-western Australia. Proc. Third Int. Conf. on Forest Fire Research/14th Conf. on Fire and Forest Meteorology, Coimbra, Portugal, ADAI, 2335–2348.
McRae
,
R.
,
J.
Sharples
,
S.
Wilkes
, and
A.
Walker
,
2013
:
An Australian pyro-tornadogenesis event
.
Nat. Hazards
,
65
,
1801
1811
, doi:.
Mell
,
W. E.
,
R. J.
McDermott
, and
G. P.
Forney
,
2010
: Wildland fire behaviour modeling: Perspectives, new approaches and applications. Proc. Third Fire Behaviour and Fuels Conf., Spokane, WA, International Association of Wildland Fire, 17 pp. [Available online at https://www.firescience.gov/projects/07-1-5-08/project/07-1-5-08_Mell_FireBehaveModeling_3rdFireFuelsConf_2010.pdf.]
Ogawa
,
S.
,
W.
Sha
, and
T.
Iwasaki
,
2003
:
A numerical study on the interaction of a sea-breeze front with convective cells in the daytime boundary layer
.
J. Meteor. Soc. Japan
,
81
,
635
651
, doi:.
Peace
,
M.
, and
G.
Mills
,
2012
: A case study of the 2007 Kangaroo Island bushfires. CAWCR Tech. Rep. 53, 58 pp. [Available online at http://www.cawcr.gov.au/publications/technicalreports/CTR_053.pdf.]
Peace
,
M.
,
L.
McCaw
, and
G.
Mills
,
2012
:
Meteorological dynamics in a fire environment; a case study of the Layman prescribed burn in Western Australia
.
Aust. Meteor. Oceanogr. J.
,
62
(
3
),
127
141
.
Potter
,
B.
,
2005
:
The role of released moisture in the atmospheric dynamics associated with wildland fires
.
Int. J. Wildland Fire
,
14
,
77
84
, doi:.
Rothermel
,
R.
,
1972
: A mathematical model for predicting fire spread in wildland fires. USDA Forest Service Research Paper INT-115, 48 pp.
Simpson
,
C. C.
,
J. J.
Sharples
,
J. P.
Evans
, and
M. F.
McCabe
,
2013
:
Large eddy simulation of atypical wildland fire spread on leeward slopes
.
Int. J. Wildland Fire
,
22
,
599
614
, doi:.
Simpson
,
J. E.
,
1997
: Gravity Currents in the Environment and the Laboratory. 2nd ed. Cambridge University Press, 244 pp.
Skamarock
,
W.
, and Coauthors
,
2008
: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3_bw.pdf.]
Umscheid
,
M.
,
J.
Monteverdi
, and
J.
Davies
,
2006
: Photographs and analysis of an unusually large and long-lived firewhirl. Electron. J. Severe Storms Meteor.,1 (2). [Available online at http://www.ejssm.org/ojs/index.php/ejssm/article/viewArticle/6/11.]