1. Introduction
This study presents simulations of the Rocky River fire using the coupled atmosphere–fire spread model known as “WRF-SFIRE” (described in section 2). The fire occurred in early December 2007 on Kangaroo Island, off the coast of South Australia (see Fig. 1a). It was ignited by a dry lightning storm that started over 20 fires on the island, four of which continued to burn for two weeks. On 9 December, one of the fires burned rapidly up the Rocky River gully in a wilderness protection area. The gully was densely vegetated with dry, predominantly mallee fuels1 and the topography is relatively low relief. Fire managers described the fire activity as extreme. One of them (R. Ellis, 2010, personal communication) described the day as “the worst fire behaviour I’ve ever seen.” Flames to a height of 50 m were observed when the fire reached a road at the northern end of the gully [see two photographs in Peace and Mills (2012)]. The fire was inaccessible for surface suppression activities because of a lack of vehicle access, and water bombing from aircraft had no significant effect.
(a) Map of southern South Australia showing Kangaroo Island and the nested domains for the WRF and SFIRE simulations. CFSD is the nearby AWS. Colored contours show topography (m). (b) NOAA Aqua satellite MODIS image at 0610 UTC (1640 local time) 9 Dec 2007. Red pixels show hot spots detected by the satellite.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
The weather conditions were not typical for extreme fire behavior in southern Australia. Synoptic patterns are frequently used to categorize days of high fire risk (e.g., Potter 2012). In South Australia, the typical severe fire weather pattern is a hot, dry airstream from the center of the continent, driven southward by strong and gusty northerly winds, and preceding the passage of a strong southwesterly wind change behind a synoptic cold front. In contrast, the weather pattern on the day of the Rocky River fire was of a high pressure ridge to the south of the continent in the wake of a weak cold frontal passage on the previous day. A detailed description of the synoptic meteorology is given in Peace and Mills (2012).
In addition to three permanent automatic weather stations (AWS) on Kangaroo Island, four portable stations were deployed during the fire campaign. The closest AWS to the Rocky River fire was portable station Country Fire Service AWS “D” (CFSD), located in the second-innermost nest of the simulations (see Fig. 1a for location). (Figure 3, described in more detail below, shows a comparison of meteorological parameters for CFSD observations and output from nest 2 of the WRF grids.) CFSD recorded daytime temperatures ranging from 15° to 18°C, with dewpoint temperatures of 2°–8°C, resulting in relative humidities of 35%–55%. An important aspect of the weather conditions from a fire perspective was near-surface wind speeds averaging 25–35 km h−1. The wind direction was broadly aligned with the orientation of the gully, from the southwest to the northeast. Figure 1b shows the afternoon MODIS image on the day of the fire. The island is obscured by scattered to broken stratocumulus cloud, which would have inhibited insolation. The cloud bands in Fig. 1b are nearly aligned with the surface wind direction and indicative of cellular convection in the boundary layer. The heavy loads of mallee fuel on the island were very dry because of low rainfall the preceding winter. Mallee is a fuel with a strong response to wind (Cruz et al. 2013), so rapid fire spread was expected because of the prevailing winds. However, the observed extreme fire behavior was not anticipated. The forest fire danger index (FFDI) described by McArthur (1968) is the measure of fire danger used over Kangaroo Island. On the day of the Rocky River fire, the FFDI recorded at the nearby AWS CFSD peaked at 13, a value in the “high” category, indicating conditions in which fire suppression is expected to be effective.
Peace and Mills (2012) describe the meteorology of the event, and in that case study they proposed that the fire activity could be attributed to dry, heavy fuel loads, winds aligned with local topography, and, possibly, entrainment of dry air from above a subsidence inversion. Following that case study, this paper describes simulations of the fire using the coupled fire–atmosphere model WRF-SFIRE. The aim is to explore fire–atmosphere interaction processes with the coupled model and to test hypotheses developed in the case study, rather than to reproduce the observed fire behavior. A detailed verification is in any case unachievable for this fire as it was a bushfire in a location with limited access, thereby there were very limited observations of fire rate of spread and no meteorological observations at the fire ground. The modeled fire spread is reasonably consistent with the very limited observations, but should not be regarded as an effort to replicate and verify known fire activity using the coupled model. Notwithstanding the limited verification data, there is merit in simulating the Rocky River fire with a coupled model as the simulations provide an opportunity to explore the processes that may have contributed to the extreme fire behavior that was anecdotally described.
The importance of atmospheric processes impacting on a wildfire has been an area of active research in recent years. A number of recent meteorological case studies describe fires in which unexpected fire activity can be attributed to features in the lower atmosphere mesoscale environment. Several of these (e.g., Mills 2005, 2008a; Charney and Keyser 2010; Zimet et al. 2007) describe dynamical mixing of dry and high-momentum air from the mid–upper troposphere to above a fire site. In each of the events above, extreme fire behavior occurred in an environment where dry, high-momentum air was present in the midtroposphere. Each study proposed meteorological mechanisms by which the surface fire activity could be enhanced by mixing of the air mass from the mid- to upper troposphere to near the surface.
This study is the second of a series of three Australian fires events simulated with WRF-SFIRE (Peace 2014). The first study (Peace et al. 2015) describes the D’Estrees Bay fire, which also burned on Kangaroo Island during the 2007 bushfires. Although the two fires were separated by a distance of around 100 km, the unusual fire activity identified at each occurred on different days, under the influence of different mesoscale atmospheric structures. The coupled simulations show distinct and different fire–atmosphere interaction processes occurring at the two fires. The following section provides a description of the WRF-SFIRE model and the configuration used in the simulations. We then present the results of the simulations, followed by a discussion on the implications of the findings.
2. WRF-SFIRE
The model used in this study is WRF-SFIRE. The WRF (Weather Research and Forecasting) numerical weather prediction model is described by Skamarock et al. (2008). The SFIRE model is a two-dimensional fire spread model coupled to the WRF Model, as described by Mandel et al. (2011). This study was undertaken as part of the same research as Peace et al. (2015), which contains a more complete description of WRF-SFIRE, including how the atmosphere and fire components interact. It also provides a broader context for this study by reference to other simulations performed using the coupled model.
The configuration for these simulations is very similar to that described in Peace et al. (2015) and also is reported in Peace (2014). The simulations were runs for a 24-h period starting at 1200 UTC 8 December 2007. Four nests were used (see Fig. 1a for nest configuration); nest resolutions were 6 km, 2 km, 667 m, and 222 m. The fire grid resolution was 22 m, run at a 1:10 ratio on the innermost nest. The time steps were 18, 6, 2, and 0.5 s, (the time step of 0.5 s on the inner nest was required to maintain model stability). Vertical levels, physics, and dynamics options were as reported in Peace et al. (2015). Initialization was with the European Centre for Medium-Range Weather Forecasts interim reanalysis (ERA-Interim) data (Dee et al. 2011), with simulations run on Tizard at E-Research SA. As previously, simulations were run in both feedback “on” and feedback “off” mode. When feedback is on, at each time step a heat and moisture flux is calculated from the fire grids and inserted into the atmospheric grids, whereas when feedback is off no fluxes are inserted, so the atmosphere is effectively unaware of the fire and no coupling occurs. By comparing simulations with feedback on and off, we investigate how the fire changes the meteorological environment and the feedback of those changes onto the fire.
As in Peace et al. (2015) high-resolution topographic grids at 25-m DEM and high-resolution vegetation grids at 10 m were derived from data provided by the local land management agency [Department of Environment, Water and Natural Resources (DEWNR)]. Figures 2a and 2b show the fuel map and topography in the simulations, respectively. The Anderson (1982) fuel models implemented in WRF-SFIRE are dissimilar to the local mallee fuels on the island (e.g., Cruz et al. 2013). Therefore, to create a representative mallee fuel type for use in the Rothermel framework, a fuel category was created following the mallee descriptors in McCaw (1998) (documented in Peace 2014). Fire spread information provided by DEWNR showed fire spread was restricted to the east by a bitumen road, so a no-fuel area (shaded yellow in Fig. 2a) was included to constrain fire spread. Higher fuel loads in the gully compared to the adjacent ridgetop were thought to have been an important factor influencing fire behavior (Peace and Mills 2012; R. Ellis, DEWNR, 2010, personal communication). Fuel load differences were therefore created in the input file with mass loading doubled in the gully compared to the adjacent ridges, using a simplified vegetation map edited from the version provided by DEWNR. However, examination of the simulations as the fire crossed different fuel loads showed no difference in fire perimeter growth rate.
High-resolution grids for (a) fuel and (b) topography (m) as provided by DEWNR and used in the WRF-SFIRE input grids. The vegetation grids were reallocated to create a simple mosaic of three fuel types: no fuel (yellow), fuel in gullies (blue), and fuel on ridges (red). Fire perimeters at 10-min intervals from the feedback-on simulation are overlaid in black.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
A constant value of fuel moisture of 10% was used, based on estimates given by the local fire manager (R. Ellis, DEWNR, 2010, personal communication). The temperature and dewpoint temperature ranges from AWS observations (see Fig. 3) input to Table 4.6 of Cruz et al. (2015) suggest a fine dead fuel moisture content of 11.5%–13%, only marginally higher than the field estimate of Ellis. The difference may reflect climatologically dry conditions and has only a small effect on the simulated spread of the fire because rate of spread is relatively insensitive to fuel moisture content in this range. Mapping of fire perimeters during the two-week campaign was performed intermittently and with limited resources. However, it is known that on the day of interest, the Rocky River fire was active at the southern end of the gully, near the visitor center in the morning, and burned up the gully to arrive at the (Playford Highway) bitumen road at the northern end of the gully in the late afternoon. Based on this information, a fire ignition line was made at 2200 UTC (0830 local time) at the southern end of the gully in the model. Preliminary simulations showed the fire moving much too fast to be consistent with the observed the late afternoon fire arrival time at the bitumen road top of the gully. The wind scaling factor was adjusted, and a value of 0.2 presented a late afternoon fire arrival near the road. As the Rothermel spread model assumes a wind speed measured at 20 ft above the ground, a wind adjustment will be required for whichever fuel model is chosen, depending on the density and cover of the vegetation type. The appropriate wind adjustment for mallee is likely to be between 0.6 and 0.2. We have selected a factor toward the lower end of the range, which is within the range for forests within established canopy described by Baughman and Albini (1980) and justify this on the basis of the long-unburnt dense vegetation and degree of topographic shielding from the prevailing wind.
CFSD observations and WRF output from nest 2. Time is given from midnight to midnight with sampling at 10-min intervals for the observations and 30-min intervals for the WRF grids.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Initial runs were made with output at 10-min intervals. However, this gave a very fragmented time series of fire plume evolution. Accordingly, the simulations were rerun from a restart file with output at 1-min intervals for the time period 0600–0830 UTC (1630–1900 local time; data storage for a longer period was prohibitive).
3. Results
Three significant weather features were apparent through the afternoon and evening: temporal oscillations in the wind speed and temperature, a gradual backing of wind direction (southwest to south-southwest), and decreasing wind speed (Fig. 3).
Simulated fire perimeters at 10-min intervals with feedback on and off are presented in Fig. 4. Fire spread was constrained at the southeastern flank and northern edge by roads that are included as areas of no fuel (see yellow region in Fig. 2a). Isochrones for the feedback-on simulation are more tightly packed along the eastern flank of the fire than in the feedback-off simulation. This restricted fire spread occurred along a ridgeline that can be seen in Fig. 2b. In the feedback-on simulations, convergence of the fire-modified winds along the ridgeline prevented the fire from spilling over to the eastern side of the ridge (not shown). By comparison, there was no convergence in the unmodified winds in the feedback-off run, so the fire moved unrestricted to the east. Some of the difference in fire area in the two simulations is due to timing of fire spread relative to the backing environmental winds. However, the dominant mechanism for producing the different fire perimeters was interactions between topography and fire-modified winds, which prevented fire growth to the east of the ridgeline.
Fire perimeters for feedback-on and feedback-off simulations with output at 10-min intervals.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
In these simulations, the total fire area is greater for feedback off than for feedback on (see Fig. 4). This result differs from our previous simulations of the D’Estrees Bay fire (Peace et al. 2015), in which total fire area was greater with feedback on.
Simulated fire spread was sensitive to slight variation in wind direction (Fig. 5). Fuel fraction (Fig. 5, bottom) is a measure of area burned at each time step (on the atmospheric grid cells with feedback on). Note the rapid increase in fuel consumed from 0630 UTC. At this time, the backing trend in wind direction settled to a direction of 210° (Fig. 5, middle). The increase in fuel consumption occurred as the 210° wind direction steered the fire past the no-fuel region to the southeast of the fire area (yellow area in Fig. 2a). The slight change in wind direction directed the head fire to the north of the no-fuel zone and hence caused the jump in fire spread, in spite of the continuing slow decline in wind speed (Fig. 5, top). This is an important result with respect to operational wind forecasts and amendment criteria. The difference in wind direction would generally not be included in a fire weather forecast since a wind shift of 30° does not trigger amendment criteria. An operational forecast would most likely split the difference with a forecast of 220° or 230°, thus omitting information on the temporal change in wind direction, which in this case had a significant impact on fire spread direction relative to a fire break.
Times series of wind speed (10-m winds), wind direction, and fuel consumption rate (as measured by fuel fraction on the atmospheric grid) at each (10 min) time step of the simulation.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 6 shows detail of fire spread in the 1-min data. A feature of the 1-min data is that fire spread is not steady state through the period. The feedback-on and feedback-off simulations both show “pulses” where the isochrones are closer together, then farther apart, reflecting slower and faster spread periods of the fire. However, the pulses are much less evident in the feedback-off simulations than in the feedback-on simulations. It thus appears that what would otherwise be weak pulses in wind driven fire spread are amplified by the fire–atmosphere feedback. Note that in Fig. 6 the variation is generally over intervals of less than 10 min and so would be averaged out with 10-min output. Pulses in fire activity and particular time periods of interest are annotated. From the formulation of the Rothermel model and its implementation in WRF-SFIRE, the rate of spread (ROS) of the fire can only be affected by fuel description, wind, and slope. As the pulses in rate of spread occurred across a constant fuel type and topography is the same for both simulations, the differences in ROS must be due to the fire-modified winds.2 The cause of the surges and lulls in fire spread is explored in the remainder of this study.
One-minute fire isochrones for the feedback-on run, with faster runs in red and slower fire runs in blue. The 10-min intervals (UTC) are in black and annotated. Significant fire runs highlighted with circles at A, B, and C are discussed in the text.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 7 shows maximum fire spread at each 1-min output time step (151 time steps). The maximum rate of spread is from the WRF-SFIRE output parameter “maximum fire rate of spread” (F-ROS is in any direction, not necessarily perpendicular to the fire front). Fire spread at all times is faster in the simulation with feedback on, with significant oscillations. The maximum fire ROS for feedback on ranges from 3 to 6 m s−1, whereas for feedback off, maximum ROS slowly declines from around 2.8 to 2.2 m s−1, qualitatively consistent with the slow decline in environmental wind speed during this time (see Fig. 5). The much weaker oscillations in maximum ROS in the feedback-off simulations show that the surges and lulls in the feedback-on simulations result from fire–atmosphere feedback. As a consequence of the formulation of the Rothermel equations in SFIRE, the surges can only be attributed to speed of the near-surface fire-modified winds.
Maximum rate of spread (m s−1) at each time step for feedback on (blue) and off (green).
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figures 8 and 9 show vertical motion (up motion in red/yellow, down motion in blue) for the feedback-off and feedback-on simulations, respectively. A pattern of cellular convection is seen, qualitatively consistent with the satellite image in Fig. 1b and fitting a description of Rayleigh–Bénard convection cells (Stull 1988). The convection is open cellular, with up motion on the edges and down motion in the center. Cells are approximately 8 km across and extend from the surface to 500–600-m elevation. The convection cells move from southwest to northeast with time, in the same direction as the environmental winds. A comparison of the feedback-on and feedback-off runs in the two figures indicates the cells are distorted near the fire but similar away from the fire, suggesting the influence of the fire is localized. The cells were a distinguishable feature in the 10-min data series between 0500 and 0800 UTC and are consistent with the cloud pattern in Fig. 1b (taken at 0610 UTC).
Vertical velocity w (m s−1) for the feedback-off simulation at model level 6, which corresponds to a height above topography of approximately 195 m (height varies with sigma level). Times are as annotated. The fire perimeter for the time step is shown in black in each plot. The blue line provides a stationary reference here and for Figs. 9–11.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
As in Fig. 8, but for the feedback-on simulation.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 10 shows the horizontal wind speed for the feedback off simulation at the same height as the vertical motion plots in Figs. 8 and 9. Bands or waves in the wind speed, with maxima and minima oriented northwest–southeast (normal to the wind direction) are seen, with speeds varying from around 6 to 11 m s−1. The wind speed bands are linked to the convective cells, are spatially coherent, and moving at the same speed. The wind speed bands were vertically coherent and the cells were similarly vertically coherent, with no directional shear with height. Wind speed bands were also seen in the feedback-on case, although the wave structure was distorted by the fire–atmosphere interactions in the vicinity of the head fire. The wind speed bands were present for the same time period as the cells, but were identifiable to a slightly higher level (approximately 1-km elevation, a similar height to the inferred cloud base). The bands in wind speed were present at the surface, again varying by a factor of 2, but slightly reduced in speed (approximately 5–9 m s−1). This raises the question as to whether the bands in wind speed directly caused the surges in simulated fire spread described earlier. Figure 7 shows that the variation in rate of spread for feedback off was insignificant relative to feedback on. Since the fast and slow wind speed bands were present in both simulations, only a small part of the variation in fire spread can be attributed directly to variations in near-surface environmental (non-fire-modified) wind speed.
As in Fig. 8, but for wind speed (m s−1) showing bands for feedback off.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 11 shows vertical motion and fire perimeter at times of fast fire spread highlighted as A, B (b1, b2), and C in Fig. 6. Here, B includes two fast fire runs: one just before (b1) and one just after (b2) 0700 UTC, which are shown separately. The plots show snapshots of a feature that is clearly evident in an animated time series. As the leading edge (yellow up motion) of each cell passes over the head fire, a strong pulse of (blue) down motion develops just behind the head fire. In the 2.5-h simulation at 1-min, 12 individual cells pass over the head fire and a downward pulse occurs with each, coinciding with a fast fire run. This generally fits with observations that a fire spreads fastest when driven by downdrafts. The cellular structure and associated down pulses are less coherent after 0800 UTC, most likely because the daytime boundary layer structure is weakening. Kochanski et al. (2013) describe similar formation of downdrafts and surface wind acceleration observed during the FireFlux Experiment and subsequently modeled using WRF-SFIRE.
Vertical velocity w (m s−1; the height is ~195 m) at four times of fast fire spread (A, B, and C from Fig. 6 above—with two stages of B). The diagonal lines show the cross sections taken for Fig. 12.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 12 shows wind cross sections at the times shown in Fig. 11 through the lines annotated in pink. There are several features of interest in the plots. First and most significantly, at each time of fast fire spread (or just before) there is a surge in downward motion into the back of the head fire (circled in black). These surges are localized and low to the ground (500–1000 m deep). The upward plume reaches a higher elevation than the downward plume, with stronger velocities seen in the upward plume. The vertical velocities are similar in strength to those in a small to moderate size thunderstorm. As the atmospheric grid resolution is 222 m, peak instantaneous vertical velocities may be significantly faster than the 5 m s−1 seen here. The plume easily reaches a height of 2–3 km, at which height the wind speed is a westerly 15–20 m s−1. Modification of the wind fields by the fire occurred to a height of 5 km or more. Because the plume is interacting with the airmass at higher levels, there exists a mechanism for momentum entrainment.
Southwest (left)–northeast (right) cross sections of vertical velocity w (shaded) and wind speed (m s−1; vectors). The times correspond to Fig. 11. The black circles are explained in the text.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
The top panel of Fig. 13 shows back trajectories from a location in the peak of the downward motion for 15 min back in time from 0735 UTC. The descent of air parcels from an elevation of around 1100 m into the back of the fire can be clearly seen. The topography beneath the descent minimum is approximately 270 m, so trajectories are around 100 m above the surface fire. The location matches the peak in downward motion in Fig. 12 (bottom right). The bottom panel of Fig. 13 shows the directional clustering of the trajectory parcels, aligned with the southwest winds. Similar trajectories from locations displaced away from the fire showed parcels maintaining a course at near-constant elevation.
Backward trajectory plots showing downward motion of air parcels for 0720–0735 UTC and end point at 35.825°S, 136.864°E. Open circles show particle positions (10 members) at 1-min intervals. The red outline shows fire perimeter.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
Figure 14 shows vertical motion cross sections at a selected times that coincide with slower fire runs than those displayed in Fig. 12. In comparison with Fig. 12, Fig. 14 shows that when the head fire is moving more slowly, either a strong downward surge is not present or the downward motion is comparatively weaker.
As in Fig. 12, but at times of relatively slower fire spread.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
In summary, the simulations show surges and lulls in fire spread in the 1-min data. The surges in fire spread match the passage of Rayleigh–Bénard convection cells over the head fire. Velocity fluctuations in the background winds, seen as bands or waves in the wind speed, are spatially coherent with the convection cells. Fire–atmosphere interactions distort both the wind speed bands and convective cells. The fast fire spread coincides with the passage of the leading edge of the convective cells and downward pulses of vertical motion.
The hypothesis developed in the case study (Peace and Mills 2012) postulated that fire activity may have been enhanced by the mixing of dry air from above the subsidence inversion, thus reducing near-surface relative humidity and providing a mechanism for predrying of fine fuels (dependent on turbulent mixing and entrainment processes).
The hypothesis is explored in Fig. 15, which shows cross sections of the difference in relative humidity for the feedback-on and feedback-off runs. The boundary layer interface (which in the feedback-off simulation is at around 1500 m) is a region of large differences that fluctuate strongly in time. Significant perturbations due to turbulent mixing by the fire plume extend upward (moist intrusions, blue) and downward (dry intrusions, red). Fluctuations in relative humidity extend up to 10 km laterally from the head fire. The dark blue areas show that fire–atmosphere interactions increased the depth of the planetary boundary layer by more than 500 m. Some dry air entrainment from above the inversion (red and orange) can be seen. It is difficult to infer any implications for fuel moisture, as the simulations do not include a fine fuel moisture response. However, it appears likely that such effects would be secondary to the effects of wind, particularly considering that the response time scale for fine fuel moisture to changes in atmospheric relative humidity is of minutes to hours. Also, in this case, the downdrafts are seen to have a maximum impact behind the fire front, hence any drying would affect already burnt fuels. From the simulations, it is not possible to make conclusions regarding exactly how the simulated entrainment would impact an actual fire. However, given the emphasis placed on near-surface relative humidity predictions in Australian fire weather forecasting, the parcel trajectories shown in the simulations, and the links between elevated dry slots and increased fire activity, the process is worth exploring in future simulations that include fuel moisture feedback.
For the Fig. 11 cross sections and the indicated times, relative humidity difference for the feedback-on simulation minus the feedback-off simulation with wind vectors for feedback on. Blue regions are moister air; red areas show drier air.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
In this case, static stability across the interface was extremely strong and would therefore impede vertical mixing. However, in other atmospheric configurations, where dry levels are present with lesser static stability, entrainment may be more likely, so further research, including the examination of additional case studies, is necessary before making broad conclusions. Dry air above unusually active fires features in several case studies [e.g., Mills (2008b) describes a case of extreme drying]. However, in many of these cases meteorological mixing mechanisms were identified that operated on broader spatial scales and longer time scales than the fire plume circulation presented here.
The simulations provide a further interesting insight into the observations of fire activity that were made as the fire reached the northern end of the gully in the late afternoon. As the fire reached the top of the gully and burned close to observers watching from a road, they saw a spot fire ignite to the north of the main fire and get “drawn back” into the main fire, against the prevailing winds. At this time, AWS observations indicate the FFDI had fallen below 10. Figure 16 shows that the simulated fire-modified winds to the northeast of the head fire between 0800 and 0830 UTC were in the opposite direction from the background winds. The unmodified background winds were around 10 m s−1 with the opposing inflow 5 m s−1. The simulations therefore show the wind structure to be consistent with the observation of a spot fire being drawn back into the main fire front.
Wind speed (shaded; m s−1) and vectors at 0812 UTC on the lowest sigma level showing convergence into the head fire. The fire area is outlined in red, and the area of wind inflow is circled in black.
Citation: Journal of Applied Meteorology and Climatology 55, 5; 10.1175/JAMC-D-15-0157.1
4. Conclusions
The WRF-SFIRE simulations of the Rocky River fire provide insights into dynamical interactions that may occur between a fire and the surrounding atmosphere. The simulations showed that fire spread was very sensitive to small changes in speed and direction of both the environmental winds and the fire-modified winds. Simulated fire spread was constrained along a ridgeline because of convergence of fire-modified winds along topography. Output at 1-min intervals shows surges and lulls in the ROS of the fire front. This variation in ROS occurred as a consequence of interactions between the fire plume and low-level mesoscale convective cells. The faster surges occurred as the leading edge of convective cells passed over the head fire. Trajectories show the entrainment of higher-momentum air from above the surface, with air parcels accelerating downward into the back of the fire plume. The hypothesis that dry air entrainment may occur through mixing by the fire plume was shown to have a weak signal only. However, that finding may be specific to this particular case, since static stability across the subsidence inversion was very strong, and vertical entrainment of dry air may be more likely in an atmosphere with weaker stability.
Pulses in fire front propagation are consistent with the surge and stall behavior described by Dold (2011) as well as the non-steady-state fire spread described by Viegas (2004), who both described rate of spread of a fire as nonconstant. Viegas (2004) attributed non-steady-state spread to convection and radiation, whereas Dold (2011) attributed what he termed “surge and stall” behavior to subtle variations in fire line intensity. Pulses in fire front progression are also consistent with fire ground observations and photographic evidence that fire progression tends to show surges and lulls in activity.
The simulations of the Rocky River fire show surges in fire spread linked to the passage of mesoscale convective circulation cells (Rayleigh–Bénard cells) in the PBL over the fire. As the ascending branch of each cell crosses the head fire, the ascent plume is enhanced, and enhanced subsidence behind the head fire transports higher-momentum air downward and drives an increased ROS. However, the simulations shown here present output at 1-min intervals, while the Rothermel model, as with other empirical fire spread models (e.g., McArthur 1968), is intended to describe a steady-state relation between environmental wind speed and fire spread over a time period of minutes to hours. The assumptions implicit in empirical and semiempirical models of fire spread will limit their ability to represent fluctuations in fire behavior at short time scales. Although a comprehensive validation has not been performed in this study, Kochanski et al. (2013) describe a validation study of the coupled model against the field observations from the FireFlux Experiment and conclude that the overall agreement between simulations and observations over a time scale of 20 min and spatial scale of 1–2 km is “relatively good” (Kochanski et al. 2013, p. 1122).
Better understanding of the impacts of fire–atmosphere interactions at short time periods (1–5 min) may be critical for firefighter safety as sudden changes in wind strength and direction can cause abrupt changes in fire activity (e.g., Cheney et al. 2001). Further investigation and development of tools that identify meteorological environments that are conducive to non-steady-state fire spread would be of great benefit.
It is necessary to understand meteorological processes impacting fire behavior to determine the appropriate inputs for fire weather forecasts. These simulation results demonstrate how sensitive fire spread is to wind direction and, as a consequence, illustrate the critical detail required for predictions. Wind predictions for a fire are extremely challenging, since the wind propagating the fire front arises from the combination of background winds at the surface and through the lower troposphere, wind interactions with terrain, wind interactions with the fire plume, and fire interactions with topography. An additional factor is that Australian fuels are highly susceptible to spotting, which is driven by the wind and fire plume. The evidence seen here, as well as in other recent events, suggests that a greater understanding of the subtleties of wind structure and modification of the environment in the vicinity of fire plumes could add valuable information to fire weather forecasts in Australia.
Acknowledgments
Thanks are given to the Department of Environment, Water and Natural Resources (Mike Wouters, Rob Ellis, and colleagues) for fire spread and vegetation data and to Robert Fawcett, Claire Yeo, the editor, and the anonymous reviewers for their constructive comments, which helped us to improve the manuscript. Thanks also are given to the developers of WRF-SFIRE for their suggestions and assistance. This work has been supported by the Bushfire Cooperative Research Centre and the Bureau of Meteorology. Simulations were run on supercomputing facilities at E-Research SA.
REFERENCES
Anderson, H. E., 1982: Aids to determining fuel models for estimating fire behavior. U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range Experiment Station General Tech. Rep. INT-122, 22 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.
Charney, J. J., and D. Keyser, 2010: Mesoscale model simulations of the meteorological conditions during the 2 June 2002 Double Trouble State Park Wildfire. Int. J. Wildland Fire, 19, 427–448, doi:10.1071/WF08191.
Cheney, N. P., J. S. Gould, and L. McCaw, 2001: The Dead-Man Zone—A neglected area of firefighter safety. Aust. For., 64, 45–54, doi:10.1080/00049158.2001.10676160.
Cruz, M. G., W. L. McCaw, W. R. Anderson, and J. S. Gould, 2013: Fire behaviour modelling in semi-arid mallee-heath shrublands of southern Australia. Environ. Model. Software, 40, 21–34, doi:10.1016/j.envsoft.2012.07.003.
Cruz, M. G., J. S. Gould, M. E. Alexander, A. L. Sullivan, W. L. McCaw, and S. Matthews, 2015: A Guide to Rate of Fire Spread Models for Australian Vegetation. CSIRO Land and Water Flagship, 123 pp.
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:10.1002/qj.828.
Dold, J., 2011: Fire spread near the attached and separated flow transition, including surge and stall behaviour. Proc. 19th Int. Congress on Modelling and Simulation, Perth, WA, Australia, Modelling and Simulation Society of Australia and New Zealand, Inc., 200–206. [Available online at http://www.mssanz.org.au/modsim2011/A2/dold2.pdf.]
Kochanski, A. K., M. A. Jenkins, J. Mandel, J. D. Beezley, C. B. Clements, and S. Krueger, 2013: Evaluation of WRF-SFIRE performance with field observations from the FireFlux experiment. Geosci. Model Dev., 6, 1109–1126, doi:10.5194/gmd-6-1109-2013.
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:10.5194/gmd-4-591-2011.
McArthur, A., 1968: Fire behaviour in eucalypt forests. Australia Forestry and Timber Bureau, Leaflet 107, 36 pp.
McCaw, L., 1998: Research as a basis for fire management in mallee heath shrublands of south-western Australia. III International Conference on Forest Fire Research, 14th Conference on Fire and Forest Meteorology: Proceedings, D. X. Viegas, Ed., Vol. II, Associação para o Desenvolvimento da Aerodinâmica Industriais, 2335–2348.
Mills, G. A., 2005: On the subsynoptic-scale meteorology of two extreme fire weather days during the eastern Australian fires of January 2003. Aust. Meteor. Mag., 54, 265–290. [Available online at http://www.bom.gov.au/amm/docs/2005/mills2_hres.pdf.]
Mills, G. A., 2008a: Abrupt surface drying and fire weather Part 1: Overview and case study of the South Australian fires of 11 January 2005. Aust. Meteor. Mag., 57, 299–309. [Available online at http://www.bom.gov.au/amm/docs/2008/mills1_hres.pdf.]
Mills, G. A., 2008b: Abrupt surface drying and fire weather Part 2: A preliminary synoptic climatology in the forested areas of southern Australia. Aust. Meteor. Mag., 57, 311–328. [Available online at http://www.bom.gov.au/amm/docs/2008/mills2_hres.pdf.]
Peace, M., 2014: Coupled fire-atmosphere simulations of three Australian fires where unusual fire behaviour occurred. Ph.D. thesis, School of Mathematical Sciences, University of Adelaide, 193 pp. [Available online at https://digital.library.adelaide.edu.au/dspace/bitstream/2440/90794/3/02whole.pdf.]
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/technical-reports/CTR_053.pdf.]
Peace, M., T. Mattner, G. Mills, J. Kepert, and L. McCaw, 2015: Fire-modified meteorology in a coupled fire–atmosphere model. J. Appl. Meteor. Climatol., 54, 704–720, doi:10.1175/JAMC-D-14-0063.1.
Potter, B., 2012: Atmospheric interactions with wildland fire behaviour—I. Basic surface interactions, vertical profiles and synoptic structures. Int. J. Wildland Fire, 21, 779–801, doi:10.1071/WF11128.
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.
Stull, R. B., 1988: An Introduction to Boundary Layer Meteorology. Kluwer Academic Publishers, 666 pp.
Viegas, D., 2004: On the existence of a steady state regime for slope and wind driven fires. Int. J. Wildland Fire, 13, 101–117, doi:10.1071/WF03008.
Zimet, T., J. E. Martin, and B. E. Potter, 2007: The influence of an upper-level frontal zone on the Mack Lake Wildfire environment. Meteor. Appl., 14, 131–147, doi:10.1002/met.14.
Mallees are a eucalypt species that have multiple stems growing from a ground-level lignotuber. They are usually less than 10 m high and are a major vegetation group of semiarid areas of southern Australia.
The relationship between wind speed and fire spread uses the Rothermel model, which is valid for equilibrium rates of spread. Here, it is applied to winds that exhibit short-term fluctuations. While it may be less accurate in this situation, we nevertheless expect that the tendency for fires to spread faster in stronger winds expressed in the model will apply in nature.
