The meteorology of Black Saturday

Authors


Abstract

The meteorological conditions are investigated over the state of Victoria, Australia on 7 February 2009, the day of the ‘Black Saturday’ fires. Daytime temperatures exceeding 45°C, strong surface winds and extremely dry conditions combined to produce the worst fire weather conditions on record. A high-resolution nested simulation with the UK Met Office Unified Model and available observations are used to identify the important mesoscale features of the day. The highest resolution domain has horizontal grid spacing of 444 m and reproduces most aspects of the observed meteorological conditions. These include organized horizontal convective rolls, a strong late-afternoon cool change with many of the characteristics of an unsteady gravity current, a weaker late-evening cold front and propagating nocturnal bores. These mesoscale phenomena introduce variability in the winds, temperature and humidity at short temporal and spatial scales, which in turn lead to large spatial and temporal variability in fire danger. Copyright © 2012 Royal Meteorological Society

1. Introduction

On 7 February 2009 a series of fire complexes developed over the state of Victoria, Australia. The fires caused 173 fatalities and destroyed more than 2133 residences (VBRC, 2010). The fires also decimated a number of townships, including Marysville and Kinglake. The combined effects of the fires are acknowledged as the worst on record and the day has since been named ‘Black Saturday’. The extremely hot, dry and windy conditions created a day of exceptional fire danger that was exacerbated by a variety of mesoscale phenomena. The meteorology of Black Saturday is the focus of this study.

The quintessential fire weather day in Victoria comprises hot northwesterly flow of continental origin driven by a large anticyclone to the east or northeast and cyclone to the south or southwest. For example, this synoptic pattern was also associated with the catastrophic 1983 ‘Ash Wednesday’ fires (see Reeder and Smith, 1987; Mills, 2005). Days of high fire danger such as these are usually preceded by several days on which the maximum temperature exceeds 30°C, with the final and hottest day punctuated by a strong cool change associated with a synoptic-scale cold front. (In the warmer months, late afternoon or evening cold fronts in southern Australia are often called ‘cool changes’; e.g. see Reeder and Smith (1992). This expression will be used here.) Temperature falls exceeding 10°C following the passage of the cool change are not uncommon. The change in wind direction accompanying the passage of the cool change can also alter the shape, increase the length and change the direction of propagation of the fire front, all of which can contribute to the catastrophe. For example, virtually everyone killed in the 1983 Ash Wednesday fires were killed within 1 h of the passage of a cold front. Similarly, the fire fighters at Linton, Victoria in 1998 were killed after a cold front passed. Although extreme (perhaps the most extreme on record), the synoptic pattern on 7 February was typical of fire weather days in Victoria.

The maximum temperature on 7 February was 46.8°C at Melbourne Airport, which is the highest temperature ever recorded at that site. This extremely high temperature followed a series of hot days and a very significant heat wave (Table 1). Specifically, the three days from 28 to 30 January exceeded 43°C, nine of the eleven preceding days exceeded 30°C, and the two cooler days over this period still exceeded 28°C. There was no rainfall recorded at Melbourne Airport between 9 January and 7 February. This extended period of high temperatures and dry conditions, among other things, led to very dry fuels over much of the state of Victoria and contributed to the severity of the fire danger on 7 February. After the cool change passed on the evening of 7 February, the temperature fell by more than 15°C in 15 min and the relative humidity rose; the maximum temperature on 8 February was only 20.9°C. These temperature and humidity changes, along with 1.8 mm of rainfall (at Melbourne Airport on 8 February) reduced the fire danger and the fires were eventually extinguished.

Table 1. Daily maximum temperatures at Melbourne Airport. [Correction added on 13 August 2012 after original online publication: in column one of the table the January dates have been corrected.]
DateMaximum temperature (°C)
26 January 200924.8
27 January 200936.0
28 January 200943.1
29 January 200944.2
30 January 200944.2
31 January 200931.6
1 February 200934.1
2 February 200928.8
3 February 200928.7
4 February 200931.6
5 February 200932.2
6 February 200933.5
7 February 200946.8
8 February 200920.9

On 7 February there were many reports of extraordinary fire behaviour from fire agencies and the general public. The factors driving this behaviour probably fall into three groups. First, observations suggest that the potential for fires to be ignited and their rate of spread vary exponentially with temperature, relative humidity and wind speed (Noble et al., 1980). Such observations form the basis of empirical fire behaviour models (e.g. see Luke and McArthur (1978) and Cheney and Sullivan (1997), and the references therein). To the extent that empirical fire-spread models are relevant to 7 February, the extreme meteorological conditions and dry fuels may have greatly amplified the potential for ignition and the rate of spread. Second, as the fires on 7 February were extremely intense, they may have coupled more strongly to the atmosphere. As shown by Clark et al. (1996) the atmosphere circulation driven by the fire may affect the fire itself, and in particular may increase the rate of spread near the centre of the fire front. Third, the complex local airflow associated with the topography and mesoscale weather features may have been responsible for localized variations in fire spread that were unexpected and unexplained by a knowledge of only the broad-scale meteorological conditions.

Those factors that fall into this third group are the focus of this study, although it is likely that factors in all three contributed to the fire behaviour on the day. Mesoscale phenomena are thought to affect wildfire behaviour and locally enhance fire danger. Such phenomena may include orographic channelling, downslope winds (Sharples et al., 2010), diabatic mixing of dry intrusions (Mills, 2008) and circulations associated with a frontal passage. Cold fronts are a crucially important ingredient of fire weather days in Victoria. In addition to locally enhancing fire danger, cold fronts can increase the danger to fire fighters because of the effect of unexpected wind direction changes on fire propagation.

The aims of this study are to document the complex meteorological flow on Black Saturday and identify its important mesoscale constituents. These aims are achieved through a high-resolution simulation of the day using a nested version of the UK Met Office's Unified Model. The fires are not represented in this simulation; only the meteorological conditions are modelled. Incorporating the interaction between the fires and the atmosphere into the model is part of a continuing research programme. The remainder of this article is organized as follows: section 2 presents the synoptic conditions and describes the configuration of the numerical model; section 3 examines the key mesoscale phenomena during the day, including propagating bores, the frontal passages and organized boundary layer structures; section 4 quantifies the influence of the mesoscale phenomena on fire danger; and the article is summarized in Section 5.

2. Synoptic conditions

This section examines the synoptic environment that led to Black Saturday and outlines the model configuration used for the simulation.

2.1. Global analysis

Figure 1 shows part of the global analyses of mean sea-level pressure, 850 hPa potential temperature (θ), Ertel potential vorticity on the 350 K isentropic surface and 300 hPa geopotential height, produced by the UK Met Office. The analysis time is 0300 UTC 6 February 2011, which is 1300 local standard time (LST = UTC + 10 h) on the day before Black Saturday. At this time a monsoon trough lies across northern Australia, extending eastward into the Coral Sea. Three distinct lows are embedded in the trough: a tropical depression to the northwest, which would eventually become Tropical Cyclone Freddy, along with lows over the Cape York Peninsula and Coral Sea. A continental-scale heat low with relatively weak pressure gradients lies over most of the continent. Two anticyclones lie well to the southwest and southeast of the continent connected by a very broad region of low pressure over the Southern Ocean. The 850 hPa potential temperatures are largest over central and western parts of the continent, with strong gradients associated with the southern and southeastern coastlines. There is a large-amplitude trough over the Southern Ocean and a weaker trough over the Tasman Sea, which is part of a propagating Rossby wave packet. The weaker trough will subsequently amplify as the Rossby wave pattern amplifies and overturns.

Figure 1.

Global analyses at 0300 UTC (1300 LST) 6 February 2011: (a) mean sea-level pressure, (b) 850 hPa potential temperature and (c) Ertel potential vorticity at 350 K (shaded in Potential Vorticity Units (PVU); 1 PVU = 10−6 K kg−1m2s−1) with 300 hPa geopotential height (contours).

2.2. Numerical model configuration

The UK Met Office's Unified Model version 7.5 (Davies et al., 2005) is used to simulate the event. The global model is initialized from a global analysis valid at 0300 UTC 6 February; these initial conditions are at the end of the 6 h assimilation window centred on 0000 UTC (see Rawlins et al. (2007) for details of the assimilation scheme). The global model provides the initial and boundary conditions to a nested domain (Nest 1) with 12.2 km horizontal grid spacing. A series of higher resolution nests are then initiated sequentially (allowing 3 h between each initiation for model spin-up), with the initial and boundary conditions for each nest provided by the next coarsest model domain (Lean et al., 2008). As is standard with the Unified Model, the nesting is one-way and there is no feedback from the higher resolution nests to the coarser nests. The configuration of these nested domains and their locations are detailed in Table 2 and Figure 2, respectively. Specifically, the highest resolution domain (Nest 4) has a grid spacing of 444 m and encompasses the mountainous region to the north and northeast of Melbourne that was significantly affected by the fires. Orography data for the Australian region is derived from the GEODATA 9 s native resolution dataset (see Hilton et al., 2003) obtained from Geoscience Australia.

Figure 2.

(a) Nest 1 (12.2 km grid spacing) with boxes marking the locations of Nest 2 (4 km), Nest 3 (1.5 km) and Nest 4 (444 m). (b) Closer view of Nest 4 and topography contours (black using 300 m intervals from 100 m to 1600 m, and shading above 400 m) with the coastline marked thick. Also shown in (b) are locations discussed in the text.

Table 2. Outline of the model configuration for each model domain detailing the number of grid points in the zonal (NX) and meridional (NY) directions, the horizontal grid spacing (Δx), the temporal resolution of the lateral boundary conditions from the respective parent domain (ΔtBC), the model time step (Δt), the initialization time (To), and the latitude (lato) and longitude (lono) of the domain's southwest corner.
DomainNXNYΔx (km)ΔtBC (min)Δt (s)To (06/02/2011)lato (°S)lono (°E)
Global640481∼40 km (N320)7200300 UTC
Nest 1888746 12.2601500300 UTC65.0 85.0
Nest 2998778 4.060500600 UTC48.0125.0
Nest 3998998 1.530250900 UTC44.0138.0
Nest 4998998 0.441551200 UTC39.0143.5

The model configurations and physics settings for the global and nested domains are based on the UK Met Office configurations for global, 12.2 km, 4 km and 1.5 km domains that were operational between 14 July and 2 November 2010; settings for Nest 4 (444 m) are based on those for Nest 3 (1.5 km). All domains use a stretched vertical grid with 70 levels. In the global model and Nest 1 the model top is located at 80 km above mean sea level (AMSL), with 11 levels in the lowest 1 km of the atmosphere. Nests 2, 3 and 4 have model tops at 40 km AMSL, with 16 levels in the lowest 1 km. The main differences between the operational configurations and those used here are the model time steps for the high-resolution nests and the temporal resolution of the boundary conditions provided to each nest (see Table 2 for details).

The only other change from the operational configuration is the settings for the boundary layer scheme, which parametrizes vertical mixing and determines the depth of the mixed layer through fluxes of sensible heat using the Lock et al. (2000) one-dimensional scheme. The operational setting for the 4 km domain has the boundary layer scheme active at levels below about 3 km, which poses difficulties for very deep mixed layers. For this reason, in the simulation of Black Saturday the boundary layer scheme in Nest 2 is made active for all model levels. This modification is necessary because the depth of the mixed layer on 7 February exceeded 5.5 km, which is well beyond the uppermost mixed-layer depth that could be properly represented using the operational settings. In Nests 3 and 4 the Lock et al. scheme is replaced by a Smagorinsky-based turbulence closure and used throughout the model domain to parametrize mixing in both the horizontal and vertical.

The model simulation reproduces the important features of the conditions on 7 February around Melbourne. For example, comparison between time series from the highest resolution domain (Nest 4) and automatic weather station observations at Moorabbin Airport (approximately 20 km southsoutheast of Melbourne and marked in Figure 2) demonstrate agreement between simulated and observed pressure, temperature, wind speed and wind direction (Figure 3). In both the observations and simulations the surface pressure falls during the day of 7 February, reaching a minimum of approximately 995 hPa (Figure 3(a)). The 2 m temperature shows similar agreement, with the model reproducing the afternoon maximum of approximately 45°C as well as the nocturnal minima (Figure 3(b)). The afternoon wind speeds in the model are approximately 10 m s−1 at 10 m (Figure 3(c)), which are slightly lower than the observations (approximately 12 m s−1), although the wind direction (Figure 3(d)) is consistent with the observations throughout most of the day. Of importance for the current study is that the simulated cool change arrives at approximately 1630 LST, which is only about 30 min after the observed change; a 30 min error in timing for a forecast with more than 24 h lead time is far smaller than might normally be expected.

Figure 3.

Observed (grey) and simulated (black) time series at Moorabbin Airport starting 2200 LST 6 February 2009. (a) Surface pressure, (b) 2 m temperature, (c) 10 m wind speed and (d) 10 m wind direction. Model data are taken from Nest 4.

In addition to the 30 min error in the timing of the front, the time series (Figure 3) exposes a few other shortcomings of the model forecast. In particular, the strength of the cool change (in terms of temperature and pressure changes) is underestimated by the model (by about 3° and 2 hPa) and the cool change at 2100 LST is difficult to identify in the model. In the morning (around 0800 LST) the temperature increases more rapidly in the observations than the model, as does the 10 m wind speed shortly thereafter; these differences are consistent with slower growth of the simulated boundary layer. The observed 10 m wind shows larger temporal variability than the model, presumably because the 444 m grid spacing does not properly resolve the turbulent boundary layer eddies. Furthermore, the differences in wind variability are larger in the morning, which may be related to the proximity of the location to the coastline and the effect of the ocean on the simulated boundary layer. Nevertheless, despite these differences, the model simulation from Nest 4 reproduces the important features of the conditions on the day. Unless otherwise specified, data from Nest 4 will be examined in the remainder of the article.

2.3. Evolution of the synoptic-scale flow

A 24 h model forecast for 1300 LST (0300 UTC) 7 February from the 12.2 km domain is shown in Figure 4. This is 24 h after Figure 1. The mean sea-level pressure pattern is typical for strong summertime frontogenesis in the region (Reeder and Smith, 1992, 1998). The key ingredients are a surface low and the two adjacent broad anticyclones (Figure 4(a)), which together drive a low-level flow field over the southern part of the continent similar to the classic hyperbolic deformation field used in many of the theoretical descriptions of frontogenesis (e.g. see Reeder et al., 1991; Reeder and Smith, 1992; Muir and Reeder, 2010). As the cyclone approaches the heated continent, this pattern of deformation strengthens the temperature contrast between the continent and the ocean, culminating in a strong cold front.

Figure 4.

Same as Figure 1 except a 24 h model forecast at 1300 LST (0300 UTC) 7 February 2009 from Nest 1.

Berry et al. (2011) finds that the climatological frequency of cold fronts in the region is largest in summer with pronounced local maxima over southwestern and southeastern Australia. These maxima are distinct from the maximum to the south of the continent over the Southern Ocean. Although the parent surface low (Figure 4(a)) and upper trough (Figure 4(c)) lie far to the south of the continent (around 50°S) near the end of the climatological storm track (Berry et al., 2011), a surface trough and cold front extends from the low across the southeastern coastline (Figure 4(a)). There is an extremely strong temperature gradient aligned with the southeastern coastline and a much weaker gradient over the Southern Ocean associated with the low (Figure 4(b)).

3. Mesoscale structures

In this section the mesoscale features identified in the observations are examined in detail in the model simulation.

3.1. Pre-frontal bore

Around 0230 LST on 7 February the automatic weather station at Moorabbin airport recorded a sharp rise in the pressure of around 2 hPa, followed for the next hour or so by a series of oscillations in the pressure and wind (Figure 3). Such a signal is typical of the passage of an undular bore. (A bore is undular when the upstream and downstream parts are connected by a series of waves, see Benjamin and Lighthill, 1954). The simulation produces a rudimentary bore without the accompanying large-amplitude waves, although it arrives more than 2 h too early. Other studies have documented and numerically modelled analogous bores that developed ahead of cold fronts over the Great Australian Bight (Schmidt and Goler, 2010) and over the north Atlantic (Knippertz et al., 2010). Although not shown here, MODIS satellite imagery on 6 February shows a series of wave clouds over the Great Australian Bight propagating beneath broken cumulus cloud and ahead of the surface cold front.

The bore in Figure 3 is also very similar to those that are commonly generated overnight by cold fronts in subtropical Australia (Smith et al., 1995; Reeder and Smith, 1998; Reeder et al., 2000; Thomsen et al., 2009). When a cold front advances into a stable layer, large-amplitude waves are generated. If the stable layer is sufficiently deep, these waves may propagate ahead of the cold front producing a double front structure (for example see Haase and Smith, 1989, Smith et al., 1995, Thomsen et al., 2009). Alternatively, if the stable layer is shallow enough that the phase speed of the generated waves is smaller than the speed of the cold front, the waves can remain bound to the front, giving the front a highly undular appearance. The former case is said to be subcritical whereas the latter is said to be supercritical. As the bore in Figure 3 has propagated far ahead of the synoptic cold front, which at this time lies over the Great Australian Bight, it is likely to have been generated from a subcritical flow.

Figure 5 shows horizontal cross-sections of the vertical velocity at 0030 LST and 0200 LST 7 February from the highest resolution model domain at 500 m above ground level, while Figure 6 shows a vertical cross-section of potential temperature and vertical velocity along the line AB in Figure 5(a). At 0300 LST the model produces a narrow line of vertical motion oriented from just west of north to just east of south (Figure 5(a)), which subsequently propagates eastward along the southern coastal region, reaching a longitude of about 146°E by 0200 LST (Figure 5(b)). In both panels of Figure 5, there is also a prominent, stationary wave-like band of vertical motion attached to the upwind side of the mountain ranges.

Figure 5.

Horizontal cross-section of vertical velocity (coloured shading) at (a) 0030 LST and (b) 0200 LST 7 February 2009 from Nest 4 at 500 m above ground level. Also shown are the coastline (thick black line), terrain contours (black at 500 m intervals) and the location of Moorabbin airport.

Figure 6.

Vertical cross-section along the line AB in Figure 5 from Nest 4 at 0030 LST 7 February 2009. Contours of potential temperature (grey) at 1 K intervals and vertical velocity (black) at 0.4 m s−1 intervals with negative values dashed.

Figure 6 shows a low-level stable layer below about 1 km capped above by a near-neutral layer. According to linear theory (Crook, 1989), such a configuration may act as a (possibly leaky) waveguide, suppressing the upward propagation of wave energy and, thereby, promoting horizontal propagation (e.g. see Menhofer et al., 1997a; Goler and Reeder, 2004; Thomson et al., 2009). At a horizontal distance of about 21 km, the isentropes rise by about 100 m and remain elevated. Such a permanent displacement is one of the characteristics of a bore. The leading edge of the bore is marked by ascent of up to 3 m s−1 followed by subsidence of about 1 m s−1. In contrast to the observations, the modelled bore is not undular.

The cooling that accompanies the passage of an idealized bore is due to the adiabatic vertical displacement of the air. Consequently, there should be no temperature change at the surface because the vertical velocity there is zero. In practice, however, the surface temperature change can be caused by vertical mixing (Menhofer et al., 1997b). Observations over central Australia show that the passage of a bore is generally marked by warming as the potentially warmer air is mixed to the surface (Smith et al., 1995; Menhofer et al., 1997b; Reeder et al., 2000). One important difference between the bore observed on the morning of Black Saturday and the modelled bore (Figure 3(b)) is that the observed bore is marked by surface warming whereas the modelled bore is marked by surface cooling. Haase and Smith (1989) showed that when the flow is supercritical, the leading edge of the gravity current (or front) can become detached and propagate as a horizontal roll vortex, carrying cold air with it. Hence, one possible explanation for the surface cooling produced in the model is that the surface inversion is too shallow, which has the effect of decreasing the phase speed of the waves supported by the layer, thereby making the flow supercritical. Another partial explanation is that there is too little vertical mixing in the model at low levels, which may also account for the inversion being too shallow. The detailed dynamics controlling the generation of the bore and its representation in the model will be investigated in a subsequent study.

3.2. Daytime boundary layer

The extremely high daytime temperatures and strong surface winds (Figure 3) suggest that the boundary layer should be vigorous, with influence from the strong shear and dry convection. At 1300 LST there is strong northwesterly flow at 500 m above ground level (Figure 7(a)), and over land the strongest speeds are in the lee of the mountains. As expected, the vector difference between the 500 m and 2 m winds (not shown) is very similar to the wind vectors in Figure 7(a), with the mean shear vector over that layer directed to the southeast and the magnitude varying from approximately 10 m s−1 km−1 to the north and northwest of the mountains to 18 m s−1 km−1 over the mountains.

Figure 7.

Horizontal cross-section at 500 m above ground level of (a) wind speed (coloured shading) with wind vectors also shown, and (b) vertical velocity (coloured shading) at 1300 LST 7 February 2009 from Nest 4. Also shown are the coastline (thick black line), terrain contours (black at 500 m intervals) and the location of Marysville (white circle).

The 500 m vertical velocity at 1300 LST (Figure 7(b)) shows coherent structures that are aligned with, or at a small angle to, the 500 m wind and the mean shear vector. These coherent structures are consistent with horizontal convective rolls that are known to form in sheared convective boundary layers (Brown, 1980). Like those here, horizontal convective rolls are usually aligned at a small angle to the low-level shear vector. A vertical cross-section perpendicular to the horizontal convective rolls (Figure 8) illustrates the predominant roll circulation to be approximately 2.5 km deep. Above 2.5 km, where the atmosphere is stable, perturbations in potential temperature and vertical velocity are in quadrature, consistent with gravity waves forced by the boundary layer convection (Lane and Clark, 2002).

Figure 8.

Vertical cross-section along the line CD in Figure 7 from Nest 4 at 1300 LST 7 February 2009. Contours of potential temperature (grey) at 1 K intervals and vertical velocity (black) at 0.6 m s−1 intervals with negative values dashed. The terrain elevation is shown with the thick black line.

The horizontal convective rolls in Figure 8 have horizontal wavelengths of approximately 10 km, although in some regions of the flow the wavelengths are closer to 15 km (Figure 7). A satellite image valid at 1450 LST (Figure 9) shows that shallow boundary layer clouds are organized into cloud streets. Such cloud streets are well-known signatures of horizontal convective rolls. Also marked on Figure 9 are lines depicting the separation of the cloud streets; the observed cloud streets have wavelengths between 10 and 17 km, i.e. within the range of separations produced by the model. Moreover, the cloud streets are oriented at a similar angle to those in the simulation. The agreement in orientation and scale implies that the simulated horizontal convective rolls are realistic and probably occurred on 7 February.

Figure 9.

MODIS Aqua true colour image at 0450 UTC (1450 LST) 7 February 2009. The location of Melbourne is marked. The parallel lines in the top-left of the image are approximately 17 km apart and those to the top-right are approximately 10 km apart.

The depth of the mixed layer can be inferred from Figure 8. There is weak stratification above 2.5 km and an inversion that begins around 4.5 km above mean sea level, marking the nominal boundary layer top. Later in the afternoon, the boundary layer becomes well mixed to a height of 5 km, with a sharp inversion at that height (not shown). Near the end of the day (1715 LST, Figure 10(b)) and ahead of the cool change, coherent roll structures persist in the boundary layer combined with cellular features as well. Although there are no observations of the daytime boundary layer depth (due to the timing of the routine soundings in the region), it is likely that the model reproduces the mixed-layer depth well; the routine Melbourne sounding and a model equivalent profile at 1200 UTC (2200 LST) 7 February (not shown) both show a residual layer that extends to 5 km, with a remnant inversion between 5 and 5.5 km above mean sea level. Moreover, of relevance to the next section, this comparison also shows that at this time the depth of the cool air associated with the front is approximately 2 km deep in both the model and observations.

Figure 10.

Cross-section at 500 m above ground level 1715 LST 7 February 2009 of (a) potential temperature and (b) vertical velocity (both colour shaded). Also shown are the coastline (thick black), terrain contours at 500 m intervals (black), the location of Melbourne, and in (a) 500 m wind vectors.

Weckwerth et al. (1997) describe the environmental conditions that are known to favour horizontal convective rolls, and review the characteristics of rolls documented by prior studies. The rolls produced here have horizontal wavelengths of around 10–15 km, which are longer than the roll wavelengths reported by Weckwerth et al. (1997). However, Weckwerth et al. (their figure 17) show that the wavelength of rolls increases with boundary layer depth, and based on their regression a 10–15 km wavelength might be expected for the 2.5 km deep mixed layer at 1300 LST. Nonetheless, the extreme shears and deep mixed layer (especially later in the day) place this event well outside the parameter space normally examined for horizontal convective rolls and future research at these extremes is warranted. In addition to being a prominent feature of the boundary layer on 7 February, it will be shown in section 4 that these horizontal convective rolls make important contributions to the spatial variability of fire danger as well.

Although the numerical model produces dry boundary layer structures similar to the observations, it does not produce deep moist convection in the pre-frontal airstream. On Black Saturday deep non-precipitating cumulus convection was observed and it is likely that the fires generated these clouds (i.e. they were pyrocumuli; see VBRC, 2010). Studying the role of the fires in generating pyrocumulus clouds and their feedback on the fires is part of our continuing research. (See Cunningham and Reeder (2009) for an example of the role played by wildfires in initiating severe pyrocumulonimbi in Australia.)

3.3. Cool change

The satellite image at 1450 LST (Figure 9) shows a northwest to southeast oriented cloud band located to the southwest of Melbourne. This cloud band marks the location of the cool change that propagated to the northeast and arrived in Moorabbin about an hour afterwards (shortly after 1600 LST 7 February, Figure 3). These summertime cool changes form predominantly as an enhancement of the coastal temperature gradient by the synoptic scale flow (Reeder, 1986; Reeder and Smith, 1992; Muir and Reeder, 2010). Reeder (1986) showed that as cold fronts advance onshore during the day into a deep well-mixed boundary layer, they decelerate and may become stationary at the coast. At the same time, the cross-front circulation strengthens very rapidly leading to strong coastal frontogenesis. Muir and Reeder (2010) reached similar conclusions, but also explained how these phenomena lead to a bias in cold fronts making landfall preferentially during the late afternoon or evening. The cool change on 7 February appears to have developed along the lines described by Muir and Reeder, although in this case the mixed layer was exceptionally deep and the background deformation exceptionally strong, resulting in an extreme cool change.

As shown in Figure 3, the simulated cool change arrives at Moorabbin approximately 30 min late at 1630 LST. The cool air is embedded within southwesterly flow and passes through Melbourne at approximately 1700 LST (Figure 10). As is typical, the cool change is marked by a rapid change in wind direction (Figure 10(a)) as well as a coherent updraft along the frontal boundary at 500 m above ground level (Figure 10(b)). Ahead of the front the boundary layer is still convectively active, but behind the front the cooler air is more stable.

The leading edges of very strong cold fronts often have the appearance of unsteady gravity currents (Smith and Reeder, 1988). Figure 11 illustrates the vertical structure of the advancing cool air in the vicinity of Melbourne. The leading edge of the cool air has the hallmarks of an unsteady gravity current, including a feeder flow of cool air toward the nose and an elevated head (approximately 1–1.5 km deep). Within 15 min (cf. Figure 11(a) and (b)) the leading edge of the current has propagated approximately 11 km, making the speed of propagation about 12 m s−1. Accompanying the passage of the cool air is ascent of up to 5 m s−1 at the current head, along with descent between 5 and 15 km behind the leading edge that creates a localized reduction in the current depth. Further behind the leading edge of the front, the cool air deepens to be about 2 km deep, in agreement with the 1200 UTC 7 February sounding at Melbourne (not shown).

Figure 11.

Vertical cross-section along the line EF in Figure 10 from Nest 4 at (a) 1715 and (b) 1730 LST 7 February 2009. Potential temperature is shaded (grey) and vertical velocity is contoured (black) at 0.6 m s−1 intervals with negative values dashed. The thick black line marks the terrain elevation.

Approximately 2 h after passing through Melbourne the simulated cool change arrives at Marysville around 1900 LST. In this region the actual arrival of the southwesterly cool change was between 1735 and 1805 LST (VBRC, 2010) and, therefore, the simulated front arrives approximately 1 to 1.5 h late. At this time the (dry) boundary layer convection has weakened and the 500 m vertical velocity is dominated by the ascent along the frontal boundary (Figure 12(b)). The front is mostly oriented from northwest to southeast, with the gross shape of the western Victorian coastline superimposed upon it. Ahead of the front the winds are strong northwesterlies, and behind the front the winds are predominantly from the southwest (Figure 12(a)). However, immediately behind the front the wind direction has more variability. This variability is due to a combination of: the interaction of the encroaching air with the terrain, with cool air being funnelled to low-lying areas; and the mesoscale variability caused by descending air behind the leading edge of the front (cf. Figure 13). Like earlier times, the advancing cool air has the structure of a gravity current, with eddies forming along the boundary of the gravity current in both the vertical (Figure 13) and the horizontal (Figure 12).

Figure 12.

Same as Figure 10, except at 1900 LST 7 February 2009.

Figure 13.

Vertical cross-section along the line GH in Figure 12 from Nest 4 at (a) 1900 and (b) 1920 LST 7 February 2009. Potential temperature is shaded (grey) and vertical velocity is contoured (black) at 0.7 m s−1 intervals with negative values dashed. The thick black line marks the terrain elevation.

The timing of the arrival of the cool change near Marysville is a weakness of this simulation, with the relatively small error in timing at Moorabbin being amplified in only a short period of time. Because the cool air advances as a gravity current, its propagation speed depends crucially on the depth of the cool air, the strength of the cool perturbation and how these vary across the terrain. These properties should be sensitive to model resolution and details of the boundary layer parametrization, making correct simulation of the timing of arrival challenging.

3.4. Nocturnal front and bore

Figure 14 shows a cross-section of potential temperature and vertical velocity at 500 m above ground level at 2200 LST 7 February. At this time the cool change has almost reached Yarrawonga, with strong pre-frontal northwesterlies backing to become southwesterly. The leading edge of the front is marked by a thin line of strong ascent. Further to the south, the post-frontal southwesterlies split around the southwestern end of the orography (the Great Dividing Range).

Figure 14.

Same as Figure 10, except at 2200 LST 7 February 2009.

Figure 15 shows a vertical cross-section of potential temperature and vertical motion along the line IJ (marked in Figure 14) from the highest resolution model domain at 2200 and 2220 LST 7 February. Two snapshots 20 min apart are shown as it provides an indication of the degree to which the cool change is steady.

Figure 15.

Vertical cross-section along the line IJ in Figure 14 from Nest 4 at (a) 2200 and (b) 2220 LST 7 February 2009. Contours of potential temperature (grey) at 1 K intervals and vertical velocity (black) at 0.7 m s−1 intervals with negative values dashed.

At this time the leading edge of the cool change appears to be an undular bore with strong ascent at its leading edge. However, now the temperature change at the surface is very weak, as it is modified by a strong but shallow stable layer. In the observations (Figure 16) the temperature rises with the arrival of the change before gradually cooling. As with the early morning bore, the temperature increase is due to downward mixing of potentially warmer air. Once again, the model shows little evidence of the change in the surface temperature field and fails to capture this mixing. Like nocturnal fronts over the subtropical interior of the continent, the Black Saturday cool change produces waves at its leading edge with three wave crests evident (Figure 15(b)). The flow is presumably supercritical as the waves appear to be bound to the front.

Figure 16.

Observed (grey) and simulated (black) time series at Yarrawonga starting 1800 LST 7 February 2009. (a) Surface pressure, (b) 2 m temperature, (c) 10 m wind speed and (d) 10 m wind direction. Model data are taken from Nest 4.

Although not shown here, the Bureau of Meteorology radar at Yarrawonga captured the passage of three distinct wave crests. The first and most prominent arrived at 2210 LST, while the second and third arrived, respectively, 20 and 40 min later. A time series from the model and the Yarrawonga automatic weather station (Figure 16) show the error in the timing of the bore to be only about 10 min.

4. Fire weather conditions

In this section the simulation is used to investigate the mesoscale changes in the fire danger and their potential contribution to the fire behaviour on 7 February.

First, the McArthur forest fire danger index (FFDI, Noble et al., 1980) is examined. The FFDI is used by fire agencies in Australia to estimate and predict fire danger (at 1500 LST). The FFDI is defined as

equation image

where H is the relative humidity (%), T is the temperature (°C), V is the 10 m wind speed (km h−1) and D is a drought factor between 1 and 10 (that is a function of soil moisture, the number of days since precipitation and the amount of precipitation in the last 24 h). The FFDI is an empirical relation that was originally calibrated to lie between 0 and 100, where 100 corresponds to the conditions at 1500 LST in Melbourne on Friday 13 January 1939 (the day of the ‘Black Friday’ fires). Until 2009 these conditions were considered to represent the maximum danger. The FFDI values of 0–11, 12–24, 25–49, 50–74, and 75–99 are assigned fire danger ratings of low-to-moderate, high, very high, severe and extreme, respectively. As the conditions on Black Saturday gave FFDI values far in excess of 100, a new fire danger category of catastrophic has been defined by the fire agencies when FFDI > 100.

Although indices such as the FFDI were not developed with high-resolution numerical models in mind, they are normally calculated using point measurements, which is closer to the output from the model than a time or space average. Moreover, the FFDI is at the heart of many empirical fire-spread models, which are applied on very small scales. For example, the McArthur model has been coupled very successfully to a non-hydrostatic atmospheric model by Clark et al. (1996); calculations reported using that coupled model reproduced many of the observed features of real fires. Thus, even though the FFDI does not provide a complete representation of fire behavior, it contains sufficient information that its variability provides an indication of the potential effect of atmospheric phenomena on fires.

The FFDI is calculated from the simulation using instantaneous values of 2 m temperature and relative humidity, and the 10 m wind speed. To isolate the influence of spatial meteorological variability on the FFDI, a constant drought factor of 10 has been used. Although the fire agencies cap the drought factor at 10 by convention, it is probably an underestimate given the hot and dry conditions leading up to Black Saturday and the extreme drying conditions on the day. As is common, the FFDI could also be calculated from time-averaged fields, but this approach would lose much of the spatial variability of the FFDI shown in Figure 17.

Figure 17.

Forest fire danger index (FFDI, coloured shading) at (a) 1300, (b) 1500, (c) 1700 and (d) 1900 LST 7 February 2009 calculated from Nest 4 (with only a portion of the domain shown). Only values over land are shaded, the thick black line marks the coastline, and terrain is contoured at 500 m intervals (black).

Figure 17 illustrates the spatial variability and temporal evolution of the FFDI on 7 February. During the afternoon (Figure 17(a) and (b)) the FFDI is > 75 (extreme) almost everywhere, except at high altitudes where the surface temperatures are reduced. Away from the mountains the FFDI is predominantly greater than 100, although there are embedded, localized regions of lower FFDI (still > 75). At 1300 LST, the variability appears as coherent bands that are approximately aligned with the surface flow. These mesoscale bands in FFDI are caused primarily by perturbations in wind speed associated with the horizontal convective rolls in the boundary layer; the temperature and relative humidity are responsible for broader-scale changes in FFDI associated with changes in altitude and proximity to the coast. As discussed earlier, later in the afternoon at 1500 LST (Figure 17(b)) and ahead of the cool change at 1700 LST (Figure 17(c)) the boundary layer structures evolve into more cellular patterns that result in a corresponding change to the patterns of FFDI variability. Present at all times, but more obvious at 1500 LST (Figure 17(b)), are regions of local reductions in FFDI that correspond to the lee-side of mountains, these are coincident with flow deceleration and reversal accompanying mountain wakes.

The passage of the cool change (Figure 17(c)) produces a strong spatial gradient in FFDI, with values greater than 100 ahead of the change and less than 50 behind it. The reduction in temperature and increase in relative humidity explain the reduced FFDI behind the change. Approximately 10 km behind the frontal boundary is a band of local FFDI maxima, with values around 75, that forms parallel to the front. This feature is related to the descending air behind the head of the gravity current identified in Figure 11, which leads to perturbations in surface wind speed that increase the FFDI locally. A similar feature is present later (Figure 17(d)) corresponding to the descending air in Figure 13, but much less obvious because it is complicated by the interaction of the cooler air with the terrain. Thus, given that the propagation speed of the cool air is approximately 10 m s−1, this post-frontal descent causes a transient increase in fire danger approximately 15 min after the passage of the front. In the complex terrain, this phenomenon works alongside the complexities in the direction and timing of the wind change caused by the interaction between the propagating cool air and the orography. The decrease in 2 m temperature in the early evening (Figure 17(d)) leads to a gross reduction in FFDI ahead of the cool change, however, localized regions of high FFDI are still present along the frontal boundary and remnants of the influence of coherent boundary layer structures persist.

In the previous section the passage of the nocturnal bore ahead of the front was identified and shown to cause a transient change in wind speed and direction (and other variables, Figure 16). The bore is not simply an interesting feature of the simulation and it is possible that it played a role in the Beechworth fire (VBRC, 2010). At 2234 LST, approximately 1 km from Myrtleford, Victoria (Myrtleford is approximately 90 km southeast of Yarrawonga; see Figure 14), there were reports of extreme fire behaviour and significant spotting. (Spotting is the process by which new fires are ignited by firebrands transported from the existing fire front by the winds). The bore arrived at Yarrawonga at approximately 2210 LST, and using radar imagery (not shown) we estimate that the bore arrived at Myrtleford at approximately 2230 LST. The radar imagery also shows invigoration of the smoke plume from the Beechworth fire shortly after the passage of the bore. Even though it is difficult to determine its direct role without more data, it seems likely that the arrival of the bore contributed to the change in fire behaviour around that time.

5. Summary

On 7 February 2009, ‘Black Saturday’, fires occurred over much of the state of Victoria, Australia. This article has identified the important meteorological phenomena on that day from observations and in a high-resolution numerical simulation. Particular emphasis was placed on the mesoscale phenomena that were shown to contribute to spatial variability and localized enhancements in fire danger.

The simulation was made with a nested version of the UK Met Office Unified Model in which the smallest domain had horizontal grid spacing equal to 444 m. The model reproduced the daytime maximum temperature to within a degree and the simulated cool change arrived only about 30 min late, despite the model being initiated from a global analysis 27 h earlier. The cool change was shown to resemble an unsteady gravity current that propagated from the southwest into the extremely deep mixed layer over land. Once the cool change reached the mountains to the north and west of Melbourne it was distorted by the terrain, which produced considerable variability in its local direction of propagation.

The deep mixed layer and strong winds ahead of the cool change promoted the development of organized horizontal convective rolls that extended over much of the State of Victoria. Cloud streets with similar orientation and spacing to the simulated rolls were also identified in satellite imagery, lending confidence in the representation of these structures in the model. The model also produced two nocturnal bores, one in the early morning ahead of the front and one in the evening that was coupled to the front. These bores were also observed and, remarkably, the simulated arrival of second bore was within about 10 min of the observations.

Assessing the fire danger using the McArthur forest fire danger index (FFDI) established the contribution of the mesoscale phenomena to local variations in fire danger. To the extent that the FFDI is related to the fire spread rate, it is possible that this spatial and temporal variability in the FFDI explains some of the spatial and temporal variability in the fire behaviour reported in extreme wildfires like Black Saturday. The horizontal convective rolls created organized bands of higher (catastrophic) and lower (extreme) fire danger, on horizontal scales of around 10 km. The localized descent that comprised part of the circulation of the encroaching cool air also created a coherent band of higher (extreme) fire danger after the passage of the cool change. Finally, the passage of the second nocturnal bore approximately coincided with observed increases in fire activity. Thus, mesoscale phenomena have been shown to contribute very substantially to local variations in the fire danger, which in turn may have played a key role in the extraordinary behaviour of the fires during Black Saturday. This study has not, however, examined the interaction and feedbacks between the fires and the atmosphere, which may have been important on the day. Understanding the roles of this atmosphere–fire coupling is a topic of our continuing research.

In summary, this study has identified the contribution of mesoscale phenomena to fire danger variability, and has highlighted the capabilities of future high-resolution numerical weather prediction and its application to fire weather forecasting. Although this particular simulation reproduced the occurrence of the key mesoscale phenomena, their timing and strength remain uncertain. These uncertainties could be quantified with ensemble forecasts. Examining the uncertainties of these mesoscale phenomena, and their implications for fire danger and fire spread is also a topic of our continuing research.

Acknowledgements

This work was funded by the Australian Research Council (ARC) Discovery Project Scheme (DP1093148), Todd Lane is supported by an ARC Future Fellowship (FT0990892) and Michael Rezny is supported by the ARC Centre of Excellence for Climate System Science (CE110001028). We thank the National Computational Infrastructure facility, an initiative of the Australian Government, for access to supercomputer resources. Figure 9 was provided by NASA/GSFC, Rapid Response, and the Bureau of Meteorology provided observational data. We also thank Greg Roff, Martin Dix, Margaret Kahn and Stuart Webster for their assistance with the numerical model, and David Packham and three anonymous reviewers for comments on a previous version of the manuscript.

Ancillary