Observations of a rotating pyroconvective plume

Background

To contextualise our new fire-generated wind observations, we first provide a review of previous observations of (1) plume rotation, (2) fire-induced updrafts, and (3) pyroCu and pyroCb processes.

The fire, fuels and consumption

Our observations are from a FASMEE prescribed fire in the FNF of Utah, USA, occurring on 7 November 2019. The fire, referred to here as the ‘Langdon Mountain Burn’, was conducted by FNF personnel as a high-intensity stand-replacing fire designed to remove conifers to re-establish aspen groves at high elevation (~3 km MSL). The fire was ignited by a ‘helitorch’, which drops flaming fuel into the forest. Thus, the fire is not freely evolving but composed of sequential ignitions. The first ignition occurred at 11:53 Mountain Standard Time (MST) (18:53 Universal Coordinated Time (UTC)) with subsequent ignitions progressing from east to west across the unit. Ignitions continued until approximately sunset and we estimate that the period of peak burning was at ~14:30 UTC (21:30 UTC), wherein there was a large areal expanse of high-intensity fire due to the accumulation of the helitorch ignitions. There were no black-line or previously burned fuels flanking the plot. The burn unit was ~400 ha, though the total area burned on the day of the fire was not quantified in detail, and we estimate ~200 ha burned during the period of our observations.

As summarised in Ottmar et al. (2019), the burn occurred in a forest characterised by quaking aspen overstorey and subalpine fir understorey with high fuel load. Pre- and post-fire fuel surveys revealed large loading of downed woody debris (DWD) and duff on the forest floor, with estimated total fuel loads (fine, down and standing fuels) of >90 tons acre⁻¹ (222 tons ha⁻¹, 20.6 kg m⁻²). A large fraction of these fuels, up to 81 tons acre⁻¹ (200 tons ha⁻¹, 18.3 kg m⁻²), was consumed by the high-intensity fire. Within the fuel plots, the dominant DWD and duff fuel loads experienced 99 and 76% consumption respectively. These large fuel loads and fractional consumption values are evidenced in Fig. 1, which provides photographs of pre- and post-fire fuel distribution. The corresponding large heat release from the fire was the driver for the development of deep convective plumes (e.g. Fig. 2b).

Fig. 1.

Photographs pre-fire (a, c), and post-fire (b, d) from fuel survey plots on Langdon Mountain. Photo credit: Roger Ottmar.

Fig. 2.

Overview of the FASMEE burn and instrumentation. (a) Topographic map of the Langdon Mountain Burn including terrain contours, approximate fire perimeter (red), and instrument scan sectors (according to the legend). (b) Photograph showing the San Jose State University (SJSU) lidar truck (left) and radar (right) with deep plume development in the background. (c) University of Nevada, Reno (UNR) lidar site. (d) Photograph of the plume near the end of the day from the UNR lidar site.

In addition to manual fuel plot surveys, pre- and post-fire airborne laser scanning (ALS) data reveal the spatial distribution of consumption and consumed percentage throughout the burn unit. Following the methods in McCarley et al. (2020, 2022), these ALS data are fitted with fuel plot observations on the ground (e.g. those in Fig. 1) to produce maps of estimated fuel loads both pre- and post-fire, which subsequently can be differenced to provide mapped consumption estimates. The ALS-based data indicate pockets of very large fuel consumption of >20 kg m⁻², the distribution of which is examined in relation to plume processes in the Results section

Instrumentation

The Langdon Mountain Burn was instrumented with scanning remote sensors to observe the plume development. The instrument configuration is shown in Fig. 2 and primary instruments and observing strategies are described below.

Observational strategy

An overview of the radar and lidar scan domains is provided in Fig. 2a. The UNR lidar conducted range-height-indicator (RHI) scans (i.e. fixed azimuth, varying elevation) centred on the upright portions of the isolated plumes during the early phase of the fire (~19:00–20:00 UTC; recall ignition was at ~18:50 UTC), then transitioned to hemispheric plan-position-indicator (PPI) scans (i.e. fixed elevation, varying azimuth) wherein a sector scan was conducted at 5° elevation increments from 0° to 55° above horizontal. The RHI scans were acquired at ~1 min intervals. The PPIs capture the broader plume development associated with continual helitorch ignitions throughout the afternoon. The lidar required ~5 min to complete a full sequence of hemispheric PPI scans. A few additional RHI scans were also conducted at ~21:00–21:10 UTC to document the plume’s condensation level. The SJSU lidar primarily conducted RHI scans centred on the upright portions of the plume throughout the fire. The SJSU Ka-band radar conducted RHI scans for most of the observing period, but also included some hemispheric RHI and PPI scans later in the burn (at ~21:30 UTC), which we use to produce plume volumes during the approximate period of peak burning (i.e. largest areal expanse of fire generating deep plumes). The radar RHI scans are much faster than the lidar scans, with acquisition rates of 10–15 s for a full scan.

Ancillary data

In addition to the deployed instruments, we used GOES17 IR and visible satellite imagery and reference radiosonde observations from Grand Junction, CO, which is ~300 km east of the burn. The radiosonde data are provided in Appendix 1.

Plume evolution

The temporal plume evolution is established with a combination of radar (Fig. 3) and satellite analysis (Fig. 4).

Fig. 3.

Radar-derived time–height overview of the plume development. (a) Maximum reflectivity (shaded), plume top height (black line), and plume condensation level (green line). (b) 95% of radial velocity (shaded), condensation level (green line), and constant ascent lines (black dashed) for 10, 20, 40 m s⁻¹ every 30 min. (c) 5% of radial velocity (shaded), condensation level (green line), and constant descent lines (black dashed) for −5 m s⁻¹ every 30 min. Heights are listed in metres above sea level (MSL) and the ground surface is between 2700 and 3000 m near the plume base, such that the lowest radar data are just above the surface.

Fig. 4.

Satellite overview of the pyroCu/Cb development. (a–f) Visible satellite imagery and approximate fire contour (red line) based on a 375 K threshold at 3.9 µm. (g) Time-series of the 8.4 µm (dashed red), 3.9 µm (solid red line) brightness temperatures (left axis) along with the plume top height from the radar data (black line, right axis). The homogenous nucleation temperature (−38°C) is shown in a dashed blue line and the high t of the condensation level is shown in dashed green.

A time–height diagram of radar reflectivity (Fig. 3a, drawn from individual RHI scans) shows plume tops ascending from 4 to 5.5 km MSL (1–2.5 km AGL) in the hour following ignition (~18:50 UTC), then deepening to ~10 km MSL (7 km AGL) by 20:55 UTC via a series of step-like pulses. The condensation level, detailed later, is at ~5.5 km MSL (2.5 km AGL) such that there is considerable deep pyroCu and potentially pyroCb development by 20:20 and 20:50 UTC, respectively. These data also indicate episodic high-reflectivity cores rising through the plume, manifest as streaks sloping up and to the right due to parcel ascent with time. These reflectivity cores/streaks are likely due to higher pyrometeor loading in the more buoyant updrafts, though measurements of buoyancy are not available to test this assumption. The reduction in reflectivity in the lower plume after 20:30 UTC is in part due to the sloping nature of the convection such that the base of the plume core was not always centred in the RHI scan plane (though it is captured in volumetric scans shown later).

Fig. 3b, c shows the corresponding 95 and 5% radar radial velocities. The 95th percentile data capture the strongest outbound flows, which are dominated by updrafts. The 5th percentile data correspond to weakest outbound, or strongest inbound, flows including downdrafts. As these data are extracted from the RHI scans, the velocities at a given height are measuring different components of the flow: higher-elevation angles more closely measure the vertical velocity, whereas lower scan angles better sample horizontal flow.

The 95% velocity data indicate enhanced outbound flow (darker reds in Fig. 3b) in streaks like those in the reflectivity data, indicative of stronger updrafts in individual convective elements. The streak’s slopes (i.e. time vs height) correspond to the rise rate of the convective elements. For reference, we have added 10, 20 and 40 m s⁻¹ constant ascent lines (black dashed lines) at 30-min intervals throughout the diagram. Most slopes are between 15 and 30 m s⁻¹, with lower velocities close to the surface and near the plume top, due to accelerations and decelerations, respectively. The quasi-constant slopes in the mid-plume suggest an approximate balance between accelerations and decelerations in that region (e.g. buoyant acceleration vs drag forces).

The ‘downdraft’ data (i.e. 5%, Fig. 3c) also show streaky distributions and substantial variation with height. The same ‘up-and-to-the-right’ streaks are apparent in the lower plume (<5500 m), especially at 20:00–20:30 UTC. The ascent of the negative velocity features is due to mechanically forced circulations linked to the updrafts themselves. Inspection of individual RHIs (see Supplementary Video S1) shows that features are horizontal-axis vortices that are well known features of convective elements in both cumuli and wildfire plumes (Lareau and Clements 2017; Clements et al. 2018; Lareau et al. 2018; Rodriguez et al. 2020).

Above cloud base (5.5 km), broader ‘down-and-to-the-right’ swaths are observed (Fig. 3c). This diagram includes descent lines for −5 m s⁻¹ for reference. These downdrafts within the pyroCu layer are likely linked to two sources of negative buoyancy: (1) latent cooling due to evaporation of cloud condensate, and (2) overshooting convective elements that become negatively buoyant in their stable surrounding. Notably, none of these downdrafts sink back to the surface and thus they likely do not influence fire processes. This contrasts with the idea of ‘plume collapse’ wherein strongly evaporatively cooled parcels in pyroCb with precipitation or virga can sink to ground level. Thus, even though portions of the pyroCu are subsiding and evaporating, in this case it is not ‘collapsing’. This distinction has important practical implications for firefighters who may mistake a subsiding pyroCu with a more dangerous pyroCb producing downdrafts that can impact the fire ground (e.g. Thurston et al. 2015).

GOES17 visible and IR (3.9 and 8.4 μm) imagery (Fig. 4a–f) and time series data (Fig. 4g) capture the onset and deepening of the pyroCu and the intensification of the fire corresponding to the plume deepening observed by radar. The visible imagery indicates a transition from a shallow smoke plume with shallow pyroCu (Fig. 4a, b) to growing pyroCu ~20:30 UTC (Fig. 4c). The growing pyroCu is consistent with what the World Meteorological Organisation (WMO) would categorise as a ‘cumulus congestus flammagenitus’ (WMO 2023), which some researchers and meteorologists refer to as towering or deep pyroCu. This timing corresponds with the sustained plume deepening above the condensation level apparent in the radar time–height diagram (Fig. 3a). The vertical and horizontal extent of the deepening pyroCu increases considerably at 20:57 UTC (Fig. 4d), corresponding to the period of radar-observed plume tops of ~10 km. Thereafter, a deep pyroCu (perhaps a pyroCb as discussed below) persists, with smoke, ash and cloud detraining aloft into the ambient northwesterly flow (Fig. 4e, f). We note that the lower reflectivity values during this time may be a result of RHI scans missing some of stronger convective cores near the surface, which were spread over a broader domain later in the fire, making instrument pointing more difficult.

Although we have direct measurement of the plume depth via radar, it is still useful to examine the plume top temperature via the 8.4 µm brightness temperature (blue line in Fig. 4g). The cloud top cooling starts at ~20:20 UTC, reaching a minimum temperature of 236.7 K (−36.4°C) at ~21:40 UTC, which is close to the homogeneous freezing temperature (−38°C, 235.15 K), which is the widely accepted pyroCb threshold due to de facto glaciation (Peterson et al. 2017a). It is worth noting that these GOES data likely overestimate the cloud top temperature (i.e. underestimate height) owing to the small scale of the plume relative to the large IR pixels (~2 × 2 km). Indeed, using the radar-observed plume height (~10 km) and the nearest available sounding data (Grand Junction, CO, which is ~300 km to the east), the plume may have reached −50°C (see sounding data in Fig. A1).

Such a cold cloud top associated with a spreading anvil would categorise this cloud as a pyroCb in some frameworks, for example the approach of Peterson et al. (2017a, 2017b) based on the homogeneous freezing temperature. However, there was no lightning detected or thunder heard and both radar and visual inspection did not indicate any precipitation processes, which is relevant to the preceding discussion of the absence of downdrafts reaching the surface. As such, it might be best to consider this event as culminating in towering or deep pyroCu. Accordingly, we refer to the cloud as towering pyroCu throughout the remainder of the manuscript, though reserve the possibility that this cloud may be classified as a pyroCb in some settings.

To summarise changes in the fire intensity during the plume and cloud growth, we examine the 3.9 µm brightness temperature (solid red line, Fig. 4g), which is sensitive to fire thermal emissions (Dozier 1981). These data first indicate warming beyond background temperature at 19:10 UTC, which is ~20 min after the helicopter ignition. This lag is due to the pixel resolution and sensor sensitivity, which often require the fire to grow upscale or intensify before detection, though hot fires can sometimes be detected very quickly (Lindley et al. 2016, 2020). A subsequent rapid increase in 3.9 µm brightness temperature occurs at 20:00 UTC, shortly preceding the development of the deep pyroCu. The 3.9 µm brightness temperature eventually plateaus owing to sensor saturation at ~415 K, with some decreases due to pyroCu obscuration.

Based on these combined radar and satellite data, we next subdivide our detailed plume analyses into three phases: (1) early plume rise (18:50–20:00 UTC); (2) onset of rotation (20:00–20:20 UTC); and (3) towering pyroCu processes (20:50 UTC onward).

Early plume rise

Following ignition, an isolated, single-core, upright plume developed as observed by the radar (Fig. 5a–e, k–o) and lidar (Fig. 5f–j, p–t). Note that these sensor’s RHI scan planes intersect the plume at somewhat different angles (Fig. 5u). A few key points emerge from inspection of these data.

Fig. 5.

Summary of the early plume development as observed by the SJSU radar and UNR lidar. (a–e) Radar reflectivity. (f–j) UNR lidar attenuated backscatter coefficient. (k–o) Radar radial velocity. (p–t) Lidar radial velocity. (u) Terrain map showing instrument RHI scan planes. (v) Photograph of the plume development from the UNR lidar vantage point.

First, the radar and lidar provide similar information about the plume. Contemporaneous radar reflectivity (Fig. 5a–e) and lidar backscatter (Fig. 5f–j) both show a ‘starting plume’ (Turner 1962) characterised by: (1) a broadening ‘plume cap’ ascending gradually with time, (2) plume width broadening with height, (3) lobes of smoke and ash hanging from the plume edges (e.g. ring vortices), and (4) ash and smoke concentration decaying with height owing to entrainment dilution. Despite the different scan angles, many of these features are common between the radar and lidar, indicating a quasi-symmetric structure, which makes sense given the isolated nature of the combustion near the time of ignition.

There are, however, some differences in the radar and lidar observations including: (1) lidar sensitivity to diffuse smoke outside the plume core, and (2) differences in the upper plume due to wind shear advecting the upper plume eastward out of the lidar scan plane while remaining in the radar’s view (compare Fig. 5e, j). This shear is also evident in the photograph (Fig. 5v), where a separation between the upper and lower plume is evident. The photograph also shows a shallow (tens of metres) initial pyroCu developing at the plume top. The radar and lidar data place the plume top at ~5500 m, which we will later confirm is the plume’s condensation level.

Both sensors also reveal similar kinematic structures throughout the plume (Fig. 5k–t). For example, comparing panels n and s in Fig. 5, outbound flow >20 m s⁻¹ along the ‘back edge’ of the plume and extending into the broadening plume cap is apparent in both the radar and lidar data. Despite these strong updrafts, the plume cap is only ascending at a rate of ~6 m s⁻¹ (inferred from change in plume top heights between panels m and n). This is consistent with the established dynamics of a starting plume wherein ‘the cap moves with a constant fraction of the plume velocity’ (Turner 1962). To maintain mass continuity, the plume fluid must thus spread laterally at the cap, forming the broadening plume top apparent in the observations.

One final feature of note is the weak ‘clear air’ inbound flows (1–3 m s⁻¹) apparent in the lidar observations between the lidar location and the plume base (blue shading in Fig. 5p–t). These data show that the ambient flow in this region is weak and has not yet been perturbed by the plume-base convergence, which would necessitate outbound flows (rather than the noted inbound flow) toward the plume. As we will show in the next section, as the plume further develops, a profound modification of this flow occurs.

Development of plume rotation

Amongst the notable aspects of the Langdon Mountain Burn was the development of strong ‘whole-column’ anticyclonic plume rotation and ~10-fold increase in horizontal winds near the plume base as compared with that in the ambient environment. Fig. 6 summarises the plume structure and rotation at 20:14–20:15 UTC, and a time-lapse video is provided in the Supplementary material (see Supplementary Video S1, courtesy of Holt Hanley, SJSU).

Fig. 6.

Overview of plume rotation. (a) Annotated photograph showing the anticyclonic rotation. (b) Lidar attenuated backscatter isosurfaces showing the plume structure. (c) Out-bound (red) and in-bound (blue) radial velocity isosurfaces on top of the backscatter isosurfaces. (d, e, f) Radial velocity (shaded) and backscatter (contoured) lidar PPI scans through the plume base at 5°, 10° and 15° elevation tilts.

The first panel (Fig. 6a) shows a still photograph of the plume annotated with the sense of rotation derived from the video. Also apparent is the helical nature of the updraft, with individual convective components spiralling upward. The corresponding lidar-volume rendering (Fig. 6b) shows an upright plume core with a slight tilt toward viewers’ left (to the east in physical space) and two plume components: (1) the main updraft column, which is closer to the lidar, and (2) a more distant plume element wrapping around the back of the primary plume core. These features are also apparent in the photograph. The sense of rotation is subsequently visualised by including red and blue isosurfaces at 5 and −1.5 m s⁻¹ respectively (Fig. 6c). The blue (inbound) and red (outbound) surfaces ‘hug’ the viewer’s right and left sides of the plume, respectively, indicating that the whole column is involved in distinct anticyclonic rotation centred on the plume core. The region of outbound flows is broader than the inbound flow, though we only have data where sufficient smoke and ash scatters exist, so it is possible that the inbound flow extends beyond what is shown.

Quantification of the rotation is obtained using lidar PPI snapshots at 5°, 10° and 15° elevation slices through the plume base (Fig. 6d–f). Each slice indicates a couplet of in- and out-bound motions centred on the plume core. In each panel, we also show the maximum in- and out-bound tangential velocities (V_x, V_n) and the rotational wind magnitude (V_rot, Gibbs 2016, see details in the background sections above). The strongest V_rot and V_x,n occur at the 10° elevation tilt, reaching magnitudes of 13.3 and 15.1 m s⁻¹. These values are notable in that they are an order of magnitude stronger than the ambient winds (~1.5 m s⁻¹ measured by the scanning lidar away from the fire and shown in Fig. 5p–t), and thus represent a significant fire-generated wind that may feed back on the fire development and intensity, though we lack observations of the combustion processes to test this hypothesis. The plume diameter is ~500 m and the tangential wind maxima are separated by ~150 m. The implied vorticity (ζ) is −0.34 s⁻¹, which is slightly less than the 0.4 s⁻¹ vortex documented in Clements et al. (2018) and much less than the large rotational velocities in FGTVs (e.g. V_rot  38 m s⁻¹ in the Carr Fire; Lareau et al. 2018).

The link between the rotation and the broader plume structure is established using a combination of lidar and radar observations (Fig. 7). The time-averaged radar reflectivity and radial velocity RHIs (Fig. 7a, b) indicate the merging and contraction of two high-reflectivity plume cores in the lowest 500 m AGL. The radial velocity data confirm the converging character of the lower plume, showing ‘blue’ inbound flow in the more distant plume core and ‘red’ outbound flow in the closer plume core (Fig. 7b). Because of this merging, the plume geometry exhibits narrowing with height, reaching a ‘waist’ or ‘neck’ at ~3500 m MSL, before broadening with height aloft. We note this ‘neck’ height can be an important consideration for plume rise modelling (e.g. Tory 2018). A photograph of this merging and contracting, sometimes called ‘indrafting’ (Finney and McAllister 2011), is provided in Fig. 8 albeit at a slightly earlier time.

Fig. 7.

Summary of plume structure leading to increasing rotation. (a, b) Time-averaged radar reflectivity and radial velocity. The inset boxes in (b) are used for the in- and out-bound flow components. (c) in- and out-bound flow in the lower plume along with the estimated along-radial convergence (purple). (d–u) Lidar PPI scans for the 5° (red frames), 10° (blue frames), 15° (purple frames) lidar elevation scans. Shading shows the radial velocity and contours the attenuated backscatter coefficient. The black dashed line indicates the radar scan plane. The lidar tilts are also shown in (a) relative to the deeper plume structure.

Fig. 8.

Example of in-drafting causing a plume merger above the surface.

The temporal variation in the magnitude of the convergence is estimated using the mean radial velocity in two boxes centred on the merging plume elements (red and blue boxes in Fig. 7b; data shown in Fig. 7c). These velocity data show that both the outbound (red) and inbound (blue) flow components increase in magnitude between 20:11 and 20:14 UTC, reaching contemporaneous peaks of ±7 m s⁻¹, respectively. The confluent component of the flow is approximated as the bulk difference between the in- and outbound flows (magenta line, Fig. 7c), and thus reaches a peak of ~14 m s⁻¹.

Given the context of converging plume components, it is useful to examine the sequence of lidar radial velocity and backscatter PPI scans at successive elevation tilts (Fig. 7d–u). For ease of interpretation, the 5°, 10° and 15° elevation tilts are organised in rows, with the axes colour-coded (red,  5°; blue,  10°; magenta,  15°). There are three primary features in these data: first (moving from left to right in the figure), the strength of the rotation increases during the period of plume convergence, reaching a peak at 20:14 UTC. Second, the lidar backscatter indicates the rotation coincides with a period of interaction between plume elements (labelled A, B). Element A is a plume from the earlier ignitions whereas Element B is from a new helitorch ignition at ~20:00 UTC. This ignition rapidly intensifies, as apparent in development of higher attenuated backscatter and contemporaneous perturbation of the winds (e.g. Fig. 7e–h). As the new plume (B) develops, the older plume (A) is drawn inwards toward the new plume. This is apparent as the decreasing gap in backscatter between the plume elements with time. It is not immediately clear why ignition B increases more rapidly than previous ignitions.

The third aspect of the plume development and rotation apparent in these data is the contraction of the plume with height into the aforementioned ‘waist’ or ‘neck’ region and the associated tightening of the circulation. Taken together, these lidar PPIs show the complex fire geometry near the surface (i.e. spatial distribution of backscatter) is reduced with height, yielding a single plume core aloft with a radius of ~500 m (compare Fig. 7 panels h, n, and t). For reference, the 5°, 10° and 15° lidar scan planes are superimposed on the radar figure (Fig. 7a), showing that these features match up nicely with the contracting and merging plume elements (note that the lidar backscatter contours in Fig. 7d–i provide a proxy for the location of the active fire at this time).

Collectively, these data suggest that the interaction of plumes from discrete combustion regions (and thus complex fire line geometry) and ‘indrafting’ are important in generating the whole-column plume rotation in this case. This appears to contrast with cases of flow–plume interactions during stronger winds that seem to favour counter-rotating vortex pairs (CVPs) associated with recent FGTVs (Lareau et al. 2022). However, we lack detailed environmental wind observations that could further support or refute this hypothesis. Obtaining more comprehensive wind speed measured in complex terrain should be included in future campaigns if possible.

Towering pyroCu processes

Plume condensation level (PCL) is an important threshold and lower boundary condition for pyroCu and pyroCb initiation. We estimate the PCL by comparing lidar (Fig. 9a) and radar (Fig. 9b) volume renderings of the deep pyroCu-topped plume at ~21:30 UTC. The radar data reveal the full plume structure, including the sloping pyrometeor-laden updraft cores rising from the surface and the broadening pyroCu aloft. The lidar, in contrast, captures only the lower portions of the plume because it is attenuated in liquid water droplets above the PCL. Fig. 9c reinforces this interpretation by displaying radar and lidar maximum reflectivity (shaded) and backscatter (contoured) plume cross sections. These data indicate very good agreement in the lower plume but a contrasting representation aloft due to attenuation of the lidar beam along the cloud base and edges (e.g. ~5500 m for the downwind portion of the plume).

Fig. 9.

Summary of lidar (a), and radar (b) volumetric plume observations. (c) Maximum reflectivity (shaded) and backscatter (contoured) in the x-direction. Black dashed line shows the plume’s condensation level.

A more detailed examination of the condensation level is accomplished by extracting radar and lidar RHI scans at 21:15 UTC (Fig. 10). All three instruments show similar updraft core structure in the lower plume, but both lidars attenuate at ~5500 m in the downwind portion of the plume (green dashed line). There is a local enhancement in the lidar backscatter where it reflects from cloud droplets and before it is attenuated (see inset in Fig. 10a). This signal is similar to that observed at the PCL in other pyroCu/Cb-topped plumes (see Lareau and Clements 2016). The radar data, in contrast, span the transition into the cloud and continue to cloud top (Fig. 10b). This is because the radar beam is not attenuated in the cloud and retains strong backscatter from the mixture of pyrometeors and hydrometeors aloft.

Fig. 10.

RHI scans examining plume condensation level. (a) UNR lidar RHI scan; (b) SJSU ka-band radar RHI; (c) SJSU lidar RHI. Shown in each panel is the estimated condensation level (green dashed line). The inset panel shows the maximum UNR lidar backscatter as a function of height.

One might expect that the polarimetric radar observations would reveal changes in the shape of scatters proximal to the condensation level owing to the presence of quasi-spherical cloud droplets commingled with irregular ash particles. This is not, however, observed. As was shown in a previous analysis of polarimetric properties of the same plume in Aydell and Clements (2021), no significant change in polarimetric variables occurs above the condensation level. This lack of differentiation is attributable to the small size of the cloud droplets, and the lack of large precipitation-sized hydrometeors, as compared with the abundant larger pyrometeors lofted into the cloud by the powerful fire-generated updrafts. Thus, in the present, non-precipitating towering pyroCu, the lidar signal attenuation remains the most useful method for diagnosing cloud base. The lack of precipitation processes is also likely linked to the absence of lightning in this pyro-cloud.

The kinematics of the cloud layer are examined using the time-maximum (red/grey shades) and -minimum (blue contours) radial velocity for RHI scans spanning 20:51–21:15 UTC (Fig. 11a). These data show a kilometre-wide updraft core with outbound speeds >25 m s⁻¹ flanked by downdrafts on the plume edges. The cellular distribution of the strongest updrafts, and assessment of corresponding animations (see Supplementary Video S2), suggests the plume possesses a ‘chain of thermals’ comprising individual buoyant elements such as occur in cumulus convection (Morrison et al. 2020). This central-updraft, flanking-downdraft structure is further reinforced by examining 3D updraft and downdraft isosurfaces (Fig. 11b) and is broadly consistent with other observations of deep wildfire convective plumes (e.g. Lareau and Clements 2017; Rodriguez et al. 2020). The maximum in-cloud outbound speed is ~35.5 m s⁻¹ at ~6200 m MSL, indicating extreme rising motion within the cloud layer. The updrafts decay above ~7000 m MSL, likely owing to overshooting their level of neutral buoyancy and interaction with stronger winds aloft. This decay was also evident in the time–height diagram (Fig. 3b).

Fig. 11.

Overview of plume updraft and downdraft structures. (a) Time-maximum (red and grey shading) and -minimum (blue contours) radial velocity. Also shown are the reflectivity contours (black lines), the approximate updraft trajectory (white dashed line), plume condensation level (green dashed line), and radar scan radials (light grey lines). The inset axes show the maximum velocity as a function of height (red line) and the estimated flow along the plume trajectory (purple line). (b) Volume rendering of the updraft (orange and red) and downdraft (blue) isosurfaces along with the −20 dbZ reflectivity surface (transparent grey). Also shown is the estimated fuel consumption as coloured shading along the terrain surface according to the colour bar.

The outbound flow at cloud base is >25 m s⁻¹, which implies considerably stronger updrafts due to the beam angle of <45° (see grey dashed lines showing beam scan angles). To account for this underestimation of the pure updraft, we use the slope of the updraft core (white dashed line Fig. 11a) and compute the flow along that line that would result in the observed radial velocity component at each height in the updraft core (inset panel). The results suggest that the sub-cloud updrafts exceed 35 m s⁻¹. These values are contextualised by previous observations by Banta et al. (1992) and Rodriguez et al. (2020) of sub-cloud updrafts in pyroCu/Cb plumes of 24  and 30–60 m s⁻¹, respectively. These cloud base updrafts are notable in that their inertia can overcome appreciable convective inhibition and thus drive pyroCu/Cb development even in unfavourable thermodynamic conditions.

Downdrafts (blue contours) are preferentially observed above the condensation level and along the flanks of the updraft core, with peak values of ~10–11 m s⁻¹, which is approximately one-third of the maximum updrafts. The downdrafts along the leading plume edge are confined to a narrow (~500 m) corridor whereas the downdrafts on the downwind side of the plume occur throughout a broader ~2 km region. The occurrence of downdrafts in the cloud layer suggests a few processes likely contribute to their occurrence. First, mechanically forced horizontal axis eddies along the plume edge likely produce some of these downdrafts (inspection of Supplementary Video S1 shows these clearly). Second, evaporation into the dry air aloft (see sounding in Fig. A1) of the characteristically small pyroCu/Cb cloud droplets likely produces negative buoyancy, like subsiding shells in ordinary cumulus (e.g. Heus and Jonker 2008). Lastly, the plume is likely overshooting its level of neutral buoyancy, causing some downwind sinking of negatively buoyant plume.

Linking plume processes to fuel consumption

Fig. 12 summarises the pre-fire fuel distribution (satellite image and fuel map, Fig. 12a, b) and the spatial distribution of fuel consumption estimated from ALS observations (Fig. 12c). These data show a mix of heavier fuels (e.g. greens correspond to conifer stands intermixed with aspen overstorey in Fig. 12a) and fine fuels in meadows (beige regions in Fig. 12a). Within the forested regions, pre-fire ALS data indicate total fuel loads up to and exceeding 40 kg m⁻² (colour shading Fig. 12b). The fuel consumption (Fig. 12c) indicates spatially heterogeneous consumption patterns corresponding to foci of the helitorch ignitions progressing from east to west across the burn unit. Unfortunately, we do not have quantification of the ignition timing and extent, something that future experiments should endeavour to provide. The statistical distribution of consumption is shown in the histogram in Fig. 12e. The median and mean consumption values amongst burned pixels (i.e. with non-negligible consumption) were 5.8 and 6.8 kg m⁻², respectively, and more intensely burned pixels (e.g. 75% values) exhibit consumption >11 kg m⁻². Notably, there are three regions (coloured outlines in Fig. 12b) with local consumption maxima >20 kg m⁻².

Fig. 12.

Summary of fuel consumption and plume processes. (a) Google imagery of pre-fire fuels. (b) Terrain map overlaid with ALS-estimated pre-fire fuel loads (coloured contours). (c) ALS-estimated fuel consumption and estimated sensible heat flux (see values in parentheses). (d) Time-maximum lidar attenuated backscatter. Each panel also shows the terrain contours at 10 m (light black) and 50 m (solid black) intervals. (e) Distribution of fuel consumption and estimated fluxes.

To link consumption to plume processes, we estimate the sensible heat fluxes, which were not explicitly measured. This is accomplished using the ALS-estimated fuel consumption along with an estimated consumption time scale (Δt) of ~1 h using:

where H_s is the sensible heat flux (W m⁻²), w is the total ALS-derived fuel consumption (kg m⁻²) burned in the increment Δt (3600 s), FMC is the fuel moisture content (~0.2; see fig. 18 in Ottmar et al. 2019), and h is the heat content of dry fuel (18 700 J kg⁻¹). The sensible heat flux values, shown in parentheses for the map and the inset histogram, indicate peak fluxes of ~90 kW m⁻² associated with the local maxima in consumption. These values are consistent with those estimated for other deep convective columns from wildfires (Lareau and Clements 2017). Noting that our time scale is a crude estimate, it is likely that instantaneous fluxes were much higher.

Next, we examine the spatial distribution of time-maximum lidar attenuated backscatter from the 10° elevation PPI scans throughout the burn, which are slightly above the surface (Fig. 12c). These data indicate that the local maxima in backscatter from plume cores agree well with the consumption maps, albeit with plume cores displaced somewhat east–southeast (ESE) from the consumption maxima. This ESE tilt to the plume is particularly evident in the large consumption regions in Region A, which corresponds with the region of strong rotation and the slightly tilted plume shown in Fig. 12a, b. We note that the consumption maxima in Region C are not linked to the strong backscatter signal because they occurred after the lidar and radar observations concluded for the day.

Finally, the consumption maxima in Regin B can also be directly related to vigorous plume cores observed with the radar and the deepest plumes of the afternoon. Returning to Fig. 11b, note that the updraft core locations are well correlated with the independently derived ALS consumption data, which are mapped onto the terrain using coloured markers. Note the consumption values within this region are not only large, but are also spatially expansive compared with the values in Region A. We hypothesise that this broad region of intense combustion was a critical factor in producing the deep pyroCu plume tops reaching 10 km and updrafts exceeding 35 m s⁻¹. In future observing campaigns, resolving heat fluxes (e.g. fire radiative power) as a function of time along with the consumption and plume rise observations should be attempted, thus providing data sets that can be used for comprehensive model validation and refinement.