Fire-Generated Tornadic Vortices

Fire-generated tornadic vortices (FGTVs) linked to deep pyro-convection, including pyrocumulonimbi (pyroCbs), are a potentially deadly, yet poorly understood wildfire hazard. In this study we use radar and satellite observations to examine three FGTV cases during high impact wildfires during the 2020 fire season in California, USA. We establish that these FGTVs each exhibit tornado-strength anticyclonic rotation, with rotational velocity as strong as 30 m s -1 (60 kts), vortex depths of up to 4.9 km AGL, and pyroCb plume tops as high as 16 km MSL. These data suggest similarities to EF2+ strength tornadoes. Volumetric renderings of vortex and plume morphology reveal two types of vortices: embedded vortices anchored to the fire and residing within high reflectivity convective columns and shedding vortices that detach from the fire and move downstream. Time-averaged radar data further show that each case exhibits fire-generated meso-scale flow perturbations characterized by flow splitting around the fire’s updraft and pronounced flow reversal in the updraft’s lee . All the FGTVs occur during deep pyroconvection, including pyroCb, suggesting an important role of both fire and cloud processes. The commonalities in plume and vortex morphology provide the basis for a conceptual model describing when, where, and why these FGTVs form.

Recent exemplars of these extremes include California's Carr Fire in 2018, which produced pyroCb and a deadly pyrogenetic tornado with winds >140 mph (Lareau et al. 2018), and the Loyalton Fire in 2020, which necessitated the first-ever National Weather Service (NWS) fire tornado warning (Cappucci 2020).
Despite their impacts, the dynamics of Fire-Generated Tornadic Vortices (FGTVs) are not well established, having only been comprehensively documented in two cases to date (Fromm et al. 2006;McRae et al. 2013;Lareau et al. 2018). For example, it is not understood where in the fire FGTVs form, how they are linked to the convective plume and vigorous pyro-convection, including pyroCb, and how consistent their radar signatures are from one event to the next. This knowledge gap motivates this paper, which establishes commonalities in the location, morphology, and evolution of FGTVs during three high impact wildfires.

Understanding Vortices Generated by Fires
Fire Generated Vortices (FGVs) span many spatial, temporal, and intensity scales (Forthofer and Goodrick 2011;Tohidi et al. 2018). While FGVs can have both vertical and horizontal axes of rotation (e.g., fire-whirls vs. horizontal roll vortices, Haines and Smith 1987), the focus of this study is on FGVs with predominantly vertical axes. Small FGVs (~10 m) are Unauthenticated | Downloaded 04/22/22 01:34 AM UTC near the Convective Condensation Level (CCL; Lareau and Clements 2016), and more precisely is determined by the plume's temperature and moisture (Tory et al. 2018). Updrafts near pyroCb cloud base can be as high as 60 m s -1 (Rodriguez et al. 2020) and plume tops can penetrate the stratosphere (Fromm et al. 2006;2010;Peterson et al. 2021). Accordingly, vigorous pyro-convection, including pyroCb, have been linked to violent firestorms (Fromm et al. 2006;Peterson et al. 2015;Peace et al. 2017;Terrasson et al. 2019) and FGTVs, wherein it is hypothesized that pyroCbs provide enhanced column stretching that contributes to FGTV spin up (Cunningham and Reeder 2009;McRae et al. 2013;Lareau et al. 2018).
While there are strong indications that "jet in a crossflow" dynamics and vigorous pyroconvective processes both contribute to FGTV development, to date there have been few observations of vortex and plume morphology with which to confront these theories. This sets the stage for the analyses that follow.

Radar Data
NEXRAD radar data are used to quantify wildfire plume processes, including FGTV winds.
These 10-cm wavelength radars are sensitive to the large (mm-cm scale) particulate ash and debris, called pyrometeors, lofted in wildfire convective plumes (McCarthy et al. 2019). The metadata for the radars used are included in Table 1. For analyses of three-dimensional plume structures these radar data are interpolated to common cartesian grids whereas for analyses of the near surface winds data are kept on a native polar grid (azimuth, range). Some of the 7 Accepted for publication in Bulletin of the American Meteorological Society. DOI 10.1175/BAMS-D-21-0199.1. velocity data are aliased, requiring an algorithmic and manual dealiasing (See Supplemental Material S1).
After dealiasing, FGTV strength is quantified using the rotational velocity, given by where and are the strongest out/inbound radial velocities, respectively, proximal to the vortex center, which is manually determined (Gibbs 2016).
is correlated with, but different from, the actual vortex strength. Due to the need for dealiasing there is some uncertainty in the values. Examples of the dealiasing and its uncertainties are available in S1.

Satellite Data
Data from GOES17 are used to examine fire and plume processes. We use a "Fire-RGB" approach, which blends data from the near-infrared (1.6, 2.2, 3.9 µm) channels and allows viewers to differentiate between more and less intense fires (red is cooler, white is hotter). (https://rammb.cira.colostate.edu/training/visit/quick_guides/Fire_Temperature_RGB.pdf).
Similarly, we use "true-color RGB" imagery to examine smoke and pyroCb processes. The true color images combine data from the 0.47 µm (blue), 0.64 µm (red), and 0.86 µm ("veggie") channels. The spatial resolution of the fireand true-color-RGB data are 2 and 1 km, respectively.

Ancillary Data
Data from the high-resolution rapid refresh (HRRR; Benjamin et al. 2016)  radiosonde are also used in the case study of the Loyalton Fire. Fire perimeter data are obtained from the national infrared observations program (NIROPs).

Embedded and Shedding Vortices During the Loyalton Fire
The lightning started Loyalton Fire consumed ~20,000 acres (8100 ha) on 15 August 2020, yielding a deep pyroCb and a sequence of FGTVs (Table 2). The fire's growth occurred during southwest surface winds, which backed with height, becoming more southerly in the mid-troposphere (Fig. 2c). The thermodynamic environment was conducive to elevated convection ( Fig. 2a,d) and consistent with the climatology of pyroCb environments (Peterson et al. 2017a). Namely, there is a deep, dry, well-mixed layer extending from the surface to ~550 hPa, which is conducive to active fire behavior and vertical plume development, overtopped with a moist mid-to-upper troposphere, which is supportive of moist-convection.
The evolution of the Loyalton Fire's FGTVs and pyroCbs are summarized in Fig. 3 (also see animation S4). The time-height diagram of radar reflectivity (Fig. 3a) indicates rapid plume growth from 6.5 to ~13 km MSL. During the plume deepening, cores of high reflectivity air (>30 dbZ) ascend with time, indicative of vigorous convective updrafts.
Noting that the CCL was at ~5 km (black dashed line in Fig. 3a), the entire upper portion of the plume was involved in deep-moist convection, as is apparent from photographs ( Fig. 3c) and satellite imagery (Fig. 3d) corresponding vortex depths were notable, with one vortex (~2035 UTC) reaching ~6.5 km MSL (4.9 km AGL), and multiple vortices extending above the condensation level (see Supplemental Information S1). This means that some, but not all, of the vortices extend from the surface into the pyroCb.
Radar snap shots of the strongest vortices at 2030, 2125, and 2205 UTC (Fig. 4) indicate distinct in-and outbound velocity couplets (Fig. 4b,d,f) near the advancing left flank of the head fire (black dashed lines; Fig. 4a,c,e). These radar data are from KRGX's second lowest scan elevation (0.5°), yielding 500-1500m AGL radar beam heights in the vicinity of the FGTVs. The first two FGTVs were anchored to the head fire and reside within high reflectivity updraft cores ( Fig. 4a,b,c,d, see also movie S2). In contrast, the third vortex was detached from the fire, residing in a lower reflectivity region downstream (i.e., to the northeast; Fig. 4e,f, see also movie S3). These FGTV and plume morphologies are also apparent in the 3D plume structure, as shown with radar reflectivity iso-surfaces and vertical vortex lines (Fig. 5c,d). These data indicate that the convective plume is bent over in the wind, with evidence for bifurcation (see shed) from the fire's primary updraft. This region of low reflectivity is also apparent as the narrow "weakness" in the reflectivity plan-view map in Fig. 5a, which occurs in the region between the updraft and the ash fall downwind. The time mean radar reflectivity also indicates a counter-clockwise curving ashfall region (black dashed line, Fig. 5a), which is evidence of the backing wind profile (shown in Fig. 2a,c,d).
Photographs and videos help confirm these radar observations, showing that the earlier FGTVs (e.g., before 2130 UTC) were embedded in an anticyclonically rotating smoke and ash filled convective column linked directly to the fire (P1, Fig. 5e, also see movie S2).
In contrast, the later "shedding" FGTV, shown in Fig. 5f, was funnel-like, pendant from the plume aloft, and separated from the primary fire front, consistent with the 3D radar renderings (see also movie S3).
Taken together, the observations from the Loyalton Fire provide rare insight into the location and morphology of FGTVs and show distinct similarities to laboratory experiments with jets/plumes in crossflows in terms of vortex locations, flow features, and plume geometry (c.f., Fig. 1).

Large Embedded Vortices during the Creek Fire
The Creek Fire generated explosive pyroCb activity, with cloud tops reaching ~16 km MSL, and multiple strong FGTVs (30 m s -1 ) on 5 September 2020 under the influence of diurnally varying upslope and up-valley winds (Fig. 6c, Table 2). Like the Loyalton Fire, a pronounced backing wind profile impacted the plume (Fig 6a,c,d), who's growth is summarized in Fig. 7 (see animation S5). These data indicate progressive plume deepening (from 8 to ~16 km), periods with deep convective cores, and sustained pyroCb activity (as shown in Fig. 7c,d).
Plume tops easily surpassed the CCL at ~5.9 km and the homogenous freezing level at ~11 km. The cloud tops were close to the tropopause height, which was ~16,800 m, as determined from a sounding at Reno, NV. The pyroCb went on to produce lightning, precipitation, and downdrafts (a complete analysis of which are beyond the scope of this manuscript). These radar data also indicate a secondary pyroCb event in the evening (~0245 UTC on 6 Sept) wherein high reflectivity cores (~40 dbZ) reached ~12 km and plume tops 14 km.
The time series (Fig. 7b) and vortex depths (black squares, Fig 7a) show that the three deepest plume pulses were associated with FGTVs with exceeding 20 m s -1 (40 kts) at ~2050, 2200, and 0310 UTC (on 6 Sept). The peak twice reached 30 m s -1 (60 kts, see also S1), which is ~5 m s -1 (10 kts) stronger than in the Loyalton Fire despite the diminished beam-to-beam azimuthal resolution (1 km vs 480 m, see Table 1). The corresponding vortex depths (black squares in Fig anticyclonic circulation is much larger during the Creek Fire (~5 km diameter) than during the Loyalton Fire (~1-2 km diameters). These broader circulations suggest the potential for more significant wind impacts.
Apart from the FGTVs, the radar-observed airflow indicates prominent flow splitting around the fire flanks (red) and flow reversal zones (green) downwind of the head fire ( Fig During the second vortex period (2130-2158 UTC) the plume cores have moved laterally away from one another, and the left (anticyclonic) plume is more bent over, while the cyclonic updraft remains more upright and deeper (Fig. 8k). As before, the vortex cores remain embedded in the anticyclonic updraft. In contrast, for the tertiary, nocturnal FGTV (0240-0327 UTC) the cyclonic updraft is less established, and the deepest part of the plume is linked to the anticyclonic vortex region (Fig. 8l). One reason for this change may be decoupling of the near-surface winds after dark (note inbound flow adjacent to the fire in Fig.   8h).
In summary, the Creek Fire produced long-duration, high rotational velocity, Time-averaged radar maps, along with vortex snapshots, establish the dominant flow features during the Bear Fire ( Fig. 11a,b). Like the previous fires, these data indicate prominent flow reversal (red shading) extending >10 km downwind of the head fire, with strong convergence between the northeasterly winds (15-25 m s -1 ) and the reversed flow (10-15 m s -1 ; Fig. 11b). The northeasterly flow splits around the head fire, yielding cyclonic and anticyclonic shear zones along the northern and southern periphery of the flow reversal zone, respectively. The anticyclonic shear zone is the stronger of the two (i.e., a tighter gradient), and hosts the compact, but vigorous, anticyclonic FGTVs (Fig. 11c,d,e). The radar snapshots also show that the FGTVs emerge from near the head fire, then migrate downstream along the anticyclonic shear maxima (Fig. 11c,d, Fig. 5a,b). As with the Loyalton Fire these vortices may carry embers and flaming gases, leading to accelerated fire spread through this region.
The accompanying radar volume and vortex-line renderings show that the vortices diminish in depth as they move downstream and detach from the left-flank of the head fire (i.e., moving right to left in the image; Fig. 11g,h). The vortices also occur downwind from where the flanking plume merges with the head fire's updraft and lifts from its near-surface trajectory (annotation arrows in Fig. 11g,h), which is consistent with the location of wakelike vortices found in laboratory experiments (e.g., Fric and Roshko 1994). The accompanying webcam snapshot shows the approximate location of these FGTVs, though the vortices are cloaked in smoke and ash (Fig. 11i).
Both the volumetric and near-surface reflectivity data also indicate counter-clockwise curvature in the ash fall region extending away from the head fire ( Fig. 11a,g,h). As with the previous cases, this curvature is indicative of the backing winds, which turn from northeast near the surface to northerly aloft (as shown in Fig. 9a,c). This is also apparent in the photograph, which shows dense smoke and ash spreading southward above the vortex zone.
In summary, the Bear Fire provides an interesting case of strong, near-surface winds and strong, but transient, FGTVs that propagate away from the head fire along an anticyclonic shear zone. Thus, there are similarities to the subset of shedding vortices observed during the Loyalton Fire and to the broader disruption of the flow apparent in all three cases. These similarities set the stage for the following synthesis of these FGTV events.

Common Radar Signatures
Commonalities amongst the Loyalton, Creek, and Bear Fires provide the building blocks for a FGTV conceptual model. These common features, summarized schematically in Fig. 12 (1) Anticyclonic vortices (triangles) with rotational velocity exceeding 20 m s -1 (40 kts) on the left flank of the asymmetric head fire (black oval in upper panels) with two distinct morphologies: (a) Embedded FGTVs within the high-reflectivity updraft cores and anchored to the fire (red triangles).
(b) Shedding FGTVs moving away from the fire along the periphery of the reversed flow (magenta triangles) and pendant from the bent-over plume.
(2) Flow splitting (blue arrows) and flow reversal (red arrows) around the head fire indicative of CVPs (blue and red circles). The flow reversal can extend >10 km downwind from the fire.
(3) Counter-clockwise curving ashfall extending downwind from the head fire indicative of a backing wind profile (see inset wind barbs).
(4) Bent-over and bifurcating plume structures associated with the CVP (as shown in earlier volume renderings, e.g., Fig. 5d). with the "tornado-like" wake vortices described in Fric and Roshko (1994). Our embedded and shedding vortex morphologies are also broadly consistent with quasi-steady on-source and unsteady off-source whirls, respectively, discussed in Tohidi et al. 2018, wherein the source refers to the fuel bed.
While laboratory studies provide intriguing analogs to our FGTV cases it is important to acknowledge that these real-world scenarios include additional complexities. These include, but are not limited to, (1) the influence of stratification, apparent as the descending branch of the plumes, (2) the contribution of latent heating in the pyroCb to the plume structure and kinematics, (3) ambient turbulence in the convective boundary layer, (4) unsteadiness in the combustion, and (5) a host of terrain-flow effects, some of which are discussed below. Future work will need to isolate the importance of these processes.
An additional complexity in our cases is the tendency for fire-flow interactions to favor FGTVs on one flank of the fire, in this case, the anticyclonic flank. This may provide important context for identifying when and where a fire will yield an FGTV. We note that the angled head fire structures in our cases are similar to that of oval jets inclined to the crossflow, which produce asymmetric vortex structures in laboratory experiments (Wu et al. 1988). Fire-geometry and crossflow interactions have also been linked to vortex generation in other laboratory and wildfire studies (e.g., Kuwana et al. 2013;Peace et al. 2015). It is also possible that backing wind profiles favor anticyclonic vortices via linear dynamic pressure perturbations akin to those in mesocyclonic thunderstorms forming in sheared environments (Markowski and Richardson 2011). Indeed, simulations of buoyant plumes from hydrothermal vents in sheared flows (i.e., Eckman layer) also generate asymmetric CVPs (Lavelle 1997 To this end, observations from other fires suggest a possible sensitivity to the wind profile. For example, Fig. 13 shows radar observations of two other pyroCb plumes (King and Apple fires, see Table 1) that produced CVPs with flow splitting and flow reversal (arrow annotations in Fig. 13), but did not produce FGTVs. Notably, these cases have only speed shear, evident in the ash fall extending in a straight, rather than curved, trajectory from the head fire (black dashed line). They also have weaker flow reversal, which may be indicative of plumes less conducive to FGTV development due to less disruption of the crossflow. This may be analogous to identifying difference between non-tornadic and tornadic supercells where environmental factors (e.g., sheer, moisture, etc.) modulate the potential for tornadoes or in our cases, FGTVs. Future idealized modeling studies should be conducted to explore these shear-plume interactions and sensitivities, which may yield a better understanding of what tips the balance between the common CVP signature and rare FGTV formation.

FGTVs in context
It is important to place FGTV strength ( ), depth, and damage in the context of ordinary tornadoes (Fig. 14). This is accomplished using a database of tornado , debris signature (TDS) heights, and "enhanced Fujita-scale" (EF) damage ratings (https://www.spc.noaa.gov/efscale/ef-scale.html; Emmerson et a. 2019;. For the FGTVs we use the estimated vortex top rather than TDS (Supplemental Information S1), which is not defined for FGTVs, and limit the analysis to the strongest and deepest FGTVs. 98, 60, 23% probabilities of exceeding EF1, 2, and 3 damage, respectively (see Fig. 7 in Smith et al. 2020). We note that the Carr Fire FGTV, documented in Lareau et al. (2018), has not been included in Fig. 14  The radar estimated and observed impacts of FGTVs underscore their threat and the need to warn for their development, as was done with the first-ever NWS tornado warning for the Loyalton Fire's FGTVs. Future dialogue amongst wildfire stakeholders and weather forecasters will be needed to establish and refine warning criteria for these events.

Site Specific Factors Influencing FGTVs
Site-specific factors, including terrain, fuels, and micro-to meso-scale flows can impact arrangement of fuel loads (Zhou and Wu 2007). To examine these factors, Fig. 15 shows the terrain (hill shaded) and satellite imagery, representing the pre-fire fuel distributions, for each fire. The Loyalton Fire FGTVs occurred over a 10 km span on lee slopes (in southwest winds) and moved from heavier fuels at upper elevations to lighter, flashier fuels (Table 2) at lower elevations. The Creek Fire FGTVs occurred along a >10 km span along the west edges of the deeply incised San Joaquin River valley, and then into higher elevation terrain. The fuels ranged from brush and grasses to heavy timber ( Table 2). The Bear Fire's FGTVs occurred along a plateau, moving through a patchwork of previously logged plots. While informative, these limited observations are insufficient to establish the importance of terrain and fuels on FGTV development. That said, we believe the commonalities in plume and vortex structures amongst our cases suggest that terrain and fuels are not the dominant factor in these FGTVs. For example, the Creek Fire generated FGTVs over a span of 9 hours as the fire progressed ~20 km, moving through varying terrain and fuel loads. Clearly then, no one specific terrain feature or fuel configuration could explain the persistent FGTVs, which remained in a fixed location relative to the fire and plume.

Summary
We have presented three cases of large, high-impact wildfires in California that produced fire-generated tornado-strength vortices (FGTVs) and pyroCb. Using radar and satellite observations we documented FGTV strengths, depths, and locations and placed those data in the broader context of the wildfire plume structure and fire evolution. The observations indicate long-lived anticyclonic vortices with rotational velocity up to 30 m s -1 (60 kts), vortex depths as great as 4.9 km AGL, and plume tops as high as 16 km MSL.
From these observations we have identified two distinct FGTV morphologies: (1) Embedded vortices residing within one branch of the counter rotating vortex pair and anchored to the fire, and (2) shedding vortices, which detach from the fire and progress downstream while pendant from the bent-over plume. In addition, we have documented common flow and plume features linked to the FGTVs, which include prominent meso-scale flow reversal downstream of the head fire, flow splitting around the fire's updraft, and bentover plume structures due to the interaction of the plumes with the cross wind. We have also shown that the vortex cores, in two cases, reach pyroCb cloud base and that vortex strength covaries with pyroCb plume depth, suggesting two-way links between the cloud processes aloft and the vortex processes at the surface.
The inferences from this study compliment the understanding gained from previous           (c,f,i) Fire-RGB satellite imagery showing the fire location and relative intensity along with estimated fire perimeters and FGTV locations. (j,k,l) Radar reflectivity iso-surfaces of the time-averaged plume structure looking from the southwest. The solid black lines and filled circles indicate vortex lines, with the marker size scaled to the rotational velocity. Figure 9. Overview of the meteorology during the Bear Fire on 9/9/2020. (a) HRRR model sounding, (b) 500 hPa heights (in meters) and 700-400 hPa layer averaged relative humidity (shading), (c) wind barbs for the surface (blue) and 700 hPa (red) along with the fire perimeters (black line), approximate vortex zone (red shaded), and topography (terrain shaded), and (d) time series of wind speed and direction from a location just north of the Fire.