In this study, data from a Halo Photonics scanning Doppler lidar are examined (Pearson et al., 2009). The lidar emits a 1.5 µm laser beam and records two range-resolved quantities: (1) the attenuated backscatter coefficient (m-1 sr-1), which is a range corrected measure of backscattered energy, and (2) the Doppler velocity (m s-1). The lidar also reports the signal-to-noise ratio (SNR), which is useful for discriminating between meteorological targets and instrument noise. The lidar range is 9.6 km and the range-gate resolution is 18 m. Azimuth and elevation motors allow for scans of the full upper hemisphere and the lidar is level-mounted to the bed of a pickup truck, enabling rapid deployments to wildfires (Clements and Oliphant, 2014).

Data from four National Weather Service (NWS) radars are used to examine plume structure. These 10 cm radars are sensitive to large ash and precipitation particles in the convective column but insensitive to cloud droplets and sub-micron smoke. Radars have been used in numerous wildfires studies (Banta et al., 1992; Hufford et al., 1998; Fromm et al., 2006; Rosenfeld et al., 2007; Jones and Christopher, 2010a, b). Recently, dual-polarization radars have been used to examine the microphysics of wildfire plumes and clouds (Melnikov et al., 2008, 2009; Lang et al., 2014).

In this study we leverage three aspects of the NWS radars. First we examine the radar echo tops to estimate the maximum cloud height. The echo tops are the highest level at which the radar reflectivity exceeds 18 dbZ (Lakshmanan et al., 2013). Second we combine radar reflectivity from multiple radars to generate volume renderings of the pyroconvective plumes. These volumes are constructed by creating a gridded interpolant from all the available contemporaneous radar data. Data from the Medford, Reno, Beale, and Sacramento radars are combined for the Bald Fire, and from the Beale and Sacramento radars for the Rocky Fire. The radar locations relative to the fires are shown in Fig. 1.

Finally, we inspect the differential reflectivity (Zdr) data from the Medford, OR radar during the Bald Fire. Zdr is the logarithmic ratio of the reflectivity from the horizontally and vertically polarized radar beams (Markowski and Richardson, 2011). When Zdr is large and positive it indicates the presence of large horizontal targets, including needle-like ash particles (Melnikov et al., 2008, 2009). When Zdr is near zero the targets are more spherical (e.g., hydrometeors), and when negative the targets are vertically oriented (e.g., graupel).

Visible satellite observations from the Geostationary Operational Environmental Satellite (GOES-15) are used to characterize the presence of pyroCus above each fire. These data have a spatial resolution of 1 km and a nominal temporal resolution of 15 min, depending on the scan schedule. Data from the Moderate Resolution Imaging Spectroradiometer (MODIS) Terra and Aqua satellites are also used. These data include both true color visible images and fire-radiative power (FRP). The nominal resolution is 500 m. FRP is derived by differencing fire pixels from adjacent non-fire pixels using infrared radiance (Wooster, 2002) and has been shown to provide high fidelity representation of fire activity during California wildfires (Koltunov et al., 2012; Peterson et al., 2015). FRP data from GOES are also examined.

Thermodynamic profiles were collected adjacent to both fires using a GRAWTM GS-E radiosonde system. These sondes measure temperature, humidity, and wind from the surface to the tropopause, ascending at a rate of ∼ 3 m s-1. The balloons were launched after sunset to avoid interfering with daytime fire-suppression aircraft operations, and as a result the temperature profiles include surface-based stable layers that are not representative of daytime conditions. To address this shortcoming, the afternoon temperature from the truck weather station is used to infer the convective boundary layer (CBL) depth using the “parcel method” (Holtzworth, 1964).

The sonde data are used to examine the ambient condensation level by considering three lifted parcels (1) the most unstable (MU) parcel, (2) the mixed-layer (ML) parcel, and (3) the convective (CONV) parcel. The MU parcel is the parcel with the highest convective available potential energy (CAPE), whereas the ML parcel is based on the mean temperature and mixing ratio in the lowest 150 hPa. The CONV parcel reflects the surface temperature required for free convection based on the surface mixing ratio. The condensation level for each of these parcels is compared in the analyses below.

The truck-mounted Doppler lidar was situated ∼ 7 km southwest of the fire where it conducted RHI scans of the windward edge of the developing pyroCu from 13:50 to 15:32 PDT on 2 August. During this time, 95 RHI scans were completed, providing a time- and space-resolved measure of the plume evolution, including the height of the condensation level. The lidar location relative to the fire perimeter is indicated in Figs. 3 and 4e. Figure 4f provides photograph from the lidar vantage point showing the windward edge of the cloud capped smoke column.

Figure 5 shows a sequence of lidar scans spanning the 5 min period prior to the MODIS-Aqua overpass. These data are expressed as the logarithmic attenuated backscatter coefficient (hereafter backscatter) in units of m-1 sr-1. Red and blue dots represent the leading plume edge and attenuation point, respectively, along each individual beam. The backscatter is due to smoke and debris in the lower portion of the plume (below 5500 m) and due to cloud droplets in the pyroCu aloft. The laser beam attenuates in both the smoke and the cloud water, but the cloud edge is marked by a distinct increase in the backscatter and more rapid attenuation. For example the distance between the leading plume edge (red dots) and the attenuation point (blue dots) tends to be much larger in the lower portion of the plume, whereas above 5500 m the attenuation occurs over just a few range gates. These aspects of the data give the pyroCu cloud returns a “crisper” edge.

While pyroCus were already present at the beginning of the scan sequence, the data show the development of a new cloud element. Figure 5a and b for example, show only a few points of rapid attenuation and high backscatter aloft, whereas starting at Fig. 5c a new, upright cloud edge is detected. This nascent pyroCu element then rapidly expands during the subsequent RHI scans, reaching a height of at least 8500 m before moving out of the lidar field of view (Fig. 5d–f). As we show in the radar analysis below, the actual plume top was as high as 12 km in the 10 min following these scans.

The scans, which were roughly parallel to the mean wind direction, also reveal that the plume experienced significant variations in tilt with time, alternating between windward (Fig. 5a) and rearward sloping geometries (Fig. 5f). In fact, the windward protrusion of the plume was as much as 2 km away from its base. Large coherent vortices are also apparent along the plume edge, especially in Fig. 5a and b as the “stair step” pattern in the plume edge detections. Based on the radial velocity data (not shown) the inward clefts in the plume edge correspond to enhanced flow into the plume and outward lobes reflecting flow towards the lidar. Vortices of this sort are a well-known feature of rising thermals and plumes and play a leading role in entrainment (Scorer, 1957; Woodward, 1959).

Following the initial plume rise, sustained pyroCus were observed with the lidar until 15:32 PDT, at which point the truck was relocated for safety reasons. To determine the plume condensation level, we aggregate data from all of the lidar scans during this period. From this larger data sample, Fig. 6a presents the time-maximum backscatter as a function of height and distance, and Fig. 6b as a function of height only. In addition, Fig. 6c shows the computed percentiles (5, 50, and 95th percentiles) of the attenuation depth binned into 100 m intervals. Collectively, these data reinforce many of the aspects of the initial plume rise sequence discussed above. For example, there is a persistent transition in backscatter near 5500 m (blue line in Fig. 6a and b). Below this level, the backscatter approximately linearly decreases with height, consistent with the entrainment of clear air into an aerosol-laden plume. In contrast, at 5500 m the backscatter sharply increases (as does the SNR, not shown), corresponding to the condensation level and development of the pyroCu.  The backscatter intensity remains high there and above due to the continued presence of liquid water.

The attenuation depth also shows a sharp transition at 5500 m (Fig. 6c). Below that level the median attenuation depth increases with height, which is again consistent with the dilution of the smoke plume via entrainment. At 5500 m the attenuation depth (across all percentiles) sharply decreases, converging towards a median value of ∼ 200 m. The rapid attenuation aloft is consistent with the presence of liquid water drops and supports our interpretation that the change in backscatter intensity is due to condensation in the plume. From these data, we therefore conclude that the observed condensation level occurs very near 5500 m and was nearly constant throughout the 1.5 h observation period despite many changes in fire intensity.

Since the pyroCu cloud tops exceeded the lidar range, the maximum smoke injection depth is obtained from the radar echo tops product from the NWS radar in Medford, OR (KMAX), which is ∼ 200 km to the northwest. Figure 7a shows the time maximum of the echo tops above the Bald Fire on 2 August. These data indicate that smoke reached altitudes in excess of 12 km, and thus the convective column rose an additional 3.5 km above the maximum height resolved in the lidar scans. The radar returns also show that the highest echo tops occur in a localized region above the fire perimeter where pyroCus were most prevalent. In contrast, the smoke layers without pyroCus correspond to plume heights closer to 6 km.

An additional interesting aspect of the radar data is the presence of deep echo tops southwest (e.g., upwind) of the infrared fire perimeter (solid contours, Fig. 7a). This observation is consistent with the periodic forward tilt of the plume as observed in the lidar backscatter (Figs. 5 and 6). We hypothesize that the forward tilt relates to large-scale vortices that form as the plume penetrates through a stable layer at the top of the boundary layer (Saunders, 1962), and due to the deflection of the ambient flow around the plume.

Figure 7b shows the corresponding time series of the maximum radar echo tops. The pyroCu initiation at 13:30 PDT, as shown in Fig. 4b, corresponds to a rapid rise in echo tops from 6500 to 10 000 m. Following the initial plume growth, the plume tops slowly rise until 14:20 PDT at which point a second period of rapid plume growth occurs, pushing the convective column to heights above 12 km. The onset of this deep plume penetration closely corresponds to the lidar plume rise sequence shown in Fig. 6, as well as the MODIS-aqua image. The plume heights subsequently subside, remaining near 10 km for the balance of the afternoon before diminishing more substantively at night.

A 3-D volume rendering of radar reflectivity from the Bald Fire at the time of maximum injection height (14:29 PDT) reveals additional aspects of the plume structure (Fig. 7c). The isosurfaces for 30, 28, 26, 24, and 18 dbZ are shown, along with the fire perimeters (red shading), lidar scan plane (black dots), lidar plume edge detections (yellow dots), and the lidar-derived condensation level (green contour). These volume data show an expansive region of high reflectivity immediately above the fire perimeter. The reflectivity and plume height diminish towards the northeast, consistent with the fall out of the larger soot and ash particles in the downwind direction (e.g., southwest flow aloft). We note that since the radar is not sensitive to cloud droplets or micron-sized smoke, it is possible that the cloud edges and some smoke reside outside of the radar volume rendering. It is also clear from these data that the lidar sees only the leading edge of the plume before attenuating in dense smoke and cloud water, consistent with the analyses presented above.

The shape of particles within the plume can be inferred by considering the differential reflectivity (Zdr) from the Medford, OR radar at different heights (Fig. 8). Three elevation angles are inspected (0.5, 1.5, and 2.4∘), intersecting the updraft core at heights of 4115, 7742, and 11 009 m, respectively. The lowest scan shows very high Zdr, indicative of large, horizontally oriented particles, which is consistent with ash (Melnikov et al., 2008, 2009; Lang et al., 2014) (Fig. 8a). In contrast, the mid-elevation scan intersects the plume above the condensation level and shows a significant reduction in Zdr, with values between 0 and 2.5 in the updraft core (Fig. 8b). These values correspond to more spherical particles and small ice, suggesting the presence of large hydrometeors. Finally the upper-most portion of the plume, at ∼ 11 km, exhibits negative Zdr, presenting the possibility of vertically oriented graupel particles (Fig. 8c).

The lidar-observed condensation level and radar-estimated plume tops provide valuable constraints on the plume structure when contextualized with atmospheric profiles collected adjacent to the fire. Figure 9a, for example, shows data from a radiosonde launched at 21:00 PDT from ∼ 15 km to the southwest of the fire (location shown in Fig. 3).

The sounding shows that the afternoon CBL extends from the surface (1364 m) to ∼ 4000 m and is capped by a pronounced stable layer. Within the CBL, the water vapor mixing ratio is roughly constant at ∼ 5 g kg-1, whereas above the CBL, a layer of very dry air is observed with a mixing ratio of only ∼ 0.5 g kg-1. Further aloft, near 400 hPa, a layer of higher humidity air, reflecting monsoonal moisture, is found. The height of the tropopause is ∼ 13 km.

Relative to the observed profile, the “in cloud” profile is estimated by pseudo-adiabatically lifting a parcel from the lidar-observed condensation level at 5500 m. The resulting parcel possesses 910 J kg-1 of CAPE, which is an upper bound on the energy available for buoyant ascent. The equilibrium level (EL) of the pyroCu parcel is 11 742 m, which is in close agreement with the radar-estimated echo tops, but does not account for the inertial overshoot of the parcel, which is likely reflected in the localized region of radar plume heights exceeding 12 km (Fig. 7a).

Also of note, the homogeneous freezing level (-38 ∘C) in the plume profile occurs at 10,158 m and the temperature at the EL is -52 ∘C, indicating that the upper portion of the cloud must be glaciated. As such, this particular pyroconvective cloud should be classified as a pyroCb. In fact, pyroCbs from other nearby fires on that day were known to produce lightning as well as a significant and destructive fire-whirl (Muller and Herbster, 2014).

Lidar RHI scans where conducted between 15:45 and 20:08 PDT from an already burned area within the Rocky Fire perimeter (Fig. 10). This location allowed for scans of four separate pyroCu plumes rising from the complex fire perimeter. A total of 267 RHI scans were performed.

PyroCu were first observed with the lidar starting at ∼ 16:00 PDT rising from the northwest flank of the fire (319∘ azimuth, scan path #1 in Fig. 10). Figure 12 shows a sequence of photographs (top panels) and contemporaneous lidar scans (bottom panels) detailing the onset and expansion of this cloud-topped plume. The plume was initially observed as it penetrated through a stable layer at the top of the CBL, evident as a lateral smoke layer at 2600 m in the backscatter data and as a diffuse haze in the photographs. During this time, a thin pileus cloud accompanied the developing pyroCu and the lidar cloud returns were limited to a few points near the plume top (Fig. 12f).

By 16:03 PDT, however, a distinct cumuliform cloud had developed (Fig. 12b) and the lidar backscatter showed a commensurate increase in intensity and attenuation along the pyroCu edge (Fig. 12g). Based on these data the cloud base was at ∼ 4200 m. The subsequent scans show the rapid pyroCu development, and by 16:09 PDT cloud edges were detected as high as 7500 m. Interestingly, soon thereafter the pyroCu detrained from the convective column and dissipated (not shown).

Another pyroCu event at 18:00 PDT is detailed in Fig. 13 corresponding to a lidar azimuth of 86∘ (scan path #3 in Fig. 11). This plume initiated ∼ 2.5 km east of the lidar location. As in the previous case, the rapidly growing plume was first recorded as it rose through the boundary layer top, now at ∼ 2300 m, and expanded into the free troposphere (Fig. 13a and e). Subsequent scans revealed the onset of pyroCu with a condensation level of 4200 m, which is unchanged from the earlier Rocky Fire pyroCu event detailed in Fig. 12. In this case, however, the cloud top was not as well documented because attenuation at the cloud base shielded the lidar view of the upper plume.

The photographs detailing the plume rise show changes in smoke color near the base of the convective column (Fig. 13a–d). For example, at 18:05 PDT the smoke is a dark gray (Fig. 13a), whereas later the smoke is increasingly white (Fig. 13d). We believe the change in smoke coloration is associated with changes in the completeness of combustion: flaming combustion produces smoke dominated by black carbon aerosols, whereas smoldering combustion generates more organic carbon aerosol, which more effectively backscatter sunlight and appear whiter (Bellouin, 2014; Saleh et al., 2014).

An additional aspect of the observed plume rise is the relationship between the updraft strength and the ambient wind. This relationship is examined in Fig. 14, which displays VAD wind profiles (Fig. 14a and b) and RHI radial velocities detailing the plume structure (Fig. 14c–e). The wind profiles show significant shear over the lowest 2 km of the atmosphere. Strong (5–7 m s-1) northwesterly winds near the surface transition to weak flow at the boundary layer top (0–1 m s-1 near 2300 m), then reverse to easterly flow aloft (Fig. 14a and b). The observed near-surface wind speed maximum is atypical in the atmospheric boundary layer and “adverse” wind profiles of this character have previously been linked to blow-up fires (Byram, 1954).

Compared to the ambient wind, the flow within the plume is characterized by much stronger velocities (Fig. 14c and d). For example, outbound speeds in excess of 15 m s-1 are recorded at numerous locations within the plume at 18:09 PDT. The actual updraft speed likely exceeds these values since the radial velocity data only reflect the projection of the updraft onto the oblique lidar beam. Significant downdrafts are also observed in the upper portion of the plume, especially at 18:13 PDT (blue shading, Fig. 14c and d).

The RHI velocity data also show that the strength of the updraft diminished with time. For example, comparative histograms demonstrate that strong outbound velocities were both higher and more common at 18:09 than 18:13 PDT (Fig. 14e). This observation is consistent with the change in smoke coloration described above: flaming combustion is likely to produce stronger updrafts due to more rapid heat release.

Changes in plume geometry also accompany the reduction in updraft strength. The plume is at first mostly upright (Fig. 14c) and later becomes more sheared (Fig. 14d). Specifically, the leading plume edge becomes tilted downwind within the boundary layer, while the upper portion of the plume bends back towards the observing location. Based on these data, we hypothesize that as the fire's updraft weakens, it becomes more susceptible to the ambient shear. The role of wind shear as a limiting factor in plume development is further discussed below.

A more robust examination of the plume condensation level during the Rocky Fire's multiple pyroCu events is presented in Fig. 15. These analyses leverage the aggregated data from all of the RHI scans on 30 July. Unsurprisingly, the time-maximum backscatter exhibit a sharp transition near 4200 m (Fig. 15a and b), as was indicated in the earlier plume rise sequences (Figs. 12 and 13). Below 4200 m, the backscatter decays roughly linearly with height, and above that level the backscatter converges to a value of near -4 m-1 sr-1 (Fig. 15b). Likewise, the attenuation depth linearly increases from the surface up to 4200 m, then abruptly decreases to a median depth of ∼ 200 m. This pattern is consistent with the dilution of the smoke plume by entrainment and the onset of condensation aloft. Importantly, these analyses are remarkably similar to those during the Bald Fire suggesting a clear lidar signature of pyroCu onset. Moreover, the condensation level is once again found to be constant throughout the observing period, indicating that ambient atmospheric conditions rather than variations in water released during combustion likely control its height.

Radar data are again used to estimate the maximum smoke injection depth. The Rocky Fire was within ∼ 100 km of both the Sacramento (KDAX) and Beale Air Force Base (KBBX) NWS radars (Fig. 1), and data from both sites are in agreement.

The maximum echo tops (from KDAX) occur between 7000 and 7500 m, consistent with the lidar cloud detections (Fig. 16a). The spatial pattern of echo tops indicate that plumes of similar height developed on all of the expanding flanks of the fire. Interestingly, the corresponding time series demonstrates the plume transience, showing rapid variations in plume height throughout the late afternoon (Fig. 16b). Each spike corresponds to a short-lived pyroCu with durations ranging from 10–30 min. Satellite data confirm the episodic nature of these plumes (not shown).

The variability in echo tops is also due to the presence of multiple updrafts. For example, a volume rendering of the reflectivity data at 16:09 PDT shows the two distinct updrafts associated with the complex fire perimeter (Fig. 16c). The narrow updraft rising from the northwestern flank of the fire is the same plume shown in Fig. 12, and the lidar plume detections agree well with the radar data (yellow dots, Fig. 16c). A second broader plume rises from the north and northeastern flanks of the fire at the same time. Above 5000 m the upper portions of both plumes are tilted to the north–northwest due to southeasterly flow in that layer. Later in the fire's evolution the plume growth shifted towards the east and southeast (not shown).

The Rocky Fire pyroCu development is interesting in that the thermodynamic environment theoretically supports much deeper convection than was observed. Using radiosonde data from ∼ 15 km southwest of the fire at 21:05 PDT, Fig. 17a shows that moist adiabatic ascent from the observed 4200 m cloud base would generate 2035 J kg-1 of CAPE and that the plume equilibrium level would be ∼ 13 km, impinging on the tropopause. The radar and lidar data indicate, however, that the plumes ascended to no higher than ∼ 7.5 km, corresponding to plume top temperature of -20 ∘C. As such, these clouds are best classified as pyroCu, and never developed as deep pyroCb.

What then limits the growth? There appear to be two related limiting factors in the plume rise: (1) wind shear, and (2) dry air entrainment. The lidar wind profiles, presented above in Fig. 14, indicate significant wind shear between the CBL and free troposphere. This wind shear is also apparent in the radiosonde wind profile, which shows a 180∘ wind shift at 2300 m (wind barbs and hodograph Fig. 17a). The flow below this level is from the west–northwest, whereas the flow above, and extending up to ∼ 7 km, is from the east–southeast. The layer of southeasterly flow is associated with a surge of monsoonal moisture.

A second layer of significant wind shear at 7000 m separates the monsoon flow from southwesterly flow in the upper troposphere. This shear also coincides with a rapid decrease in dew point temperature, and thus relative humidity. It is notable then that the maximum echo tops occur only about 500 m above the upper shear layer. Visual observations throughout the afternoon and early evening suggest this shear zone affected the pyroCu development, tending to sweep the upper portion of the cloud away from the updraft core. The detraining upper portions of the cloud subsequently developed ragged and wispy edges indicative of dry air entrainment as opposed to the crisp crenelations of growing cumulus congestus.

The effect of the wind shear on a buoyant parcel is easily visualized by examining the ascent track of the radiosonde, which rose at a mean rate of 2.7 m s-1 (Fig. 17b). The ambient shear causes a pronounced zigzag pattern that is clearly detrimental to sustained upright convection despite the substantive CAPE. This result is not surprising in that CAPE is known to overestimate convective development and updraft strength (Markowski and Richardson, 2011).

Figure 19a shows the GOES-15 and MODIS FRP in for the Rocky Fire on 29–30 July 2015. From these data it is clear that the diurnal cycle of fire intensity is similar during the first two days of fire growth, with peak FRP values near 1500 MW in the late afternoon and fire activity extending into the late evening. Interestingly, despite comparable fire intensity, pyroCus were not observed on 29 July but were widespread on 30 July.

To better understand this disparity Fig. 19b–e compares the ambient meteorological conditions between days. These data are from a weather station just outside the fire perimeter, the location of which is shown in Fig. 10 (data obtained from MesoWest, Horel et al., 2002), and the 17:00 PDT upper air soundings at Oakland International Airport (KOAK, location shown in Fig. 1). The high temperature on both days was ∼ 39 ∘C and afternoon winds were from the west with peak gusts near 6 m s-1. The relative humidity was higher on 30 July than 29 July.

Figure 19d and e show the time series of the water vapor mixing ratio and the differences in the relative humidity from the KOAK soundings for two afternoons. From these data it is apparent that the onset of pyroCus on 30 July corresponds to the arrival of much higher humidity air, both at the surface and aloft. For example, the mixing ratio increases from 4.5 to 8 g kg-1 while the relative humidity at 5500 m jumps from 7 to 66 %. The corresponding change in the CCL is substantial, dropping from 5848 m on 29 July to 4267 m on 30 July. Since the fire intensity was similar on both afternoons it is likely that reduction in the height of CCL due to the influx of monsoon moisture was the driving factor in pyroCu formation. These observations support the conclusions of Luderer et al. (2006, 2009) that environmental moisture, not water released in combustion, is the primary control on pyroCu development.