Abstract

Intense wildfires occasionally generate fire-triggered storms, known as pyrocumulonimbus (pyroCb), that can inject smoke particles and trace gases into the upper troposphere and lower stratosphere (UTLS). This study develops the first pyroCb detection algorithm using three infrared (IR) channels from the imager on board GOES-West (GOES-15). The algorithm first identifies deep convection near active fires via the longwave IR brightness temperature, distinguishing between midtropospheric and UTLS injections. During daytime, unique pyroCb microphysical properties are characterized by a medium-wave brightness temperature that is significantly larger than that in the longwave, allowing for separation of pyroCb from other deep convection. A cloud-opacity test reduces potential false detections. Application of this algorithm to 88 intense wildfires observed during the 2013 fire season in western North America resulted in successful detection of individual intense events, pyroCb embedded within traditional convection, and multiple, short-lived pulses of pyroconvective activity. Comparisons with a community inventory indicate that this algorithm captures the majority of pyroCb. The primary limitation is that pyroCb anvils can be small relative to GOES-West pixel size, especially in regions with large viewing angles. The algorithm is also sensitive to some false positives from traditional convection that either ingests smoke or exhibits extreme updraft velocities. A total of 26 pyroCb events are inventoried, including 31 individual pulses, all of which can inject smoke into the UTLS. Six of the inventoried intense pyroCb were not previously documented. Near-real-time application of this algorithm can be extended to other regions and to next-generation geostationary sensors, which offer significant advantages for pyroCb and fire detection.

1. Introduction

Smoke plumes observed during intense fire activity occasionally become capped by cumulus clouds [pyrocumulus (pyroCu)]. Depending on meteorological conditions, pyroCu can continue developing into a larger fire-triggered thunderstorm, known as pyrocumulonimbus (pyroCb; http://glossary.ametsoc.org/wiki/Pyrocumulonimbus). This extreme form of deep “pyroconvection” is characterized by its distinctive cloud microphysics when compared with traditional convection. Smoke plumes carry large concentrations of cloud condensation nuclei, which yield a large quantity of small ice particles (e.g., Rosenfeld et al. 2007; Reutter et al. 2014). Precipitation development is therefore slow or nonexistent, allowing a large quantity of smoke particles and small ice crystals to reach high altitudes without being scavenged. This increases the lifetime of pyroCb anvil clouds (Lindsey and Fromm 2008), drastically enhancing downwind smoke transport.

PyroCb are known to inject a significant quantity of aerosol mass into the upper troposphere and lower stratosphere (UTLS)—sometimes more than 7–10 km above the tropopause (Fromm et al. 2005, 2008a,b). Several stratospheric aerosol layers that were previously presumed to be of volcanic origin have recently been reclassified as originating from pyroCb activity (Fromm et al. 2010). A volcanic eruption can inject a significantly larger quantity of stratospheric aerosol mass than can a single pyroCb event, but pyroCb activity is observed with a much higher frequency (e.g., Fromm et al. 2010), originating almost exclusively in the mid- and upper-latitude forests of North America, Asia, and Australia. PyroCb are therefore highly relevant for understanding stratospheric aerosol mass loading and its effects on global climate.

Stratospheric smoke layers and their connection to pyroCb activity were first identified in the early 2000s (Fromm et al. 2000; Fromm and Servranckx 2003). Many studies have since employed a wide variety of ad hoc remote sensing and modeling techniques to identify and characterize pyroCb events. Ground-based lidar and radar have been employed to measure pyroCb height and smoke-plume evolution downwind (Rosenfeld et al. 2007; Fromm et al. 2006, 2008b, 2010; Lareau and Clements 2016). Reverse trajectories from high-altitude smoke clouds observed by lidar and other sensors are often computed using atmospheric transport and dispersion models, such as the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (http://www.arl.noaa.gov/HYSPLIT_info.php). When these trajectories intersect locations of active fires, pyroCb occurrence can be confirmed during preceding days (Fromm et al. 2008b, 2010; Guan et al. 2010). Many intense-pyroCb events develop in remote regions where ground-based observations are unavailable. Spaceborne observations are therefore paramount for pyroCb detection and monitoring.

As viewed from space, intense-pyroCb activity is typically displayed as a cluster of fire pixels adjacent to a cumulus cloud, which is anchored to the fire and oriented downwind. The typical pyroCb life cycle involves initiation in midafternoon and termination a few hours after sunset (Fromm et al. 2010; Peterson et al. 2017). Many pyroCb are relatively short events (e.g., maturity period < ~1 h) and consist of a single anvil cloud. Sometimes, pyroCb will evolve into a multihour, multipulse event, producing two or more anvils (Rosenfeld et al. 2007). Cessation of pyroconvection is made apparent by the separation of the cold-cloud anvil from the fire pixel cluster and movement downwind.

Observations from polar-orbiting sensors, such as the Moderate Resolution Imaging Spectroradiometer (MODIS), Advanced Very High Resolution Radiometer (AVHRR), and Ozone Mapping Profiler Suite Aerosol Index (OMPS AI), can provide detailed information on pyroCb size, cloud-top brightness temperature (BT), effective radius, and the characteristics of the aerosol layer (e.g., Fromm et al. 2005, 2008a, 2012; Rosenfeld et al. 2007). Sensors with multiangle and profiling capabilities, such as the Multiangle Imaging Spectroradiometer (MISR) and Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP), provide information on the vertical structure of pyroCb and associated high-altitude smoke particles (Fromm et al. 2008a, 2012; Paugam et al. 2016). Geostationary sensors, such as the Geostationary Operational Environmental Satellites (GOES), provide a higher temporal sampling of pyroCb evolution and cloud-top BT (Fromm et al. 2005) but at the expense of a coarser spatial resolution. Polar-orbiting and geostationary sensors with multispectral capability have also been employed to separate the unique microphysical signature of a pyroCb from traditional convection (Fromm et al. 2005; Rosenfeld et al. 2007; Lindsey and Fromm 2008).

The combined information from the available suite of sensors, models, and remote sensing techniques has greatly advanced understanding of pyroCb in recent years, and it has even motivated a collaborative effort to begin archiving observed events. Much of the available literature stems from individual case studies of pyroCb, however, with several key studies focused on a single large pyroCb in Canada (i.e., the 2001 “Chisholm Fire”; Fromm and Servranckx 2003; Trentmann et al. 2006; Rosenfeld et al. 2007; Fromm et al. 2008a,b; Luderer et al. 2009). PyroCb have not been investigated over large spatiotemporal scales, and a unified, remote sensing–based pyroCb detection strategy is currently unavailable. It is therefore reasonable to assume that impacts from global pyroCb activity, including stratospheric intrusions of smoke particles, are greatly undersampled.

The goal of this study is to improve systematic detection and characterization of pyroCb through the development of an automated algorithm in western North America. Geostationary satellite observations are required because of the relatively short and sporadic life cycle of pyroCb. The algorithm takes advantage of the unique cloud microphysics of pyroCb anvils, which are characterized by extremely small ice particles relative to unperturbed traditional convection (Rosenfeld et al. 2007; Reutter et al. 2014; Chang et al. 2015). PyroCb microphysics increase solar reflectivity in the anvil region, thereby overwhelming the thermal radiative component of 4-μm BT during most of the daytime (e.g., Ellrod and Bailey 2007). The result is an unusually large 4-μm BT in the anvil region (Fromm et al. 2010). This attribute forms the foundation for separation of pyroCb from traditional convection, but only for daytime scenes. Additional tests for convective intensity and cloud opacity are also employed. Algorithm performance is assessed using documented pyroCb events during the fire season (June–August) of 2013.

Results highlight the ability of the algorithm to capture individual intense-pyroCb events, pyroCb embedded within traditional convection, and multiple, short-lived pulses of pyroCb activity. Sensitivity to other aerosol sources and traditional convection is also investigated. The pyroCb algorithm, in combination with satellite fire detections, is used to build the first systematic inventory of pyroCb events in western North America. A companion study combines this output with meteorological data to build the first physical conceptual model for pyroCb development (Peterson et al. 2017). These studies serve as a first step toward improved global detection, monitoring, and forecasting of pyroCb, ultimately providing increased understanding of high-altitude smoke plumes, their frequency of occurrence, and their potential impacts on global climate and the UTLS aerosol system.

2. Study region and data

The northern and eastern boundaries of the study domain are determined by the field of view for the operational Geostationary Operational Environmental Satellite observing western North America (GOES-West, currently GOES-15). The study domain includes all fire-prone regions in the western continental United States and northern Mexico (MCONUS), as well as the remote boreal forest of western Canada (Fig. 1). The western boundary is the border between Alaska and the Canadian Yukon territory (141°W). This region features a large latitudinal extent and complex topography. Individual fires are examined for pyroCb within this region during June–August 2013.

Fig. 1.

Study-region map that is primarily based on fire-prone regions within the effective field of view for GOES-West. Red dots represent 88 intense fires that were inventoried for pyroCb activity. Blue triangles indicate the locations of 10 convective control cases. Thick orange curves indicate the viewing zenith angle of GOES-West. Thin contours indicate surface elevation, with green shading representing regions of forest or mixed forest and chaparral vegetation. Fires that produced specific pyroCb events highlighted in this study are labeled in light blue.

Fig. 1.

Study-region map that is primarily based on fire-prone regions within the effective field of view for GOES-West. Red dots represent 88 intense fires that were inventoried for pyroCb activity. Blue triangles indicate the locations of 10 convective control cases. Thick orange curves indicate the viewing zenith angle of GOES-West. Thin contours indicate surface elevation, with green shading representing regions of forest or mixed forest and chaparral vegetation. Fires that produced specific pyroCb events highlighted in this study are labeled in light blue.

a. Satellite sensor description

GOES-West is in geostationary orbit, with an equatorial subsatellite point of 135°W. The GOES-West Imager features five spectral channels, with central wavelengths ranging from 0.63 to 13.30 μm (Table 1; Hillger and Schmit 2011). The pyroCb detection algorithm is based on BT values for infrared (IR) channels 2 (4 μm), 4 (11 μm), and 6 (13 μm). The nominal pixel size of these IR channels is 16 km2 (4 km). PyroCb detection is based on processing of 15-min GOES-West imagery and quantitative information for daylight hours (roughly 6000 scenes), including additional scenes during rapid scan mode (http://www.ospo.noaa.gov/Operations/GOES/west/rso.html). GOES-West data for this study were obtained from the NOAA Comprehensive Large Array-Data Stewardship System (CLASS; http://www.class.ncdc.noaa.gov/).

Table 1.

GOES-West (GOES-15) spectral-channel characteristics adapted from Hillger and Schmit (2011).

GOES-West (GOES-15) spectral-channel characteristics adapted from Hillger and Schmit (2011).
GOES-West (GOES-15) spectral-channel characteristics adapted from Hillger and Schmit (2011).

As the viewing zenith angle (VZA) of GOES-West increases away from nadir, pixel size also increases from 24–39 km2 in MCONUS (VZA < 60°) to 55–109 km2 in the boreal regions of Canada (60° < VZA < 75°; Fig. 1, orange lines). All pixels with a VZA greater than 75° (Fig. 1, dashed orange line) are excluded. A large portion of Alaska (west of 141°W) is also excluded because the majority of observed fires in that region were located either in close proximity to or beyond the 75° VZA limit (Fig. 1). PyroCb detection above this limit becomes very challenging because of extreme pixel growth (e.g., >110 km2) at large VZAs.

b. Identification of intense-burning events

PyroCb development requires fire activity burning with enough intensity to drive a smoke column to the top of the boundary layer (e.g., Peterson et al. 2014). Any regional analysis of pyroCb activity must therefore begin with a detailed examination of the most intense observed fire events. The GOES Wildfire Automated Biomass Burning Algorithm (WF_ABBA; Prins and Menzel 1994; Prins et al. 1998) retrieves fire pixels by taking advantage of the spectral contrast between a pixel containing fire and the surrounding nonfire cloud-free region using the 4- and 11-μm IR channels (channels 2 and 4). WF_ABBA also provides an estimate of instantaneous fire radiative power (FRP; MW) for detected fire pixels, which is a measure of radiant heat output and is commonly used as a proxy for fire intensity (e.g., Kaufman et al. 1998; Giglio et al. 2003; Schroeder et al. 2010; Peterson et al. 2013). The FRP calculation in WF_ABBA employs the middle-IR method described by Wooster et al. (2005).

Additional FRP information is available from the fire products for the MODIS sensors aboard the Terra and Aqua satellites (MYD14/MOD14, collection 5), which also employ 4- and 11-μm channels similar to WF_ABBA (Giglio et al. 2003; Giglio 2010). Differences in orbit and spatial resolution allow MODIS (~1-km resolution) to detect smaller fires than WF_ABBA, but with less-frequent observations. Whereas saturation of the 4-μm channel can affect WF_ABBA FRP retrievals, the higher saturation temperatures of MODIS allow FRP to be retrieved for nearly every fire pixel detected.

Within the GOES-West study region (Fig. 1), FRP retrievals from the WF_ABBA and MODIS fire products are employed to select clusters of active fire detections, where a large total FRP is retrieved over an area of 625 km2 during either a single day or a period of up to 5 days. This time flexibility ensures inclusion of periods during which opaque cloud cover may have obscured the fire. Locations of peak FRP emissions are isolated for each cluster with a lateral precision of roughly 5.0 km. A threshold of 140 000 MW is employed for both WF_ABBA and MODIS to select the most-intense events. This threshold is low enough to include the majority of large wildfires but high enough to exclude smaller agricultural fires, which are unlikely to produce pyroCb. Application of this threshold during the 2013 period of study provides 88 intense-FRP fires (Fig. 1, red dots) in MCONUS and Canada to examine for pyroCb occurrence.

The list of intense-FRP dates and locations was cross referenced to fires from 2013 in the U.S. Forest Service Incident Information System (Inciweb; http://inciweb.nwcg.gov), thus ensuring that large fires within the GOES-West domain were not missed. Inciweb also serves as the source of the U.S. fire names used in this study (Fig. 1). Canadian fires are identified by province or territory. GOES-West data were processed for each of these fires, with pyroCb detection attempted for all scenes within 12 h after a satellite active fire detection (section 5). Although pyroCb detection requires the same 4- and 11-μm channels used by each fire detection algorithm, the thresholds employed are much different, with pyroCb specifically defined as cloudy pixels. It is therefore impossible for pyroCb detections and valid FRP retrievals to occur at the same time for the same pixel.

c. PyroCb community archive

An online discussion group focused on pyroconvection was founded in 2004 and began daily tabulation of pyroCb activity in 2013 (https://groups.yahoo.com/neo/groups/pyrocb/info). This community effort employs polar-orbiting and geostationary satellite imagery from both passive and active sensors to monitor pyroconvective onset, duration, and residual smoke-plume characteristics. A pyroCb is generally defined as a convective cloud remaining anchored to a wildfire, and it is identified by multiple criteria including an 11-μm BT of less than an approximated homogeneous liquid-water freezing threshold from −35° to −40°C, a relatively large 4-μm BT (Fromm et al. 2010), and reduced visible reflectance when compared with traditional convection (Rosenfeld et al. 2007).

The confidence of each pyroCb detection is augmented with the aid of additional data and tools such as the ultraviolet absorbing aerosol index (e.g., Guan et al. 2010), CALIOP backscatter profiles, and back-trajectory calculations. PyroCb detection to date has therefore been based on many data sources and varying detection criteria. This lack of automation can introduce ambiguity to the archive, likely allowing some pyroCb events to go unnoticed. Despite these limitations, observations from the community provide the only existing archive of known pyroconvection events to date.

The pyroCb community identified several pyroconvective events in Russia, Canada, and the United States during June–August 2013. Of these, 37 events (Table 2) were identified within the domain of GOES-West (Fig. 1), ranging from brief pyroconvective pulses with some observational evidence to support the required low 11-μm BT to multihour or multipulse deep pyroCb. These events coincide with 15–20 individual fires within the study region (Fig. 1), all of which are included in the intense-FRP dataset described above. The pyroCb group’s activity resulted in near-real-time observation of a pyroCb spawned by the tragic Yarnell Hill Fire in Arizona on 30 June 2013 (Hardy and Comfort 2015) and pyroCb activity observed during the California Rim Fire in August 2013 (Peterson et al. 2015). Dissemination of pyroCb detection and plume information also provided guidance for aircraft flights during NASA’s Studies of Emissions and Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC4RS) field mission (https://espo.nasa.gov/home/seac4rs/; Peterson et al. 2015; Toon et al. 2016). In this study, the community archive is applied for developing, testing, and evaluating an automated GOES-West pyroCb detection algorithm, as described in subsequent sections.

Table 2.

Community archive of 37 pyroconvective events during June–August 2013.

Community archive of 37 pyroconvective events during June–August 2013.
Community archive of 37 pyroconvective events during June–August 2013.

3. PyroCb detection algorithm structure

While detection of pyroCb is focused on observations of active fires, the GOES-West algorithm can be run on all pixels in a given scene or subset region. Figure 2 provides a flowchart of the full pyroCb detection algorithm. Inputs include IR BT values for the 4- (BT4), 11- (BT11), and 13-μm (BT13) channels (Table 1). This quantitative information is used in three primary steps: 1) detection of deep convection, 2) separation of pyroconvection from traditional convection, and 3) confirmation that all detected convection is optically thick. During this process, pixels are sorted into four groups: 1) no deep convection, 2) deep convection or high thin clouds, 3) marginal pyroCb, and 4) intense pyroCb.

Fig. 2.

Flowchart of the GOES-West pyroCb detection algorithm (daytime only), with the four output groups highlighted in red or green ovals. Each algorithm step also includes the number of the section where it is described.

Fig. 2.

Flowchart of the GOES-West pyroCb detection algorithm (daytime only), with the four output groups highlighted in red or green ovals. Each algorithm step also includes the number of the section where it is described.

The algorithm produces both imagery and statistical output. Imagery is generated for each input channel, algorithm test, and pyroCb detection category (groups 3 and 4). Summary statistics are calculated on the basis of the center of each fire location identified by analysis of satellite active fire detections (described in section 2b), including all pixels within a 40-, 50-, and 60-km radius. Statistics are calculated for each of the four output groups, including the minimum, maximum, mean, median, and standard deviation of BT4, BT11, and BT13, as well as several additional algorithm tests. The following sections (denoted in Fig. 2) provide details of each algorithm step and test, focusing on both imagery and statistical outputs.

a. Thresholds for intense convection

The cloud-top temperature of all convection can be identified using BT11, with colder values linked to more intense convection and/or high-altitude cloud cover (e.g., Fromm et al. 2010). To ensure the presence of deep convection, potential pyroCb pixels must have a BT11 of less than −20°C. Marginal-pyroCb (MPCB) pixels are defined as deep non-ice or mixed-phase clouds that loft smoke particles to the midtroposphere but fail to reach the tropopause (−35°C < BT11 < −20°C). Periods of sporadic pyroCu are unable to be detected because either the BT11 is greater than −20°C or the spatial extent is small in comparison with the coarse (>4 km) pixel resolution of GOES-West (section 4b).

Following the pyroCb community criterion, intense-pyroCb (IPCB) pixels are defined as ice clouds with BT11 values near and below the homogeneous freezing level (e.g., BT11 < −38°C; Wallace and Hobbs 2006; Rosenfeld et al. 2007). A GOES-West pixel in western North America actually provides the mean BT11 over an area between 20 and 110 km2. The pyroCb algorithm employs an IPCB BT11 threshold of −35°C to account for averaging of subpixel convection (Fig. 2). IPCB cloud-top height is often inferred by matching the minimum observed BT11 against environmental temperature profiles (Fromm et al. 2010). Tropopause temperatures regularly range from −50° to −60°C during IPCB activity (Peterson et al. 2017). Therefore, IPCB pixels coincide with the highest likelihood of injecting smoke particles into the UTLS.

Figure 3 provides BT11 imagery for two IPCB events observed during large wildfires in Idaho and Colorado (included in Table 2), with an average pixel size of 33 km2 within the 50-km statistics radius. In both examples, several pixels have BT11 values of less than −50°C, representative of an IPCB anvil reaching well into the upper troposphere. Figure 4 provides a time series of statistical output over a 36-h period centered on the Pony/Elk Fire pyroCb event displayed in Fig. 3a. The minimum BT11 within the 50-km radius (Fig. 4, solid black curve) is compared with the temperature at the lifting condensation level (LCL) derived using the North American Regional Reanalysis (NARR; http://nomads.ncdc.noaa.gov/data.php#narr_datasets; Mesinger et al. 2006). When BT11 is cooler than the LCL temperature, it indicates convective development and/or high clouds (Fig. 4, blue dots). The BT11 criterion for MPCB and IPCB activity is met for several consecutive time steps during the evening hours of 10 August.

Fig. 3.

GOES-West BT11 imagery for IPCB activity observed during (a) the Pony/Elk Fire in Idaho and (b) Papoose Fire in Colorado. The pink circle indicates the 50-km pyroCb statistics radius.

Fig. 3.

GOES-West BT11 imagery for IPCB activity observed during (a) the Pony/Elk Fire in Idaho and (b) Papoose Fire in Colorado. The pink circle indicates the 50-km pyroCb statistics radius.

Fig. 4.

Statistical output (50-km radius) for the Pony/Elk Fire IPCB event displayed in Fig. 3a. Solid black curve indicates the minimum BT11, with blue dots representing values below the LCL temperature (clouds). The dashed black curve indicates the maximum BTD4–11, and the red curve indicates normalized hourly FRP. IPCB activity meeting the high-LCL microphysics threshold (BTD4–11 > 60°C) is denoted by green vertical lines. False IPCB detections excluded by the opacity test (BTD11–13 < 3.0°C) are denoted by brown vertical lines. Bars at the top indicate hours excluded from pyroCb detection (SZA > 80°), with gray indicating sunset and sunrise and black indicating nighttime.

Fig. 4.

Statistical output (50-km radius) for the Pony/Elk Fire IPCB event displayed in Fig. 3a. Solid black curve indicates the minimum BT11, with blue dots representing values below the LCL temperature (clouds). The dashed black curve indicates the maximum BTD4–11, and the red curve indicates normalized hourly FRP. IPCB activity meeting the high-LCL microphysics threshold (BTD4–11 > 60°C) is denoted by green vertical lines. False IPCB detections excluded by the opacity test (BTD11–13 < 3.0°C) are denoted by brown vertical lines. Bars at the top indicate hours excluded from pyroCb detection (SZA > 80°), with gray indicating sunset and sunrise and black indicating nighttime.

The IPCB and MPCB BT11 thresholds remove pixels that are affected by thin, transparent cirrus clouds, which are usually modulated by warmer BT11 values from the surface. These thresholds also generally preclude fire detection and prevent retrieval of FRP, both of which require warm BT11 values. The red curve in Fig. 4 illustrates the impact on FRP. Large values (>4000 MW) observed for several hours prior to IPCB development rapidly become suppressed (<2000 MW) as cloud cover partially obscures the fire, thereby reducing the number of fire pixels with valid FRP. Nonconvective translucent mid- and high-altitude clouds may not be excluded by the BT11 thresholds and may still exhibit a 4-μm signal large enough for fire detection (section 3c).

The Pony/Elk pyroCb event (Fig. 3a) has the classic visual signature of an anvil anchored to a cluster of fire pixels, but its BT11 values are similar to nearby traditional cumulonimbus clouds (Cb). In many cases, pyroCb anvils are observed embedded within or in very close proximity to traditional Cb (Fig. 3b). It can therefore become difficult or impossible to identify pyroCb events solely on the basis of BT11 thresholds.

b. Microphysical signature of pyroCb

Convection with large aerosol loading will experience a microphysical shift (caused by indirect aerosol effects) toward relatively small cloud droplets and ice particles (Rosenfeld et al. 2007; Reutter et al. 2014; Chang et al. 2015). These unique microphysical properties form the foundation of the pyroCb detection algorithm. As outlined in Fig. 2, the algorithm employs a 4- and 11-μm brightness temperature difference (BTD4–11) to distinguish smoke-polluted IPCB and MPCB activity from unperturbed Cb. This is effective because daytime BT4 contains contributions from both an emitted thermal component and a reflected solar component. Small cloud droplets and ice particles produce a relatively high solar reflectivity, which increases the daytime BT4 when compared with the radiative thermal component alone (e.g., Ellrod and Bailey 2007). Any deep convective tower with a large quantity of small ice particles in the anvil region (cloud top) will therefore produce increased BTD4–11, resulting from a large relative daytime 4-μm reflectivity and increased BT4 values (Melani et al. 2003a,b; Lindsey et al. 2006).

BTD4–11 is uniquely sensitive to ice microphysical perturbations caused by pyroCb, taking advantage of the fact that development occurs over a limited number of global domains where meteorological conditions are favorable and convection is relatively weak (Peterson et al. 2017). Over western North America, for instance, there is a distinct tendency for convection to yield cloud-top heights that fall well below the local tropopause (Sassen and Campbell 2001). Such “warm” anvil shields coincide with relatively large background ice crystals (e.g., Heymsfield et al. 2014), against which smoke-induced microphysical perturbations are more conspicuous. The majority of this “high-based convection” develops in an environment in which a layer of moisture and instability is advected over a dry, deep, and unstable mixed layer (inverted V sounding profile; e.g., Nauslar et al. 2013; Peterson et al. 2017).

Two BTD4–11 tests have been derived on the basis of empirical testing (Fig. 2). The standard test applies a BTD4–11 threshold of 50°C for initial detection over all pyroCb-prone regions. A second BTD4–11 threshold of 60°C is applied specifically to cases coinciding with a very high cloud base [LCL temperature < 0°C or LCL height > 3000 m above ground level (AGL)]. This “high LCL” test is necessary because enhanced daytime 4-μm reflectivity is inherent to high-based convection in some regions of interior western North America (Lindsey et al. 2006). These regions coincide with increasingly higher cloud bases, which reduce the LCL temperature, thereby reducing the residence time of cloud droplets prior to reaching the homogenous freezing level. The result is a larger quantity of ice particles with small effective radii (Lindsey and Grasso 2008), ultimately increasing BTD4–11. Although this secondary high-LCL test is not required for pyroCb detection, its use increases detection confidence in such cases and limits false detections.

A control dataset was built for deep convection in the absence of pyroCb and smoke (Table 3). This dataset includes seven cases of deep convection in MCONUS and Canada with surface elevations ranging from less than 1000 m above mean sea level (MSL) to more than 3000 m MSL (Fig. 1, blue triangles). Several of these cases produced more than one individual convective event during the time intervals considered in Table 3. Three severe-weather events in the high plains (surface elevations of 900–1200 m) are also included. This is a transitional region in which pyroCb are possible but convection is more vigorous than it is at locations farther west. Mean 4-μm daytime reflectivity is thus higher, enhanced by strong mean updraft speeds (Lindsey et al. 2006). Similar to the statistical output of the pyroCb algorithm, maximum BTD4–11 values are calculated for pixels obtained over a 50-km radius, centered on each convective event. Subsequent analysis of these data is based solely on cloudy daytime pixels with a BT11 of less than −20°C.

Table 3.

List of deep-convection control cases (no pyroCb or smoke) during 2013. All control-case analysis is based on pixels within a 50-km radius from the center latitude and longitude.

List of deep-convection control cases (no pyroCb or smoke) during 2013. All control-case analysis is based on pixels within a 50-km radius from the center latitude and longitude.
List of deep-convection control cases (no pyroCb or smoke) during 2013. All control-case analysis is based on pixels within a 50-km radius from the center latitude and longitude.

Figure 5 provides comparisons between various thresholds of BTD4–11 and relevant thermodynamic variables from the 10 control cases. All variables are derived from the NARR using the most unstable parcel below 400 hPa (Peterson et al. 2017). These distributions show that only 5% of the data have maximum BTD4–11 values that are greater than the standard pyroCb microphysics threshold of 50°C. The majority of these data points coincide with much lower LCL temperatures (Fig. 5a) and higher LCL heights (Fig. 5b) than those with BTD4–11 values that are below 50°C. All data points with maximum BTD4–11 values of greater than 60°C coincide with an LCL temperature of less than 0°C and an LCL height of more than 3000 m AGL, thus serving as the basis for the high-LCL microphysics threshold.

Fig. 5.

Distributions of maximum BTD4–11 (50-km radius) are compared with (a) LCL temperature, (b) LCL height, and (c) CAPE, all of which are calculated by lifting the most unstable parcel below 400 hPa. Every pixel from the 10 control cases (Table 3) that passes the deep-convection test (BT11 < −20°C) is included. The boxes are bounded by the 25th and 75th percentiles, with the median indicated as a line bisecting each box. The whiskers indicate the 10th and 90th percentiles, and mean values are displayed as triangles. The corresponding number of data points is included at the top of each box plot. Dashed green lines indicate the two BTD4–11 microphysics thresholds that were used for pyroCb detection. Dashed red lines define a high-LCL environment.

Fig. 5.

Distributions of maximum BTD4–11 (50-km radius) are compared with (a) LCL temperature, (b) LCL height, and (c) CAPE, all of which are calculated by lifting the most unstable parcel below 400 hPa. Every pixel from the 10 control cases (Table 3) that passes the deep-convection test (BT11 < −20°C) is included. The boxes are bounded by the 25th and 75th percentiles, with the median indicated as a line bisecting each box. The whiskers indicate the 10th and 90th percentiles, and mean values are displayed as triangles. The corresponding number of data points is included at the top of each box plot. Dashed green lines indicate the two BTD4–11 microphysics thresholds that were used for pyroCb detection. Dashed red lines define a high-LCL environment.

An increasing relationship also exists between BTD4–11 and convective available potential energy (CAPE; Fig. 5c). An environment with large CAPE will have stronger updraft speeds, which also reduces the residence time between the LCL and the homogeneous freezing level. The CAPE distributions in Fig. 5c display less distinction at high BTD4–11 values than do LCL temperature and height. While thermodynamics favorable for high-based convection and increased mean BTD4–11 are common in mountainous locations, there is little direct dependence on surface elevation (not shown).

Figure 5 shows that 82% of pixels examined from the 10 control cases exhibit BTD4–11 values that are between 20° and 40°C. The empirical microphysical thresholds were thus derived primarily through comparison of these cases, since the details of pyroCb structure and microphysical composition are almost exclusively unknown and very challenging to replicate in a model. Despite this limitation, it can be safely assumed that pyroCb effective radii are much smaller than the control cases because of extreme aerosol loading (Rosenfeld et al. 2007; Reutter et al. 2014; Chang et al. 2015). For instance, MODIS observations show that effective radius values in a Canadian pyroCb anvil were 5–10 μm (Kablick et al. 2015). Radiative transfer model simulations of an unperturbed thunderstorm initialized using a midlatitude summertime standard atmosphere (not shown) suggest that BTD4–11 greater than 50°C in opaque ice-phase cloud tops will likely be observed only when effective radii are smaller than 10 μm. Only 5% of pixels from the control cases exhibited BTD4–11 values above 50°C, which were exclusively a result of high cloud base or high CAPE (Fig. 5). Therefore, while the modeling effort is highly uncertain [on the basis of Khain et al. (2001)], the results are reasonably consistent.

Figure 6a displays BTD4–11 imagery for the same Idaho fire example as Fig. 3a. Blue and red shading indicates pixels with cloud microphysics characteristic of pyroCb. The map in Fig. 6b displays coincident LCL temperatures (shaded) and heights, showing that the majority of traditional Cb observed in interior western MCONUS, including the domain of Fig. 6a, likely have a very high cloud base and therefore an inherently large mean BTD4–11. The fire location has an LCL temperature that is below −6.0°C and a height that is near 4000 m AGL. Applying the high-LCL microphysics test (BTD4–11 > 60°C; Fig. 6a, red shading) isolates the pyroCb anvil and removes most of the nearby traditional Cb, aside from a few intense updraft cores.

Fig. 6.

Analysis of cloud microphysics observed during the Pony/Elk Fire IPCB event in Idaho on 10 Aug using (a) GOES-West BTD4–11 imagery and (b) a map quantifying LCL characteristics. The pink circle in (a) indicates the 50-km pyroCb statistics radius. Shading in (b) indicates LCL temperature, orange and red contours represent LCL height, and brown contours indicate surface elevation. The black box indicates the approximate domain displayed in (a), and the black triangle indicates fire location.

Fig. 6.

Analysis of cloud microphysics observed during the Pony/Elk Fire IPCB event in Idaho on 10 Aug using (a) GOES-West BTD4–11 imagery and (b) a map quantifying LCL characteristics. The pink circle in (a) indicates the 50-km pyroCb statistics radius. Shading in (b) indicates LCL temperature, orange and red contours represent LCL height, and brown contours indicate surface elevation. The black box indicates the approximate domain displayed in (a), and the black triangle indicates fire location.

In Fig. 4, the dashed black curve shows the maximum BTD4–11 for this pyroCb event calculated over the 50-km pyroCb statistics radius. Individual scenes in which at least one pixel meets IPCB criteria (BT11 < −35°C and BTD4–11 > 60°C) are marked by green vertical lines, showing that two distinct pyroCb pulses were observed. Gaps between green vertical lines from the same pyroCb pulse either result from the opacity test (brown vertical lines, described in the next section) or from instances in which all individual pixels within the 50-km statistics radius failed to pass both the ice cloud (BT11 < −35°C) and microphysics (BTD4–11 > 60°C) thresholds required for IPCB detection (Fig. 2).

Analysis periods just before sunset and after sunrise (Fig. 4, gray bars) are excluded because BTD4–11 is unreliable (Lindsey et al. 2006). For instance, BTD4–11 often becomes very large for cloud cover observed in Canada at sunrise. This anomaly results from specular reflection (sun glint) in areas where the solar zenith angle (SZA) and VZA are similar and the relative azimuth is high. To avoid false detections, all potential MPCB and IPCB pixels must have an SZA of 80° or less (Fig. 2). Although pyroCb have been observed during sunset and nighttime (e.g., Fromm et al. 2010), the meteorological conditions that are supportive of development and favorable for intense fire activity are most common during the daytime period considered for detection (Peterson et al. 2017). This suggests that pyroCb are likeliest to begin developing prior to the 80° SZA limit. A subset of pyroCb will persist into the evening hours when detection is no longer possible. If the entire life cycle of a pyroCb occurs after sunset, it will not be detected.

Imagery and statistical output for MPCB and IPCB pixels (Fig. 2, groups 3 and 4) are provided on the basis of both cloud-microphysics thresholds. Individual pixels are only considered to be MPCB or IPCB when they pass both the intense-convection (BT11) and cloud-microphysics (BTD4–11) tests. A pyroCb anvil can therefore contain a mixture of MPCB and IPCB pixels. Large viewing angles will affect the accuracy of BTD4–11 (Schmidt et al. 1995), which is the primary reason for the 75° VZA cutoff highlighted by the dashed orange curve in Fig. 1. In addition, the dependence of 4-μm reflectivity on the scattering angle between the sun, clouds, and satellite suggests that relatively large BTD4–11 values may occasionally appear in the absence of pyroCb (Lindsey and Grasso 2008), especially near the 80° SZA limit. Traditional Cb in close proximity to a fire can also ingest a large quantity of smoke particles (e.g., Barth et al. 2015). The potential effect of this phenomenon on BTD4–11 is explored in section 4c.

c. Cloud-opacity test

All deep convection, including pyroCb, contains an optically thick (opaque) cloud core. The periphery of the anvil region is commonly optically thin (translucent), however. These cloud-edge pixels can trigger false pyroCb detections because bright and hot surfaces with large BT4 values can be observed through translucent clouds, occasionally resulting in BTD4–11 values that are above the pyroCb microphysics thresholds. Removal of cloud-edge pixels and other translucent cloud impacts is therefore essential for accurate pyroCb detection via the BTD4–11 microphysics tests.

Cloud opacity and type are commonly determined using the 11- and 12-μm brightness temperature difference (BTD11–12), known as the split window (e.g., Prata 1989; Barton et al. 1992; Schmit et al. 2001; Ellrod and Bailey 2007). Since the launch of GOES-12 in 2001, however, the 12-μm channel (channel 5) has been replaced with a 13-μm channel (channel 6; Table 1; Schmit et al. 2001; Hillger and Schmit 2011). Located near the edge of the longwave atmospheric window, the 13-μm channel has stronger absorption from carbon dioxide than does its 12-μm counterpart (Schmit et al. 2001). The 13-μm channel on GOES-West (GOES-15) has the same nominal spatial resolution (4 km) as the other IR bands (Table 1; Hillger and Schmit 2011). It results in degraded detection of low clouds but is useful for detection of thin cirrus (Schmit et al. 2001).

Previous studies using the traditional split window show that BTD11–12 values are typically between −1.0° and 1.0°C for opaque cloud types, with larger positive values coinciding with translucent clouds and clear sky (e.g., Barton et al. 1992; Ellrod and Bailey 2007). Opacity thresholds for the BT13 modified split window (BTD11–13) from GOES-West are less certain. Figure 7a provides BTD4–11 imagery for a large pyroCb (Papoose Fire) in close proximity to traditional Cb and other cloud cover in southwestern Colorado (same as Fig. 3b), highlighting a large microphysical signature via the high-LCL test. Figure 7b compares these BTD4–11 values with BTD11–13 for all cloudy pixels (BT11 < −20°C) from the scene in Fig. 7a. Similar to traditional BTD11–12 values, BTD11–13 is low for opaque cloud and increases as clouds become more transparent. The majority of pixels meeting MPCB and IPCB criteria have BTD11–13 values of less than 3.0°C (Fig. 7b, green vertical line). The few pixels meeting MPCB and IPCB criteria with BTD11–13 values that are greater than 3.0°C are primarily cloud-edge pixels. Investigation of several additional known pyroCb events from Table 2 reveals similar results (not shown). Therefore, all pixels passing the initial deep-convection test must also have BTD11–13 values that are less than 3.0°C (Fig. 2).

Fig. 7.

Analysis of cloud-opacity effects on pyroCb detection during the Papoose Fire in Colorado: (a) BTD4–11 for an IPCB event during a period with substantial translucent cloud cover, (b) BTD4–11 vs BTD11–13 for the IPCB case using all pixels in (a) passing the deep-convection test (BT11 < −20°C), (c) BTD4–11 during a period with thin clouds and no pyroconvection, and (d) as in (b), but for the thin-cloud case. Dashed green lines indicate the BTD4–11 and BTD11–13 thresholds for pyroCb detection. Red arrows indicate fire locations.

Fig. 7.

Analysis of cloud-opacity effects on pyroCb detection during the Papoose Fire in Colorado: (a) BTD4–11 for an IPCB event during a period with substantial translucent cloud cover, (b) BTD4–11 vs BTD11–13 for the IPCB case using all pixels in (a) passing the deep-convection test (BT11 < −20°C), (c) BTD4–11 during a period with thin clouds and no pyroconvection, and (d) as in (b), but for the thin-cloud case. Dashed green lines indicate the BTD4–11 and BTD11–13 thresholds for pyroCb detection. Red arrows indicate fire locations.

Along with reducing the potential for cloud-edge errors, the empirically based opacity test also eliminates false pyroCb detections caused by thin clouds passing over a hot fire. Figure 7c shows several pixels with very large BTD4–11 values associated with the same fire as Fig. 7a. This false pyroCb signature is caused by mid- and high-level translucent clouds allowing the signal of the fire (large BT4) to pass through. Corresponding quantitative comparisons (Fig. 7d) show that these few pixels have large BTD11–13 values (>6.0°C) and will therefore be excluded by the opacity test. The large increase in maximum BTD4–11 values observed after the pyroCb in Fig. 4 is also caused by thin clouds over a hot fire. IPCB detections excluded by the opacity test are displayed as brown vertical lines. These false detections were avoided as a result of the combination of the BTD11–13 opacity test and proximity to sunset.

All anvil clouds gradually diffuse as they detach from the primary updraft core and advect downwind. These decaying “orphan anvils” can be a contaminating factor when multiple fires are located in close proximity to each other. All pyroCb orphan anvils will be excluded by the opacity test at some distance downwind from the contributing fire or Cb core, however, and are therefore unlikely to produce false pyroCb detections when passing over fires far downwind. These situations highlight the importance of a BTD11–13 opacity test for reducing pyroCb detection errors, but it may have difficulty screening mountain-wave clouds, which tend to be optically thick and populated by very small ice crystals (e.g., Neiman and Shaw 2003), resulting in large BTD4–11 values.

4. Initial algorithm assessment

PyroCb have been observed in several regions of western North America using varying VZA and pixel size (Fig. 1). In all detection scenarios, pyroCb magnitude, anvil size, microphysics, and proximity to traditional Cb must be considered. The following sections address the performance of the pyroCb detection algorithm on the basis of these considerations using several events from the 2013 community archive (Table 2) and an event that was previously undetected.

a. PyroCb detection capabilities

Output imagery from the detection algorithm is based on BTD4–11 values for each MPCB and IPCB pixel (Fig. 2, groups 3 and 4) after application of the cloud-top BT11 and opacity tests. Figure 8a provides example pyroCb detection imagery for the Carpenter Fire in Nevada (pixel size = 27 km2). Orange and pink shading indicates pixels that meet IPCB criteria, and yellow and green shading indicates pixels that meet MPCB criteria. The majority of shaded pixels are IPCB, highlighting the pyroCb updraft core. A few pixels along the periphery of the core meet MPCB criteria. Much of the thin anvil region is excluded by the opacity test, however. Figure 8b shows IPCB activity observed during a large fire in Manitoba, Canada (pixel size > 100 km2). Similar to Fig. 8a, the majority of pixels are IPCB, but the core of the anvil is larger. This event shows that pyroCb detection is possible for high-latitude fires, despite degraded GOES-West performance (large VZA and pixel size).

Fig. 8.

Output imagery from the GOES-West pyroCb detection algorithm for IPCB events observed during (a) the Carpenter Fire in Nevada, (b) a large fire in Manitoba, (c) the Papoose Fire in Colorado, and (d) the Yarnell Hill Fire in Arizona. Gray/white shading indicates BT11 for nonpyroconvective pixels, yellow/green shading indicates BTD4–11 for pixels that meet MPCB criteria (group 3 in Fig. 2), and orange/red shading indicates BTD4–11 for pixels that meet IPCB criteria (group 4 in Fig. 2). Pink circles indicate the 50-km pyroCb statistics radius.

Fig. 8.

Output imagery from the GOES-West pyroCb detection algorithm for IPCB events observed during (a) the Carpenter Fire in Nevada, (b) a large fire in Manitoba, (c) the Papoose Fire in Colorado, and (d) the Yarnell Hill Fire in Arizona. Gray/white shading indicates BT11 for nonpyroconvective pixels, yellow/green shading indicates BTD4–11 for pixels that meet MPCB criteria (group 3 in Fig. 2), and orange/red shading indicates BTD4–11 for pixels that meet IPCB criteria (group 4 in Fig. 2). Pink circles indicate the 50-km pyroCb statistics radius.

The pyroCb events in Figs. 8a and 8b also highlight the application of both microphysics tests, with the Carpenter Fire pyroCb meeting high-LCL criteria (LCL height = 3039 m and temperature = −1.0°C). The Manitoba pyroCb developed with similar meteorological conditions (Peterson et al. 2017), but its cloud-base characteristics (LCL height = 2024 m and temperature = 9.0°C) did not require the high-LCL microphysics test, which is typical of low-elevation Canadian events (Lindsey et al. 2006).

Each pyroCb in Fig. 8 is located in a convectively active region, with several traditional Cb located somewhere in the image domain. Cloud-microphysics tests allowed the algorithm to detect a pyroCb in very close proximity to traditional Cb during the Colorado Papoose Fire (Fig. 8c; same event as Figs. 3b and 7a), including a small portion of an orphan pyroCb anvil detached from the New Mexico Silver Fire, which is still thick enough to pass the opacity test. On occasion, pyroCb are completely embedded within a cluster of traditional Cb, such as during the Yarnell Hill Fire in Arizona (Fig. 8d). The algorithm identified a pyroCb core near the time when 19 firefighters lost their lives because of a rapid shift in fire-spread direction that was triggered by outflow from nearby traditional Cb (Hardy and Comfort 2015).

The pyroCb displayed in Figs. 8a, 8c,and 8d are significant events included in the community archive (Table 2). The large Manitoba pyroCb (Fig. 8b) is not included in the archive, likely because of surrounding high, thin clouds disguising the anvil in standard BT11 imagery. Automated application of the algorithm microphysics and opacity tests to thousands of GOES-West scenes allowed this event to be detected. Aside from the Yarnell Hill event (Fig. 8d), each anvil core is expansive, composed primarily of IPCB pixels, with a few MPCB pixels along the edges. The following section explains detection limitations associated with the size of the pyroCb anvil.

b. Anvil-size limitations

While pyroCb are known for reaching the UTLS and producing large anvils (Figs. 8a,c,d), several reach only the mid–upper troposphere. These events typically have a reduced lifetime and smaller anvils, which can impact pyroCb detection. The BT11 imagery in Fig. 9a shows pyroconvection developing during the Thompson Ridge Fire in New Mexico (pixel size = 31 km2). The cloud-top BT11 is between −35° and −40°C, which is cold enough to meet IPCB criteria, but the anvil region is very narrow and is warmer than the large anvils associated with nearby traditional Cb (<−50°C). This produces a small sample of pixels available for pyroCb detection, which is reduced further by the opacity test. The few remaining candidate pixels fail to pass both microphysics tests (Fig. 9b). Therefore, while this potential pyroCb is included in the community archive (Table 2), it is not detectable by the GOES-West algorithm.

Fig. 9.

Analysis of anvil-size limitations for pyroCb detection: (a) BT11 over the Thompson Ridge Fire in New Mexico on 4 Jun, (b) BTD4–11 imagery for the Thompson Ridge case, (c) BT11 over an IPCB event during the Pony/Elk Fire in Idaho on 9 Aug, and (d) BTD4–11 imagery for the Pony/Elk case. Red triangles indicate fire locations.

Fig. 9.

Analysis of anvil-size limitations for pyroCb detection: (a) BT11 over the Thompson Ridge Fire in New Mexico on 4 Jun, (b) BTD4–11 imagery for the Thompson Ridge case, (c) BT11 over an IPCB event during the Pony/Elk Fire in Idaho on 9 Aug, and (d) BTD4–11 imagery for the Pony/Elk case. Red triangles indicate fire locations.

Small and narrow anvils can also affect the efficacy of microphysics tests during detectable events. The pyroCb observed during the Pony/Elk Fire on 9 August (Fig. 9c; Table 2) occurred in an environment that was favorable for a very high cloud base and large mean BTD4–11 (LCL height = 5293 m and temperature = −19.0°C). Anvils produced by nearby traditional Cb had BTD4–11 values that were greater than 50°C (Fig. 9d), suggesting the high-LCL microphysics test (BTD4–11 > 60°C) was most appropriate. The maximum BTD4–11 for pixels within this anvil passed the standard microphysics-test threshold (BTD4–11 > 50°C) but fell short of the high-LCL threshold. Similar to the Thompson Ridge event (Figs. 9a,b), the narrow anvil produced a small sample of pixels to test for pyroCb microphysics, none of which exceeded the high-LCL threshold.

The small and narrow pyroCb anvils in Fig. 9 highlight the need for including output from both microphysics tests, but any pyroCb occurring under high-cloud-base conditions that fails to pass the high-LCL microphysics test has reduced detection confidence. As VZA increases, the size of the anvil relative to pixel size also becomes an important limitation. PyroCb anvils in Canada (pixel size = 55–109 km2) must therefore be especially large to be detectable via the GOES-West algorithm (Fig. 8b). The only exception is a pyroCb that is completely embedded within traditional Cb. As highlighted by the Yarnell Hill event (Fig. 8d), detection of small pyroCb cores is possible when surrounded by other opaque high-altitude clouds with little smoke influence.

c. Sensitivity to traditional convection and smoke particles

Analysis of the control dataset (Table 3) shows that a few traditional Cb produce BTD4–11 values that are above both pyroCb microphysics thresholds (Fig. 5). These convective events must be examined in greater detail to determine potential for false pyroCb detections. Figure 10a shows a complex of deep convection in western Montana on 18 June with a large cluster of pixels producing BTD4–11 values that are above the high-LCL pyroCb microphysics threshold. Supercell thunderstorms in eastern Colorado are also known for producing large BTD4–11 values near the updraft core (Fig. 10b). Although the LCL is lower in the supercell case, both are associated with large CAPE values (1000–2500 J kg−1). This enhances updraft velocity (Lindsey et al. 2006) and produces smaller cloud particles, which increase BTD4–11 values (Fig. 5c).

Fig. 10.

GOES-West pyroCb detection algorithm output imagery highlighting false detections during (a) traditional Cb in Montana, (b) a supercell in Colorado, (c) traditional Cb downwind from several California fires, and (d) traditional convection downwind from Idaho fire activity. Gray/white shading indicates BT11 for nonpyroconvective pixels, yellow/green shading indicates BTD4–11 for pixels that meet MPCB criteria (group 3 in Fig. 2), and orange/red shading indicates BTD4–11 for pixels that meet IPCB criteria (group 4 in Fig. 2). The red arrow in (c) and (d) indicates the mean low- to midlevel wind vector (700–500 hPa), and yellow triangles indicate fire locations.

Fig. 10.

GOES-West pyroCb detection algorithm output imagery highlighting false detections during (a) traditional Cb in Montana, (b) a supercell in Colorado, (c) traditional Cb downwind from several California fires, and (d) traditional convection downwind from Idaho fire activity. Gray/white shading indicates BT11 for nonpyroconvective pixels, yellow/green shading indicates BTD4–11 for pixels that meet MPCB criteria (group 3 in Fig. 2), and orange/red shading indicates BTD4–11 for pixels that meet IPCB criteria (group 4 in Fig. 2). The red arrow in (c) and (d) indicates the mean low- to midlevel wind vector (700–500 hPa), and yellow triangles indicate fire locations.

Recent studies show that few pyroCb develop with CAPE values of greater than 1000 J kg−1 and all are devoid of meteorological conditions that are favorable for supercells (Fromm et al. 2010; Peterson et al. 2015, 2017). It is therefore unlikely that these microphysically unique convective events will occur at the same time as a pyroCb. Intense updraft cores from traditional Cb may show as a small cluster of IPCB or MPCB pixels as CAPE values increase, however, such as during the Pony/Elk pyroCb on 10 August (Fig. 6b). This phenomenon is most likely to be observed in fire-prone regions associated with the highest mean CAPE during the fire season, including the Front Range of Colorado, Black Hills of South Dakota, and some regions of central Montana (Potter and Anaya 2015). The potential for using CAPE as an additional algorithm constraint will be explored in future research.

Along with thermodynamic effects, microphysical variability relative to traditional Cb developing in a smoky environment must be considered. Recent airborne field campaigns suggest that Cb can ingest a large quantity of smoke particles and loft them to the UTLS (e.g., Barth et al. 2015). Two potential cases were discovered within the GOES-West study region in 2013. Figure 10c highlights a small pyroCb observed during the Rim Fire on 21 August (Peterson et al. 2015) in close proximity to several additional wildfires. A large complex of traditional Cb, located immediately downwind with similar meteorological conditions, was identified as IPCB pixels. In a similar way, Fig. 10d shows a large complex of traditional Cb detected as IPCB pixels downwind of several large fires in Idaho (e.g., Pony/Elk and Beaver Creek). Although cloud-base characteristics vary, CAPE values (10–700 Jkg−1) are not supportive of extreme updrafts. This suggests that the cloud microphysics may have been modified from ingestion of ambient smoke particles, ultimately producing false pyroCb detections. The GOES-West algorithm may therefore be applicable for identifying other convective avenues for lofting smoke particles. Additional research is required to support this hypothesis.

5. 2013 inventory of intense pyroCb

The combination of the GOES-West pyroCb detection algorithm (Fig. 2) and the intense-FRP dataset (section 2b) provides an opportunity to build the first systematic inventory of pyroCb events in western North America for the most active portion (June–August) of the 2013 fire season. Every daytime scene (SZA < 80°) is examined for potential pyroCb activity within the three pyroCb statistics radii (40, 50, and 60 km) during the entire lifetime of all 88 fires in the intense-FRP dataset (Fig. 1). The inventory is based on the 60-km radius whenever possible. The 40- and 50-km radii are only applied when the fire being inventoried is located in close proximity to other significant fires. To be included in the inventory, some FRP must be observed in at least one scene during the preceding 12 h. The time of a given pyroCb is defined as the first GOES-West scene in which at least one pixel in the appropriate statistics radius meets IPCB or MPCB criteria. All scenes meeting IPCB or MPCB criteria within a 6-h window are considered to be part of the same pyroCb event. This inventory is used by Peterson et al. (2017) as the basis for building the first conceptual model for pyroCb development.

Figure 11 provides an example pyroCb inventory time series for the Silver Fire in New Mexico (pixel size = 29 km2). Similar to Fig. 4, minimum BT11 and maximum BTD4–11 values are plotted as solid and dashed black lines, respectively. The isolated nature of this fire is conducive to application of the 60-km statistics radius. The corresponding FRP time series (Fig. 11, red curve) is plotted hourly using the method described by Peterson et al. (2015). Inventoried initiation times of four pyroCb events are displayed as dashed green vertical lines, with solid yellow lines indicating each individual scene that contains at least one IPCB pixel within the 60-km radius. The high-LCL microphysics threshold (BTD4–11 > 60°C) is applied because all four pyroCb occur with meteorological conditions that are favorable for a high cloud base and large mean BTD4–11. Green arrows show that the pyroCb algorithm detects four of the five pyroconvection events included in the community archive (Table 2). The undetected event is a result of a short-lived thin and narrow anvil with no pixels exceeding the BTD4–11 pyroCb detection thresholds. While 21 and 22 June also appear to meet pyroCb criteria during daytime, these were cases of thin-cloud contamination, which were removed by the opacity test (BTD11–13 < 3.0°C).

Fig. 11.

PyroCb algorithm statistical output (60-km radius) for the Silver Fire in New Mexico. The solid black curve indicates the minimum BT11, with blue dots representing values below the LCL temperature (clouds). The dashed black curve indicates the maximum BTD4–11, and the red curve indicates normalized hourly FRP. IPCB activity that meets the BTD4–11 microphysics thresholds (>50° or 60°C) is denoted by yellow vertical lines. The time of each individual event included in the IPCB inventory (Table 4) is marked by a dashed green vertical line. Green arrows indicate pyroconvection that is included in the community archive (Table 2). A brown star indicates the multiple-pulse IPCB event displayed in Fig. 12, below. Black bars at the top indicate hours that were excluded from pyroCb detection (SZA > 80°).

Fig. 11.

PyroCb algorithm statistical output (60-km radius) for the Silver Fire in New Mexico. The solid black curve indicates the minimum BT11, with blue dots representing values below the LCL temperature (clouds). The dashed black curve indicates the maximum BTD4–11, and the red curve indicates normalized hourly FRP. IPCB activity that meets the BTD4–11 microphysics thresholds (>50° or 60°C) is denoted by yellow vertical lines. The time of each individual event included in the IPCB inventory (Table 4) is marked by a dashed green vertical line. Green arrows indicate pyroconvection that is included in the community archive (Table 2). A brown star indicates the multiple-pulse IPCB event displayed in Fig. 12, below. Black bars at the top indicate hours that were excluded from pyroCb detection (SZA > 80°).

Extending this method to the remaining fires in Fig. 1 provides the foundation for the pyroCb inventory. The bulk of the processing is automated, but imagery output is also checked manually to remove double counting caused by a pyroCb anvil extending into the statistics area of multiple fires. Table 4 provides the details of 26 pyroCb events that were inventoried by GOES-West from 11 individual fires. The majority of these were initiated when the SZA was less than 60° in interior sections of the western MCONUS, where traditional convection often fails to reach the tropopause (Sassen and Campbell 2001). The approximate lifetime of each pyroCb event ranged from 1 to 5 h. Two events likely persisted after sunset (80° SZA limit). Only three pyroCb events were inventoried in Canada. The systematic approach to the GOES-West inventory also identified six previously undocumented intense-pyroCb events (Table 4, last column).

Table 4.

An inventory of 26 intense events from 11 fires during June–August 2013 from the GOES-West pyroCb detection algorithm. The number of individual pyroCb pulses = 31.

An inventory of 26 intense events from 11 fires during June–August 2013 from the GOES-West pyroCb detection algorithm. The number of individual pyroCb pulses = 31.
An inventory of 26 intense events from 11 fires during June–August 2013 from the GOES-West pyroCb detection algorithm. The number of individual pyroCb pulses = 31.

The inventory (26 events) contains fewer total pyroCb than the community archive (37 events) does (Table 2). A few of the events in the community archive are fairly ambiguous, however, and are labeled as pyroCb on the basis of collective agreement. Several events from the community archive that are not included in the inventory result from a combination of large VZA (large pixel footprint) and small anvil size, such as the Thompson Ridge example (Figs. 9a,b). This explains the relative lack of inventoried events in Canada. The requirement for a cold, thick, and expansive anvil also explains the absence of events that are composed entirely of MPCB pixels. Therefore, the inventory (Table 4) includes only intense pyroCb events, all of which have a high likelihood of injecting smoke within the UTLS. Many midtropospheric pyroCb injection scenarios and all sporadic pyroCu are excluded.

Previous studies (e.g., Rosenfeld et al. 2007), along with the Pony/Elk Fire time series (Fig. 4), show that individual pyroCb events (6-h window) can produce multiple pulses of activity. Figure 12 shows three distinct anvil clouds exhibiting pyroCb microphysical properties (large BTD4–11 values) downwind from the New Mexico Silver Fire on 28 June (Fig. 11, brown star). Because of cloud-opacity requirements, the pyroCb detection algorithm (Fig. 12, red shading) only highlights the most recent anvil, which is attached to the fire. The earlier pulses seen in Fig. 12 were each detected in previous GOES-West scenes (Fig. 11). The combination of these imagery techniques and statistical time series outputs (Figs. 4 and 11) is employed to manually identify each individual pulse of pyroCb activity associated with the 2013 inventory. Table 4 shows that three pyroCb events produced multiple pulses, resulting in a total of 31 individual pulses from the 26 inventoried events. When considering that each individual anvil is large and cold enough to be detected by the GOES-West algorithm, it is likely that a pyroCb event with multiple pulses will also produce multiple injections of smoke particles within the UTLS.

Fig. 12.

A multiple-pulse IPCB event that was observed during the New Mexico Silver Fire on 28 Jun. Blue and green shading is indicative of clouds composed of small ice particles (BTD4–11 > 50°C). Red shading indicates all pixels that meet MPCB or IPCB criteria. The pink circle indicates the 60-km pyroCb statistics radius that was used in Fig. 11.

Fig. 12.

A multiple-pulse IPCB event that was observed during the New Mexico Silver Fire on 28 Jun. Blue and green shading is indicative of clouds composed of small ice particles (BTD4–11 > 50°C). Red shading indicates all pixels that meet MPCB or IPCB criteria. The pink circle indicates the 60-km pyroCb statistics radius that was used in Fig. 11.

6. Summary and conclusions

This study has developed a novel automated pyroCb detection algorithm for GOES-West (GOES-15), which is employed to build the first systematic inventory of pyroCb in western North America during the primary fire season (June–August) of 2013. PyroCb detection begins with identification of deep convection near active fires via the thermal infrared (11 μm) brightness temperature (BT11). The algorithm distinguishes between marginal-pyroCb (MPCB) pixels and intense-pyroCb (IPCB) pixels, which are likely associated with smoke particles reaching the UTLS.

The primary component of the detection algorithm is identification of unique pyroCb microphysical properties, characterized by unusually large near-infrared (4 μm) brightness temperature (BT4) when compared with traditional convection. Application of the 4- and 11-μm brightness temperature difference (BTD4–11) allows the unique cloud microphysics of smoke-polluted IPCB and MPCB activity to be separated from more pristine traditional convection. Two thresholds are employed on the basis of regional variability of BTD4–11, which is driven by meteorological conditions. A cloud-opacity test that is based on the 11- and 13-μm brightness temperature difference (BTD11–13) is also included to reduce potential cloud-edge noise and detection errors from thin clouds passing over a hot fire. This algorithm is suitable for use in near–real time and therefore could provide useful information for forecasting of smoke transport and air quality. The dependence of BTD4–11 on variation in 4-μm reflectivity (solar component) currently limits pyroCb detection to daytime scenes with a solar zenith angle of 80° or less.

An initial assessment of the pyroCb detection algorithm reveals that cloud-microphysics tests are useful for separating pyroCb from traditional convection. This is paramount for pyroCb detection because many 2013 events were observed in a convectively active region. The algorithm successfully captures individual intense-pyroCb events, pyroCb embedded within traditional convection, and multiple, short-lived pulses of pyroCb activity. During daytime, small anvil size is the primary limitation of the GOES-West algorithm, especially in regions with large viewing angles (large pixel size). Although the algorithm is sensitive to traditional convection with extreme updraft velocities (e.g., large CAPE), this situation is uncommon during pyroCb development. Algorithm assessment also reveals potential sensitivity to traditional convection that has ingested a large quantity of smoke particles from nearby fires.

Application of the pyroCb detection algorithm to 88 intense wildfires in western North America during 2013 provides an inventory of 26 pyroCb events. These events produced 31 individual pulses of intense-pyroCb activity, all of which were capable of injecting smoke within the UTLS. Although several known pyroconvection events with small anvils were excluded because of anvil size limitations, the systematic approach of the inventory identified six previously undocumented intense-pyroCb events. Application of the GOES-West algorithm to the entire global constellation of geostationary sensors will therefore provide an unprecedented understanding of UTLS smoke injections and their frequency of occurrence. Future studies will also explore the potential utility of using visible reflectance for improved discrimination of pyroCb relative to traditional Cb (e.g., Rosenfeld et al. 2007) as well as the conversion of BT11 to IPCB cloud-top altitude for initializing smoke-transport models (e.g., Fromm et al. 2010).

A companion study (Peterson et al. 2017) employs the inventory described in this paper to show that western North American pyroCb develop in an environment that is favorable for traditional high-based convection (dry thunderstorms). The regular occurrence of the meteorological conditions associated with pyroCb, along with the large number of inventoried intense-pyroCb anvils during 2013, suggests that injections of smoke particles within the UTLS are likely an endemic feature of regional summer climate. Improved exploitation of satellite observations for pyroCb detection and monitoring is therefore paramount for understanding the role of this phenomenon in the climate system.

The next generation of geostationary satellite sensors, including the Advanced Baseline Imager aboard the U.S. GOES-R and GOES-S satellites (http://www.goes-r.gov/), as well as the Advanced Himawari Imager aboard Japan’s recently launched Himawari-8 satellite (http://www.data.jma.go.jp/mscweb/en/himawari89/space_segment/spsg_ahi.html), will offer significant advantages for pyroCb and fire detection. These sensors feature 16 spectral channels at a finer nominal spatial resolution of 2 km in the relevant IR channels. This will facilitate improved retrievals of particle size (effective radius), which will alleviate dependence on BTD4–11 thresholds. All next-generation sensors also provide temporal sampling every 5–15 min. These combined improvements will greatly enhance characterization of cloud microphysics and opacity for both large and small pyroCb anvils, especially in regions with large satellite viewing angles (e.g., Canada). All next-generation sensors provide an excellent opportunity to refine the current GOES-West algorithm for improved pyroCb detection and inventory capability.

Acknowledgments

This research was performed while David Peterson held a National Research Council Research Associateship Award at the Naval Research Laboratory (NRL) in Monterey, California. We are grateful to Jared Marquis at the University of North Dakota for performing radiative transfer simulations to guide this study. Support was provided by the Naval Research Laboratory Base Funding Program. Additional support was provided by the NASA Air Quality Applied Science Team, under Award NNH09ZDA001N-AQAST. GOES-West data for this study were obtained from the NOAA Comprehensive Large Array-Data Stewardship System (CLASS; http://www.class.ncdc.noaa.gov/). Imagery products using the methods described in this paper are posted in near–real time at the NRL pyroCb website: http://www.nrlmry.navy.mil/pyrocb-bin/pyrocb.cgi.

REFERENCES

REFERENCES
Barth
,
M. C.
, and Coauthors
,
2015
:
The Deep Convective Clouds and Chemistry (DC3) field campaign
.
Bull. Amer. Meteor. Soc.
,
96
,
1281
1309
, doi:.
Barton
,
I. J.
,
A. J.
Prata
,
I. G.
Watterson
, and
S. A.
Young
,
1992
:
Identification of the Mount Hudson volcanic cloud over SE Australia
.
Geophys. Res. Lett.
,
19
,
1211
1214
, doi:.
Chang
,
D.
, and Coauthors
,
2015
:
Comprehensive mapping and characteristic regimes of aerosol effects on the formation and evolution of pyro-convective clouds
.
Atmos. Chem. Phys.
,
15
,
10 325
10 348
, doi:.
Ellrod
,
G. P.
, and
A. A.
Bailey
,
2007
:
Assessment of aircraft icing potential and maximum icing altitude from geostationary meteorological satellite data
.
Wea. Forecasting
,
22
,
160
174
, doi:.
Fromm
,
M. D.
, and
R.
Servranckx
,
2003
:
Transport of forest fire smoke above the tropopause by supercell convection
.
Geophys. Res. Lett.
,
30
,
1542
, doi:.
Fromm
,
M. D.
, and Coauthors
,
2000
:
Observations of boreal forest fire smoke in the stratosphere by POAM III, SAGE II, and lidar in 1998
.
Geophys. Res. Lett.
,
27
,
1407
1410
, doi:.
Fromm
,
M. D.
,
R.
Bevilacqua
,
R.
Servranckx
,
J.
Rosen
,
J. P.
Thayer
,
J.
Herman
, and
D.
Larko
,
2005
:
Pyro-cumulonimbus injection of smoke to the stratosphere: Observations and impact of a super blowup in northwestern Canada on 3–4 August 1998
.
J. Geophys. Res.
,
110
,
D08205
, doi:.
Fromm
,
M. D.
,
A.
Tupper
,
D.
Rosenfeld
,
R.
Servranckx
, and
R.
McRae
,
2006
:
Violent pyro-convective storm devastates Australia’s capital and pollutes the stratosphere
.
Geophys. Res. Lett.
,
33
,
L05815
, doi:.
Fromm
,
M. D.
,
O.
Torres
,
D.
Diner
,
D.
Lindsey
,
B.
Vant Hull
,
R.
Servranckx
,
E. P.
Shettle
, and
Z.
Li
,
2008a
:
Stratospheric impact of the Chisholm pyrocumulonimbus eruption: 1. Earth-viewing satellite perspective
.
J. Geophys. Res.
,
113
,
D08202
, doi:.
Fromm
,
M. D.
, and Coauthors
,
2008b
:
Stratospheric impact of the Chisholm pyrocumulonimbus eruption: 2. Vertical profile perspective
.
J. Geophys. Res.
,
113
,
D08203
, doi:.
Fromm
,
M. D.
,
D.
Lindsey
,
R.
Servranckx
,
G.
Yue
,
T.
Trickl
,
R.
Sica
,
P.
Doucet
, and
S.
Godin-Beekmann
,
2010
:
The untold story of pyrocumulonimbus
.
Bull. Amer. Meteor. Soc.
,
91
,
1193
1209
, doi:.
Fromm
,
M. D.
,
R. H. D.
McRae
,
J. J.
Sharples
, and
G. P.
Kablick
III
,
2012
:
Pyrocumulonimbus pair in Wollemi and Blue Mountains National Parks, 22 November 2006
.
Aust. Meteor. Oceanogr. J.
,
62
,
117
126
.
Giglio
,
L.
,
2010
: MODIS Collection 5 active fire product user’s guide, version 2.4. Science Systems and Applications, Inc., and University of Maryland Department of Geography, 61 pp. [Available online at http://www.fao.org/fileadmin/templates/gfims/docs/MODIS_Fire_Users_Guide_2.4.pdf.]
Giglio
,
L.
,
J.
Descloitres
,
C. O.
Justice
, and
Y. J.
Kaufman
,
2003
:
An enhanced contextual fire detection algorithm for MODIS
.
Remote Sens. Environ.
,
87
,
273
282
, doi:.
Guan
,
H.
,
R.
Esswein
,
J.
Lopez
,
R.
Bergstrom
,
A.
Warnock
,
M.
Follette-Cook
,
M.
Fromm
, and
L. T.
Iraci
,
2010
:
A multi-decadal history of biomass burning plume heights identified using aerosol index measurements
.
Atmos. Chem. Phys.
,
10
,
6461
6469
, doi:.
Hardy
,
K.
, and
L. K.
Comfort
,
2015
:
Dynamic decision processes in complex, high-risk operations: The Yarnell Hill Fire, June 30, 2013
.
Saf. Sci.
,
71
,
39
47
, doi:.
Heymsfield
,
A.
,
D.
Winker
,
M.
Avery
,
M.
Vaughan
,
G.
Diskin
,
M.
Deng
,
V.
Mitev
, and
R.
Matthey
,
2014
:
Relationships between ice water content and volume extinction coefficient from in situ observations for temperatures from 0° to −86°C: Implications for spaceborne lidar retrievals
.
J. Appl. Meteor. Climatol.
,
53
,
479
505
, doi:.
Hillger
,
D. W.
, and
T. J.
Schmit
,
2011
: The GOES-15 science test: Imager and sounder radiance and product validations. NOAA Tech. Rep. NESDIS 141, 101 pp. [Available online at http://rammb.cira.colostate.edu/projects/goes-p/NOAA_Technical_Report_141_GOES-15_Science_Test.pdf.]
Kablick
,
G.
,
M.
Fromm
,
S.
Miller
,
P.
Partain
,
G.
Nedoluha
,
D.
Peterson
, and
L.
Zhanqing
,
2015
: Biomass-burning aerosol effects on convective cloud properties and in the detrained UTLS environment: A pyroCb case study. Abstracts, 2015 Fall Meeting, San Francisco, CA, Amer. Geophys. Union, Abstract A34A-05.
Kaufman
,
Y. J.
, and Coauthors
,
1998
:
Potential global fire monitoring from EOS-MODIS
.
J. Geophys. Res.
,
103
,
32 215
32 238
, doi:.
Khain
,
A. P.
,
D.
Rosenfeld
, and
A.
Pokrovsky
,
2001
:
Simulating convective clouds with sustained supercooled liquid water down to 37.5°C using a spectral microphysics model
.
Geophys. Res. Lett.
,
28
,
3887
3890
, doi:.
Lareau
,
N. P.
, and
C. B.
Clements
,
2016
:
Environmental controls on pyrocumulus and pyrocumulonimbus initiation and development
.
Atmos. Chem. Phys.
,
16
,
4005
4022
, doi:.
Lindsey
,
D. T.
, and
M.
Fromm
,
2008
:
Evidence of the cloud lifetime effect from wildfire-induced thunderstorms
.
Geophys. Res. Lett.
,
35
,
L22809
, doi:.
Lindsey
,
D. T.
, and
L.
Grasso
,
2008
:
An effective radius retrieval for thick ice clouds using GOES
.
J. Appl. Meteor. Climatol.
,
47
,
1222
1231
, doi:.
Lindsey
,
D. T.
,
D. W.
Hillger
,
L.
Grasso
,
J. A.
Knaff
, and
J. F.
Dostalek
,
2006
:
GOES climatology and analysis of thunderstorms with enhanced 3.9-μm reflectivity
.
Mon. Wea. Rev.
,
134
,
2342
2353
, doi:.
Luderer
,
G.
,
J.
Trentmann
, and
M. O.
Andreae
,
2009
:
A new look at the role of fire-released moisture on the dynamics of atmospheric pyro-convection
.
Int. J. Wildland Fire
,
18
,
554
562
, doi:.
Melani
,
S.
,
E.
Cattani
,
V.
Levizzani
,
M.
Cervino
,
F.
Torricella
, and
M. J.
Costa
,
2003a
:
Radiative effects of simulated cirrus clouds on top of a deep convective storm in METEOSAT second generation SEVIRI channels
.
Meteor. Atmos. Phys.
,
83
,
109
122
, doi:.
Melani
,
S.
,
E.
Cattani
,
F.
Torricella
, and
V.
Levizzani
,
2003b
:
Characterization of plumes on top of deep convective storm using AVHRR imagery and radiative transfer simulations
.
Atmos. Res.
,
67–68
,
485
499
, doi:.
Mesinger
,
F.
, and Coauthors
,
2006
:
North American Regional Reanalysis
.
Bull. Amer. Meteor. Soc.
,
87
,
343
360
, doi:.
Nauslar
,
N. J.
,
M. L.
Kaplan
,
J.
Wallmann
, and
T. J.
Brown
,
2013
:
A forecast procedure for dry thunderstorms
.
J. Oper. Meteor.
,
1
,
200
214
, doi:.
Neiman
,
P. J.
, and
J. A.
Shaw
,
2003
:
Coronas and iridescence in mountain wave clouds over northeastern Colorado
.
Bull. Amer. Meteor. Soc.
,
84
,
1373
1386
, doi:.
Paugam
,
R.
,
M.
Wooster
,
S.
Freitas
, and
M. V.
Martin
,
2016
:
A review of approaches to estimate wildfire plume injection height within large-scale atmospheric chemical transport models
.
Atmos. Chem. Phys.
,
16
,
907
925
, doi:.
Peterson
,
D. A.
,
J.
Wang
,
C.
Ichoku
,
E.
Hyer
, and
V.
Ambrosia
,
2013
:
A sub-pixel-based calculation of fire radiative power from MODIS observations: 1 Algorithm development and initial assessment
.
Remote Sens. Environ.
,
129
,
262
279
, doi:.
Peterson
,
D. A.
,
E. J.
Hyer
, and
J.
Wang
,
2014
:
Quantifying the potential for high-altitude smoke injection in the North American boreal forest using the standard MODIS fire products and subpixel-based methods
.
J. Geophys. Res. Atmos.
,
119
,
3401
3419
, doi:.
Peterson
,
D. A.
,
E. J.
Hyer
,
J. R.
Campbell
,
M. D.
Fromm
,
J. W.
Hair
,
C. F.
Butler
, and
M. A.
Fenn
,
2015
:
The 2013 Rim Fire: Implications for predicting extreme fire spread, pyroconvection, and smoke emissions
.
Bull. Amer. Meteor. Soc.
,
96
,
229
247
, doi:.
Peterson
,
D. A.
,
E. J.
Hyer
,
J. R.
Campbell
,
J. E.
Solbrig
, and
M. D.
Fromm
,
2017
:
A conceptual model for development of intense pyroconvection in western North America
.
Mon. Wea. Rev.
, doi:,
in press
.
Potter
,
B. E.
, and
M. A.
Anaya
,
2015
:
A wildfire-relevant climatology of the convective environment of the United States
.
Int. J. Wildland Fire
,
24
,
267
275
, doi:.
Prata
,
A. J.
,
1989
:
Observations of volcanic ash clouds in the 10-12 μm window using AVHRR/2 data
.
Int. J. Remote Sens.
,
10
,
751
761
, doi:.
Prins
,
E. M.
, and
W. P.
Menzel
,
1994
:
Trends in South American biomass burning detected with the GOES visible infrared spin scan radiometer atmospheric sounder from 1983 to 1991
.
J. Geophys. Res.
,
99
,
16 719
16 735
, doi:.
Prins
,
E. M.
,
J. M.
Feltz
,
W. P.
Menzel
, and
D. E.
Ward
,
1998
:
An overview of GOES-8 diurnal fire and smoke results for SCAR-B and 1995 fire season in South America
.
J. Geophys. Res.
,
103
,
31 821
31 835
, doi:.
Reutter
,
P.
, and Coauthors
,
2014
:
3-D model simulations of dynamical and microphysical interactions in pyroconvective clouds under idealized conditions
.
Atmos. Chem. Phys.
,
14
,
7573
7583
, doi:.
Rosenfeld
,
D.
,
M.
Fromm
,
J.
Trentmann
,
G.
Luderer
,
M. O.
Andreae
, and
R.
Servranckx
,
2007
:
The Chisholm firestorm: Observed microstructure, precipitation and lightning activity of a pyro-cumulonimbus
.
Atmos. Chem. Phys.
,
7
,
645
659
, doi:.
Sassen
,
K.
, and
J. R.
Campbell
,
2001
:
A midlatitude cirrus cloud climatology from the facility for atmospheric remote sensing. Part I: Macrophysical and synoptic properties
.
J. Atmos. Sci.
,
58
,
481
496
, doi:.
Schmidt
,
E. O.
,
R. F.
Arduini
,
B. A.
Wielicki
,
R. S.
Stone
, and
S. C.
Tsay
,
1995
:
Considerations for modeling thin cirrus effects via brightness temperature differences
.
J. Appl. Meteor.
,
34
,
447
459
, doi:.
Schmit
,
T. J.
,
E. M.
Prins
,
A. J.
Schreiner
, and
J. J.
Gurka
,
2001
:
Introducing the GOES-M Imager
.
Natl. Wea. Dig.
,
25
,
28
37
.
Schroeder
,
W.
,
I.
Csiszar
,
L.
Giglio
, and
C. C.
Schmidt
,
2010
:
On the use of fire radiative power, area, and temperature estimates to characterize biomass burning via moderate to coarse spatial resolution remote sensing data in the Brazilian Amazon
.
J. Geophys. Res.
,
115
,
D21121
, doi:.
Toon
,
O. B.
, and Coauthors
,
2016
:
Planning, implementation, and scientific goals of the Studies of Emissions and Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC4RS) field mission
.
J. Geophys. Res. Atmos.
,
121
,
4967
5009
, doi:.
Trentmann
,
J.
, and Coauthors
,
2006
:
Modeling of biomass smoke injection into the lower stratosphere by a large forest fire (Part I): Reference simulation
.
Atmos. Chem. Phys.
,
6
,
5247
5260
, doi:.
Wallace
,
J. M.
, and
P. V.
Hobbs
,
2006
: Atmospheric Science: An Introductory Survey. 2nd ed. International Geophysics Series, Vol. 92, Academic Press, 483 pp.
Wooster
,
M. J.
,
G.
Roberts
,
G. L. W.
Perry
, and
Y. J.
Kaufman
,
2005
:
Retrieval of biomass combustion rates and totals from fire radiative power observations: FRP derivation and calibration relationships between biomass consumption and fire radiative energy release
.
J. Geophys. Res.
,
110
,
D24311
, doi:.

Footnotes

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).