An Investigation of Non‐Spherical Smoke Particles Using CATS Lidar

Smoke particles originating from biomass burning events are typically assumed to be spherical, yet non‐spherical smoke particles are also reported from in situ observations. The spatial and temporal distributions of non‐spherical smoke particles, which could have impacts on passive‐ and active‐based satellite aerosol retrievals, are not yet well understood. In this analysis, using NASA's Cloud Aerosol Transport System (CATS) lidar data during the biomass burning season over Africa and South America from 2015 to 2017, we studied the frequency and distribution of non‐spherical smoke particles. A supplemental smoke aerosol typing algorithm was developed to identify aerosol layers containing non‐spherical smoke particles which could otherwise be misclassified as dust or dust mixture using the CATS standard aerosol typing algorithm. Approximately 30% of smoke layers over Africa and South America are non‐spherical (depolarization ratio >0.1) and align with dry biomes of low soil moisture values. Conversely, spherical smoke layers (depolarization ratio <0.1) are in moist regions. The modified algorithm with improved discrimination of non‐spherical smoke detection using CATS depolarization ratio was further verified with the National Oceanic and Atmospheric Administration Hybrid Single‐Particle Lagrangian Integrated Trajectory model, Aerosol Robotic Network Ångström exponent retrievals, and National Centers for Environmental Prediction/National Center for Atmospheric Research Reanalysis soil moisture data. This study highlights the limitations of current aerosol typing algorithms and the potential of algorithms employing ancillary data to improve aerosol typing such as multi‐wavelength volume depolarization ratio measurements or synergy with passive sensors to further discriminate between aerosol types from spaceborne elastic backscatter lidar.

Previous studies show that vegetation properties and fire phase (flaming or smoldering) control fire emissions and the composition of smoke particles and aggregates (Andreae et al., 1998;Cachier et al., 1995;Hobbs, 1996;Martins, Hobbs, et al., 1998;Patterson & McMahon, 1984;Posfai et al., 2003;Turns, 1996).Vegetation productivity and diversity is determined by the regional hydrologic cycle (Ugbaje & Bishop, 2020), which dictates the temperature of fires once the biomass is ignited.High moisture content prolongs the smoldering combustion phase, which occurs as a slow, low temperature, often flameless fire (Rein, 2016), and decreases combustion efficiency of fires (L.W. Chen et al., 2010).Conversely, flaming combustion results in a faster spread rate, higher peak temperature, and higher combustion efficiency (Hadden et al., 2014;Martins, Hobbs, et al., 1998;Ward et al., 1992).Flaming combustion fires produce a higher number of non-spherical particles with increased black carbon content compared to their smoldering combustion counterparts (Martins, Hobbs, et al., 1998).
The assumption that smoke aerosol particles are spherical introduces uncertainties to both satellite-based aerosol retrievals and numerical modeling of smoke aerosol radiative properties (Adachi et al., 2010;Bond et al., 2013).In active lidar-based aerosol retrievals, these non-spherical smoke aerosol layers are subject to misclassification due to threshold-based aerosol discrimination that relies on assumptions of smoke particle sphericity.The frequency of occurrence and spatial distribution of non-spherical smoke aerosols need to be studied for future satellite-and modeling-based aerosol applications and studies.
In this paper, we examine non-spherical smoke aerosol properties in Africa and South America, as these are the largest global sources of biomass burning emissions (Brown et al., 2021), using lidar layer-integrated volume depolarization ratio (VDR) data retrieved from the NASA's Cloud-Aerosol Transport System (CATS) on board the International Space Station (ISS), supported by passive-based soil moisture data from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis1 project, surface-based Aerosol Robotic Network (AERONET) data, surface land type and fire and thermal anomalies data from Visible Infrared Imaging Radiometer Suite (VIIRS) and the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model.In this paper, we define non-spherical smoke as an aerosol layer detected by CATS with a layer-integrated volume depolarization ratio (VDR) greater than 0.10 within an active biomass burning region as verified by AERONET.The relationship between smoke particle sphericity and regional vegetation properties was also investigated.This study presents the first analysis using space-based lidar VDR data to characterize the complicated nature of global fire regimes and is organized as follows: data used in this study is presented in Section 2, followed by case study results and methodology for the classification of smoke aerosol layers and results of the supplemental aerosol typing algorithm in Sections 3 and 4. Finally, Section 5 includes conclusions of the study.

10.1029/2023JD038805
3 of 19 CPL (McGill et al., 2002) and CALIOP (Vaughan et al., 2009).CATS Mode 7.2 Version 3.00 Level 1B (L1B) products such as total attenuated backscatter and layer-integrated VDR are reported at 60 m vertical and 350 m horizontal resolutions, while Level 2 (L2) products are reported at 60 m vertical and 5 km horizontal resolutions (Yorks et al., 2021).In addition to L2 feature detection at 5 km using averaged L1B data, during daytime when solar background contributions are high, L1B data may also undergo additional horizontal averaging (60 km) for feature detection, but final products (i.e., aerosol layer top/base heights, aerosol type, etc.) are still reported at the standard 5 km horizontal resolution (https://cats.gsfc.nasa.gov/data)(NASA, 2019).In addition to L2 feature detection at 5 km using averaged L1B data, during daytime when solar background contributions are high, L1B data may also undergo additional horizontal averaging (60 km) for feature detection, but final products (i.e., aerosol layer top/base heights, aerosol type, etc.) are still reported at the standard 5 km horizontal resolution (https://cats.gsfc.nasa.gov/data).
The layer-integrated VDR is defined as the ratio between the feature-integrated perpendicular attenuated (molecular plus particulate) backscatter to the parallel attenuated backscatter signal.Linearly polarized light is produced naturally in pulsed laser systems, such as those used in the CATS instrument.A beam splitter in the receiver optics system is then used to separate the perpendicular and parallel polarized return signal to enable estimates of the VDR.Knowledge of the relative gain between the perpendicular and parallel channels, which is quantified by the polarization gain ratio (PGR), is needed for an accurate VDR.Further details of the layer-integrated depolarization ratio and the computation of the PGR can be found in Yorks et al. (2016) and Pauly et al. (2019).A large source of uncertainty in the VDR is the random error from solar background noise.During daytime, this uncertainty is compounded by the fact that the solar background noise is higher in both the parallel and perpendicular channels.As such, only nighttime data will be used in this analysis.In addition, CATS is very sensitive at night to faint aerosol layers with aerosol optical depth (AOD) values as low as 0.08, as demonstrated by McGill et al. (2020), ensuring this study is not biased toward optically thick aerosol layers.
The CATS aerosol typing algorithm (Figure 1a; adapted from Nowottnick et al. (2022)) builds on heritage set forth by the CALIOP aerosol typing algorithm but relies on 1064 nm retrievals instead of 532 nm as used by CALIOP.The CATS typing algorithm defines smoke aerosol layers as those with a feature integrated VDR at 1064 nm less than 0.15, feature integrated total attenuated backscatter at 1064 nm greater than 0.0005 sr −1 or sufficient carbonaceous extinction from the Modern-Era Retrospective analysis for Research and Applications, Version  (Gelaro et al., 2017) at the location of the CATS detected aerosol feature (Nowottnick et al., 2022).All of the variables used to differentiate aerosol types in the CATS typing algorithm are derived from CATS data, with the exception of surface type, carbon and sulfate aerosol simulations, and wind information.Variables used in the typing algorithm and their data sources are listed in Table S1.Full details and further discussion of the CATS V3 aerosol typing algorithm are presented in Nowottnick et al. (2022).

Aerosol Robotic Network (AERONET)
The Aerosol Robotic Network (AERONET) of sun/sky photometers directly measures solar radiation to determine aerosol optical, microphysical, and radiative properties throughout the globe (Holben et al., 1998).The extensive network provides long-term, continuous observations for over 500 sites.AOD is measured at eight wavelengths ranging from 340-1,640 nm in addition to sky radiance measurements at 440, 675, 870 and 1,020 nm used for retrieving columnar aerosol optical properties.The spectral dependence of AOD with wavelength (Ångström exponent) is retrieved for a wavelength range between 440-870 nm using 440, 500, 675, and 870 nm AOD.Level 2 Version 3 Ångström exponent values for stations within the analysis domains operating during the CATS orbital period (2015-2017) are analyzed during the biomass burning season (July-October) and can be found in Table S2.Low values of Ångström exponent (e.g., <1.0) indicate large, coarse mode particles such as dust, while high values (e.g., >1.5) are measured in the presence of fine mode aerosol particles, including smoke (Eck et al., 1999;Holben et al., 1991;Kaufman et al., 1992Kaufman et al., , 2002;;Reid et al., 1999).Further, the imaginary part of the refractive index retrieved from the almucantar scan is analyzed for the case studies presented in this study.Schuster et al. (2016) found that carbon based aerosols can be separated from dust using the imaginary refractive index as 95% of biomass burning aerosols had an imaginary refractive index greater than 0.0042 at 675-1,020 nm.It is important to note that the 440 nm AOD must be greater than 0.4 for valid retrievals of the imaginary refractive index (e.g., Giles et al., 2012).

Visible Infrared Imaging Radiometer Suite (VIIRS) Fire and Thermal Anomalies Data
The VIIRS Fire and Thermal Anomalies product shows global active fire detections and thermal anomalies available from the joint NASA/National Oceanic and Atmospheric Administration (NOAA) Suomi National Polar orbiting Partnership satellite since 2012 (Schroeder et al., 2014).This product is built on heritage from Moderate Resolution Imaging Spectroradiometer (MODIS) Fire and Thermal Anomalies retrievals (Giglio et al., 2015), however VIIRS nadir observations are at a 375 m spatial resolution which allows for detection of smaller fires than what is possible from MODIS.Data are collected daily with overpass times similar to that of Aqua MODIS (Li et al., 2018).VIIRS Fire and Thermal Anomalies data flagged as high confidence were analyzed from 2015 to 2017 during the peak biomass burning months in South America and Africa.High confidence data are associated with saturated pixels and are used to confirm the presence of active fires within the study domains and during case study overpasses of the CATS sensor.

MODIS Land Cover Type Map
The MODIS Version 6 Land Cover Type Level 3 product provides global annual maps of land cover at 500 m spatial resolution from supervised classification of reflectance data from both Terra and Aqua satellites (Friedl et al., 2002(Friedl et al., , 2010)).Land Cover Type Collection 6 relies on a hierarchical classification model which includes structured distinctions between land cover properties including vegetation, canopy and ground cover, perennial and annual, leaf type, land use, and wetlands to classify land cover type based on the International Geosphere-Biosphere Programme (IGBP) scheme (Sulla-Menashe et al., 2019).It is important to note, a known issue with this product pertinent to this analysis is the misclassification of some grassland areas as savannas (Sulla-Menashe & Friedl, 2018).

The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) Model
The NOAA Air Resources Laboratory's HYSPLIT analysis tool provides information on air parcel trajectories, including evidence of transport, dispersion, and dissipation (Draxler & Hess, 1998).Both forward and backward trajectories can be calculated based on user-specified spatial and temporal ranges.HYSPLIT has been widely used in identifying plume transport and dispersion for fire events (Brey et al., 2018;H. C. Kim et al., 2020;Stein et al., 2015).In this analysis, the NCEP Global Data Assimilation System one-degree archived model data is used for ensemble analyses of air parcels to track the airmass history and understand regional wind patterns over South America and Africa.Additionally, trajectory frequencies are presented which show the normalized number of times a trajectory passed over a grid cell in the domain.

National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis Data
The NCEP/NCAR Reanalysis 1 project performs data assimilation using past surface, ship, satellite, and aircraft data from 1948 to the present to produce a long term global record of an extensive list of atmospheric variables.Data have temporal and spatial resolutions of 4-times daily, daily, and monthly values on a 2.5° by 2.5° global grid across 17 pressure levels (Kalnay et al., 1996).This study utilizes daily composite mean 700 mb winds, 850 mb winds, surface winds, and 0-10 cm volumetric soil moisture (volume of water per unit volume of soil) data over the analysis domain during the study period.

Observations of Non-Spherical Smoke Aerosol Over Africa
The largest sources of anthropogenic aerosol emissions are open burning of forests and savannas in Africa, contributing ∼52% of total global emissions (Brown et al., 2021).CATS regularly transected the African biomass burning region, providing 1064 nm total attenuated backscatter and VDR measurements of smoke aerosol layers.Figure 2a shows a CATS transect over active fires detected by VIIRS Fire and Thermal Anomalies retrievals on 28 August 2015 with visible smoke plumes stretching throughout southern Africa.The CATS 1064 nm total attenuated backscatter of this overpass (Figure 2b) highlights the extent of the smoke plume through southern Africa.The corresponding 1064 nm VDR curtain plot (Figure 2c) shows the transition from spherical aerosol layers with low values of VDR (<0.1) to non-spherical aerosol layers (depolarization ratios >0.1) as CATS approached the southern portion of the African biomass burning sites is evident.Further, Figure 2d shows the CATS aerosol layer classifications for this overpass following the methods discussed in Nowottnick et al. (2022) and outlined in Figure 1a.Aerosol layers in the northwestern portion of the transect with relatively low layer-integrated VDR values are classified as smoke aerosol (black), however in the southern portion of the overpass, the aerosol typing algorithm classifies aerosol layers as dust mixture (orange) as the layer-integrated VDR values increase to greater than 0.15.
NOAA Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT; Stein et al., 2015) back trajectory ensemble analysis and frequency analysis was applied for two portions of the biomass burning smoke plume from 28 August 2015 (Figure 3).The first back trajectory (Figures 3a and 3c) was initiated at the location of spherical smoke aerosol (17.80°S/17.39°E;blue circle in Figures 2a and 2b) from the mean altitude of the smoke plume (3.5 km) which resulted in an airmass source region in northern Namibia and southern Angola located between 16°S-20°S and 12°E-18°E (Figure 3a), just north of the CATS ground track.The NOAA HYSPLIT frequency trajectory analysis (Figure 3c), which shows the summed frequency that a trajectory passed through each grid cell in the pictured domain, confirms a local source region for the smoke plume observed by CATS.This region of Africa is dominated by biomes of shrublands, grasslands, and savannas as classified by the MODIS Land Type map for August 2015 (Figure 4) with fractional soil moisture values <0.12 (Figure S1a).The closest AERONET station with available data during August 2015 is the Mongu Inn location (15.26°S/23.14°E;#7 in Figure 4; yellow star Figure 2a) located 670 km from the CATS overpass location.A mean Ångström exponent of 1.7, fine mode fraction value of 0.9, and 500 nm aerosol optical depth of 0.5 were reported on 28 August 2015 confirming the presence of accumulation mode aerosol.Imaginary refractive index data are available on 27 August 2015 (1 day prior to the case study date and consistent with the back trajectories) with a mean value of 0.03 for the wavelength range of 675-1020 nm.This value is well within the biomass burning aerosol range of imaginary refractive index values (>0.0042) found by Schuster et al. (2016).A monthly mean Ångström exponent of 1.5 was reported, along with fine more fraction of 0.8, which suggest the presence of accumulation mode aerosol throughout August over this region.
A second back trajectory from the depolarizing portion of the aerosol plume (27.70°S/26.30°E;purple circle in Figures 2a and 2b; classification striping between smoke and dust mixture by the CATS aerosol typing algorithm) was initiated at the mean altitude of the non-spherical plume (1.5 km) and resulted in airmass source regions of savanna and grasslands of northern South Africa and southern Botswana and Zimbabwe centered between 20°S-25°S and 20°E-30° E (Figure 3b).The frequency trajectory analysis (Figure 3d) provides a clearer view of the airmass source regions and shows >90% of the trajectories are from a localized region of grasslands surrounding the CATS overpass point in South Africa, with additional trajectories from grasslands and savannas to the North and East of that location.These results suggest the aerosol plume sampled by CATS is mostly from locally generated fires and not long-range transport of dust (although a complex mixture of aerosols dominated by smoke cannot be ruled out).The Pretoria_CSIR-DPSS AERONET station (25.75°S/28.29°E;#21 in Figure 4; yellow star in Figure 2a) located 290 km from the CATS overpass location reported a mean Ångström exponent of 1.6, mean fine mode fraction of 0.9, and mean imaginary refractive index value of 0.01 for 675-1020 nm, with mean 500 nm aerosol optical depth of 0.5 on 30 August 2015 (no data is available from this location on 28 August 2015).Additionally, for data available from this location in August within ±5 days of the case study date, a mean Ångström exponent of 1.4 and mean fine mode fraction of 0.7 are reported.The Durban_UKZN AERONET station (29.81°S/30.94°E;#23 in Figure 4), located approximately 500 km from the CATS overpass point (marked in Figure 2a), reported a mean Ångström exponent of 1.5, mean fine mode fraction of 0.9, mean imaginary refactive index of 0.02 at 675-1020 nm, and 500 nm aerosol optical depth of 0.5 on 28 August 2015, indicating the presence of smoke aerosol over this region.The mean Ångström exponent for August within ±5 days from the date of the case study was 1.5 with a fine mode fraction of 0.9.These values are consistent for accumulating mode aerosol during the biomass burning season, therefore the information retrieved from AERONET, in combination with fires identified in the VIIRS image (2a), provide confidence that CATS is observing non-spherical smoke aerosols throughout the southern portion of this transect.In this region, the mean volumetric soil moisture values are ∼0.10 (Figure S1a in Supporting Information S1).The soil moisture values are consistent with previous analyses over Africa.Reichle et al. (2017) compiled 2 years of soil moisture active passive root zone soil moisture data (April 2015-March 2017) showing the change in values of soil moisture across global biomes.In Africa, values of 0.3-0.4 are found in the tropical rain forest regions close to the equator and decrease to 0.1-0.2 in grasslands and savanna.In the desert regions along the southwest coast of Africa soil moisture values lower than 0.1 were observed.Burgin et al. (2017) found similar values of soil moisture across different biomes in Africa by retrieving values across a transect spanning from the Saharan desert in the north to the Namibian desert.Their analysis showed a drop in soil moisture values when transecting toward the south, with values >0.3 in the moist savannah decreasing to 0.15 when approaching dry savannah.Across South America, soil moisture values corresponding to individual biomes were similar to those found in Africa (Reichle et al., 2017).Dry regions exhibit more flaming combustion fires which have been found to produce non-spherical smoke particles (Martins, Hobbs, et al., 1998).Additional back trajectory analyses were completed for 3 days before and 3 days after the case study overpass and resulted in the similar source regions for depolarizing and non-depolarizing layers that were found for the case study.Reanalysis data indicate weak easterly surface and westerly 850 mb winds over the region of non-spherical smoke (Figure S2 in Supporting Information S1) which is consistent with the back trajectory analyses.Given the visual confirmation of smoke plumes over southern Africa (Figure 2a), in addition to AERONET retrievals indicating the presence of accumulation mode aerosols, we conclude that the depolarizing aerosol layers identified by CATS in the southern region of this overpass are non-spherical smoke particles that have been misclassified as a dust mixture due to their depolarization ratio values.

Observations of Non-Spherical Smoke Aerosol Over South America
South America produces the second largest global source of anthropogenic emissions from biomass burning (Brown et al., 2021).Therefore, a case study of a CATS transect over active biomass burning fires in South America on 17 September 2015 is presented.Figure 5a shows the VIIRS Fire and Thermal Anomalies over South America with the CATS transect overlaid.A curtain plot of 1064 nm total attenuated backscatter from the overpass is shown in Figure 5b with the corresponding 1064 nm depolarization ratio for this scene (Figure 5c).A ∼4 km thick aerosol plume stretches throughout the figures which is classified predominantly as smoke (Figure 5d; black) and dust mixture (Figure 5d; orange) based on the layer depolarization ratio values.
Figure 6a shows a back trajectory ensemble analysis for the biomass burning smoke plume from 17 September 2015.Note that the aerosol plume exhibits a high frequency of interchanging aerosol classification between smoke and dust mixture.The back trajectory was initiated from the center of this overpass scene (18.41°S/56.96°W;blue circle in Figures 5a and 5b; black star Figure 6a) at the mean altitude of the plume (2 km) and showed a source region in eastern Brazil centered around 10°S/45°W.This region is classified as savannas and grasslands (Figure 7) with volumetric mean soil moisture values ∼0.10 (Figure S1b in Supporting Information S1).A NOAA HYSPLIT trajectory frequency analysis (Figure 6b) shows the summed frequency of trajectories passing through each grid cell in the pictured domain.The back trajectory results initiated from the mean altitude of the plume at the CATS overpass location (Figure 6b black star) show a high frequency (>90%) of trajectories pass through grid cells just north  S2) numbered.The AERONET stations reported in the case study analysis are marked in red numbering.of the CATS observation location and suggest the plume sampled by CATS did not undergo long-range transport and is local burning of dry savanna producing non-spherical smoke particles.Multiple AERONET stations were operational during this CATS overpass.The Santa_Cruz_UTEPSA AERONET station (17.76°S/63.21°W;#10 in Figure 7), located 660 km from the CATS observed layer, reported a mean Ångström exponent of 1.7 and fine mode fraction value of 0.9 during this day.The CUIABA-MIRANDA station (15.73°S/56.07°W;#11 in Figure 7) was the closest operational AERONET station during the case study (300 km from the CATS overpass point) and reported a mean Ångström exponent of 1.8, fine mode fraction value of 0.9, and imaginary refractive index averaged over 675-1020 nm of 0.01 on 17 September 2015.Additionally, the Campo_Grande_SONDA station (20.43°S/54.53°W;#13 in Figure 7), located 330 km from the CATS observation point, reported a mean Ångström exponent of 1.6, fine mode fraction value of 0.8, and mean imaginary refractive index value of 0.01 from 675 to 1020 nm during the case study overpass.Together, these large Ångström exponent and fine mode fraction values, bolstered by the smoke plumes visible in VIIRS imagery (Figure 5a) and value of imaginary refractive index greater than 0.0042 (Schuster et al., 2016), confirm the presence of smoke aerosols near the location of the CATS overpass.Additionally, the back trajectory results align with the surface and 850 mb easterly wind patterns over the back trajectory initiation location during this day (Figure S3 in Supporting Information S1) which would usher in air from the dry Cerrado region of eastern Brazil to our case study location.

Revised CATS Aerosol Typing Algorithm
Given the results of the case studies, a supplemental smoke typing algorithm has been developed in this analysis to identify smoke layers which may have been misclassified as a dust-type layer based on depolarization ratio thresholds for aerosol typing (Figure 1b).The smoke typing algorithm developed in this paper is targeted to only  S2) numbered.The AERONET stations reported in the case study analysis are marked in red numbering.
biomass burning regions (Africa and South America) over known biomass burning seasons and is not meant for operational use over the entire globe.
As the first step, aerosol layers that are identified as smoke by the standard typing algorithm as described in Section 2 (Figure 1a) are included as smoke aerosols.As the second step, for CATS identified non-smoke aerosols over the study domain during the study period, potential candidates for either spherical or non-spherical smoke aerosol plumes are selected if the physical aerosol layer thickness (vertical distance between layer base and layer height) is above 0.5 km.The 0.5 km threshold is implemented based on performance of the CATS operational aerosol typing algorithm (Nowottnick et al., 2022) to ensure that marine aerosol layers and polluted continental aerosol layers are not included in this analysis.For smoke layers as identified from the second step, aerosol layers with VDR values less than 0.1 are categorized as spherical smoke and aerosol layers with VDR values larger than 0.1 are labeled as potentially non-spherical smoke aerosols.The VDR threshold of 0.1 was selected as this threshold value was found to capture spherical liquid water particles (Yorks et al., 2011) and is therefore assumed to also separate spherical smoke particles from non-spherical smoke aerosol particles.As the last step, for non-spherical aerosol layers as identified from step 3, daily mean surface wind speed and wind direction from the NCEP/NCAR Reanalysis over the known dust source regions as described in the previous section are used to further exclude potential dust contamination.Days with westerly or southwesterly mean surface winds greater than 2 m/s from the Namib or Kalahari Deserts are excluded from the analysis.
To validate the developed algorithm, 40 cases of identified non-spherical smoke aerosols (based on CATS VDR) over Africa were selected for HYSPLIT back trajectory analyses to confirm the non-spherical smoke aerosol was not from a major dust source region.A constraint was placed on the data to only include cases which are ≤250 km from an AERONET station, and for qualified CATS observations, a random selection was further applied for selecting the 40 cases as mentioned.Each trajectory was initiated from the mean altitude of the smoke layer and was completed for 24 hr to trace the path of the air parcel.Table S3 lists starting locations and dates for each of the cases, in addition to their source regions and VDR values.AERONET 500 nm AOD (AOD 500 ) and Ångström exponent values are also listed for the closest available station to the CATS overpass point.The majority of samples came from a source region of savannas and grasslands, while two cases (5% of total cases) came from shrubland regions.No cases were found to originate from land classified as barren by the MODIS Land Cover Type, which is dominated by sand (Sulla-Menashe & Friedl, 2018).Most (∼70%) of the cases were classified as dust mixture by the standard CATS aerosol typing algorithm, while ∼30% were classified as dust due to their large values of VDR.We also verified aerosol properties from the selected cases for aerosol plumes with AOD 500 values above background aerosols (assumed to be AOD 500 of less than 0.2 for this study).The mean Ångström exponent value for cases with AERONET AOD 500 ≥0.2 is 1.5, while the mean Ångström exponent for cases with AERONET AOD 500 ≥0.3 is 1.6.Additionally, 75% of the cases with AERONET AOD 500 ≥0.2 have an Ångströsm exponent value ≥1.5 and 85% of cases with AOD 500 ≥0.3 have Ångström exponent value ≥1.5.This result shows that fine-mode aerosol dominates during those heavy aerosol load cases, yet are incorrectly classified as dust-like type aerosols from the CATS aerosol typing algorithm.This supplemental algorithm is applied during the peak burning season over the major global biomass burning regions of southern Africa and South America.By implementing this supplemental aerosol typing algorithm, non-spherical smoke aerosol layers, which may have been misclassified as a dust-type aerosol because of their increased depolarization ratio values, are classified as smoke and analyzed in this study.

Regional Analysis Over Africa
Following the results of the case study analyses over Africa and South America, the supplemental smoke aerosol typing algorithm was applied on a larger scale to investigate the frequency and spatial distribution of non-spherical smoke aerosol layers during the biomass burning season.In central and southern Africa, 28%-31% of smoke layers observed by CATS during July-October 2015-2017 were made up of non-spherical smoke aerosols having layer-integrated VDR values >0.10 with a mean VDR of 0.16.Additionally, 6%-7% of all smoke layers were found to have a layer-integrated VDR >0.20 with a mean VDR value of 0.26.These results suggest that ∼28-31% of smoke layers in central and southern Africa are non-spherical particles.Further, the results of this analysis found that 13%-15% of depolarizing smoke aerosol layers over southern Africa are subject to misclassification as a dust-type layer due to threshold-based aerosol discrimination that rely on assumptions of smoke particle sphericity.The range presented here captures different atmospheric scenes.The lower bounds is for clear-skies with AOD >0.05, while the upper bounds is for all-sky scenes.
The spatial distribution of the results of the supplemental smoke aerosol typing algorithm applied over central and southern Africa during the peak African biomass burning months of July-October 2015, 2016, and 2017 is shown in Figure 8.The highest concentration of regularly shaped smoke aerosol layers (VDR <0.10) is centered around 10°S in central Africa (Figure 8a), which is predominantly savannas and patches of evergreen and deciduous forests (Figure 4).There is also a high concentration of spherical smoke aerosol layers off the east coast of southern Africa centered around 35°E.The eastern coast of the continent is predominantly deciduous forest and savanna biomes which contribute to the spherical layers observed.Additionally, biomass burning in Madagascar is likely contributing to these observations as both surface and 700 mb winds indicate easterly patterns over this region during the biomass burning season (Figures 9c and 9d) and previous findings have shown smoke aerosols to become more spherical with aging due to uptake of water (Kar et al., 2018).Reanalysis soil moisture data indicate values between 0.16-0.3 in regions with spherical smoke aerosol layers (Figure 9a).
Conversely, the non-spherical smoke aerosol layers with layer-integrated VDR values >0.10 are concentrated south of 20° S (Figure 8b) in dry grasslands (Figure 4) where soil moisture values are lower than in forest areas of central Africa (Figure 9a).There is also a high concentration of non-spherical smoke aerosols south of the Equator between 30° and 40°E.This region is classified as grasslands, consistent with the biome of non-spherical smoke particles to the South.Soil moisture values, shown in Figure 9a, reach a maximum of 0.14 in regions of non-spherical smoke aerosol layers.Figure 9b shows VIIRS detected fires averaged over the burning seasons from 2015-2017 throughout southern Africa (red box in Figure 9a).Note that only this subset of Africa is shown because fire counts north of this region are at least one order of magnitude greater than over southern Africa.The highest frequency of fires occurs in northern Botswana and northeastern Namibia, in addition to the eastern coast of South Africa where fire counts reach 80 per day.These regions correspond to areas of non-spherical smoke layers observed by CATS during the same time (Figure 8b) attributed to the burning of dry material.It is important to note that previous studies have found smoke physical characteristics to change due to aging (e.g., Ansmann et al., 2021;Kar et al., 2018;Ohneiser et al., 2021;Saide et al., 2022), however Figure 9 suggests that CATS smoke aerosol layers observed are of fresh smoke from locally generated fires.To confirm that smoke aerosols, not desert dust particles, are the predominant aerosol type in these regions of high VDRs, we investigate the Ångström exponent values from AERONET locations within these regions.Although an attempt was made to only use data from stations that were operational for all 3 years of the CATS sampling period, some stations were only operational for a subset of the study period.Therefore, as long as a station was operational for at least one biomass burning season sampled by CATS, it was included in this analysis.Table S2 in the Supplement lists the mean values of 440-870 nm Ångström exponent and 500 nm AOD for each of the locations included in  These findings indicate that non-spherical smoke layers originate from a region of dry vegetation with lower soil moisture values, while spherical layers are from a moist burning region.Oxygen supply is limited during flaming combustion which results in regions of the flame that are smothered before oxidation of carbon radicals is complete (Andreae et al., 1998).This process results in more prevalent observations of non-spherical soot particles in flaming combustion fires than smoldering combustion fires (Patterson & McMahon, 1984).Additionally, these results agree with previous studies which found that burning material controlled the combustion efficiency and, therefore, smoke particle composition and shape (e.g., Martins, Artaxo, et al., 1998;Posfai et al., 2003).

Regional Analysis Over South America
The peak biomass burning season in South America occurs from September-October annually.Therefore, the supplemental smoke aerosol classification scheme described above, and depicted in Figure 2b, was applied over South America during the burning season throughout the CATS operational period (2015-2017) to determine the frequency and spatial variability of non-spherical smoke aerosol layers.The results of this analysis show that 69%-73% of smoke aerosol layers have layer-integrated VDR values <0.10 with a mean value of 0.05.Conversely, 27%-31% of the smoke aerosol layers over South America were non-spherical, having layer-integrated VDR values >0.10 with a mean VDR of 0.16.Furthermore, 8%-10% of smoke layers had layer-integrated VDRs >0.20, with a mean VDR of 0.24.Based on the threshold-based aerosol layer classification algorithm used by CATS, 13%-16% of smoke aerosol layers in South America are subject to aerosol type misclassification as dust or dust mixture as their layer-integrated VDR values are above 0.15.
The spatial distribution of spherical smoke identified by CATS is shown in Figure 10a.These smoke layers are heavily concentrated in the Brazilian Amazon north of 10°S, a region with a dominant vegetation type of evergreen broadleaf forests (Figure 7) with weak surface winds and <10 m/s 700-mb easterly winds (Figures 11c  and 11d).This region is dominated by smoldering combustion (Darbyshire et al., 2019) and typically has high humidity which makes it difficult to ignite naturally (Pivello, 2011).In contrast, the non-spherical smoke particles were concentrated south of 10°S in the Brazilian Cerrado (Brazilian savanna) and in northern Argentina (Figure 10b).These regions are primarily categorized as savannas and grasslands which have lower soil moisture values than their northern counterparts, as is seen in Figure 11a (<0.20 vs. >0.30),and have a high frequency of VIIRS detected fires (>150 per day) as highlighted by Figure 11b.These results also agree with those of Darbyshire et al. (2019) in which the Amazon basin, which is comprised of both the moist Brazilian Amazon and the dry Cerrado, was found to undergo distinct fire regimes according to biome.The amount of water in the biomass has been cited as a major control factor in determining if the fire will undergo flaming or smoldering combustion (Freitas et al., 2007), which is known to determine the smoke particle shape (Martins, Hobbs, et al., 1998).During the peak South American biomass burning months, the Brazilian Cerrado and dry Chaco region of northern Argentina experience dry season in which the grasses and woodlands become very dry and flammable, and thus account for a majority of the continent's woody vegetation loss through flaming combustion (Aide et al., 2013;Darbyshire et al., 2019;Ramos-Neto & Pivello, 2000).Conversely, smoke originating from the moist Amazon was found to originate from smoldering combustion (Ward et al., 1992).
AERONET derived Ångström exponent values were analyzed over South America during the CATS operational period and are listed in Table S2 (Supplement).Locations where very few smoke layers were observed by CATS are marked with an asterisk (*) and have an overall mean Ångström exponent value of 1.1 for all months listed in the table.Regions where CATS observed smoke aerosols have a mean Ångström exponent of 1.4 during all months listed, and an increased value of 1.6 during the peak biomass burning months over South America (September-October) with high aerosol load (AOD ≥0.3).This result shows that during the biomass burning season, aerosol load is dominated by fine mode aerosols.

Conclusions
As a preliminary study, the frequency and spatial distribution of spherical and non-spherical smoke aerosol layers across South America and Africa were analyzed during peak biomass burning months using space-based nighttime lidar VDR.The CATS standard aerosol typing algorithm was supplemented to include smoke aerosol layers with VDR values >0.15 which may have been misclassified as dust or dust mixture using standard depolarization ratio threshold-based aerosol typing.The results of this analysis show that approximately 30% of smoke aerosol layers in all-sky scenes over South America and Africa are non-spherical smoke particles and approximately 15% of smoke layers are subject to misclassification based on depolarization ratio thresholds.Limiting the analysis to clear-sky scenes with an AOD >0.05, when CATS depolarization ratio estimates are most accurate, only reduced these frequency by 1%-3%.
These results show that biomes near the equator and/or biomes with higher soil moisture result in spherical smoke aerosol layers linked to smoldering combustion processes.In contrast, drier regions, which are known to exhibit flaming combustion of biomass, produce smoke layers with larger VDR values.The results presented are, to the authors' knowledge, the first to showcase non-spherical smoke layers from space-based lidar and to link the dependence of depolarization ratio to soil moisture.These results also suggest the frequency of non-spherical smoke particles is greater than reported in previous literature and should be considered for its impacts on modeling retrievals and active sensor-based aerosol classification.While our results give us confidence that the smoke typing algorithm presented in this paper is properly identifying spherical and non-spherical smoke particles in the biomass burning regions of Africa and South America, more work must be done to determine how to incorporate such detection into operational space-based and airborne lidar algorithms.For example, coincident in situ and remote sensing measurements of these aerosol layers are needed to further examine physical properties of those "erroneously" detected dust-like aerosols from lidar observations.Previous studies have found dust particulate depolarization ratios to be relatively spectrally flat, while smoke exhibits a spectral dependence between 1064 and 532 nm (Burton et al., 2015), underscoring the need for multi-wavelength VDR measurements.Additional information content to aid in aerosol typing using spaceborne elastic backscatter lidar, such as synergy with passive sensors, is needed particularly for discriminating between non-spherical smoke and dust which has downstream impacts on extinction retrievals used to determine aerosol radiative impacts.

Figure 1 .
Figure 1.A schematic of the Cloud Aerosol Transport System standard (a) and supplemental (b) aerosol typing algorithms.

Figure 2 .
Figure 2. Visible Infrared Imaging Radiometer Suite high confidence fire and thermal anomalies imagery on 28 August 2015 with the Cloud Aerosol Transport System (CATS) overpass of southern Africa overlaid (a).The closest available Aerosol Robotic Network stations utilizing in this case study date are also marked (yellow stars).The CATS 1064 nm total attenuated backscatter curtain plot for the highlighted overpass (b) showing an aerosol plume stretching from the surface to ∼6 km, along with the associated 1064 nm volume depolarization ratio (c) and layer type (d) for this scene.Two points along the CATS flight track are indicated using colored circles (a) and correspond to the locations marked in the backscatter curtain plot (b).

Figure 3 .
Figure 3. National Oceanic and Atmospheric Administration Hybrid Single-Particle Lagrangian Integrated Trajectory model ensemble back trajectory analyses and frequency back trajectory analyses initiated from the mean altitude of the spherical (a), (c) and non-spherical (b), (d) smoke portions of the Cloud Aerosol Transport System observed plume on 28 August 2015 over southern Africa (black stars).The spherical smoke (a), (c) shows a localized origin of Namibia (frequency >90%) and southern Angola (frequency >50%), while the non-spherical smoke (b), (d) shows the highest frequency of trajectories from South Africa (>70%) and residual trajectories from southern Botswana and Zimbabwe (>30%).

Figure 4 .
Figure 4.A Moderate Resolution Imaging Spectroradiometer Land Cover Type map of Africa with locations of Aerosol Robotic Network (AERONET) stations (listed in TableS2) numbered.The AERONET stations reported in the case study analysis are marked in red numbering.

Figure 5 .Figure 6 .
Figure 5. Visible Infrared Imaging Radiometer Suite high confidence fire and thermal anomalies imagery on 17 September 2015 with the Cloud Aerosol Transport System (CATS) overpass of South America overlaid (a).The closest available Aerosol Robotic Network stations utilizing in this case study date are also marked.The CATS 1064 nm total attenuated backscatter curtain plot for the highlighted overpass (b) showing an extensive aerosol plume stretching from the surface to ∼4 km, along with the associated 1064 nm volume depolarization ratio (c) and layer type (d) for this scene.Origin point of Hybrid Single-Particle Lagrangian Integrated Trajectory model back trajectory is indicated by bue dot on panels (a) and (b)

Figure 7 .
Figure 7.A Moderate Resolution Imaging Spectroradiometer Land Cover Type map of South America with locations of Aerosol Robotic Network (AERONET) stations (listed in TableS2) numbered.The AERONET stations reported in the case study analysis are marked in red numbering.

Figure 8 .
Figure 8.The spatial distribution of spherical (a) and irregularly shaped (b) smoke aerosol layers identified by Cloud Aerosol Transport System (CATS) over Africa during the biomass burning season (2015-2017) using the supplemental CATS aerosol typing algorithm.Spherical smoke layers (a) are centered closer to the equator while non-spherical smoke layers (b) are located in the southern portion of the continent.

Figure 9 .
Figure 9. National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis 1 soil moisture values over Africa during the 2016 burning season (a; red box indicated domain shown in panel (b) and daily Visible Infrared Imaging Radiometer Suite high confidence detected fires throughout the burning season over southern Africa during the Cloud Aerosol Transport System (CATS) operational period (b).Low values of soil moisture align with observations of non-spherical smoke layers identified by CATS, while higher values are regions of CATS observed spherical smoke layers.Highest fire counts are present along the eastern coast of South Africa and northern Botswana and Namibia, while fewer fires are observed in the southern tip of South Africa.NCEP/NCAR Reanalysis 1 mean surface (c) and 700 mb (d) winds during the 2016 burning season indicate no large-scale transport of desert dust into the analysis domain.

Figure 10 .
Figure 10.The spatial distribution of spherical (a) and non-spherical (b) smoke aerosol layers identified by Cloud Aerosol Transport System (CATS) over South America during the biomass burning season (2015-2017) using the supplemental CATS aerosol typing algorithm.Spherical smoke layers (a) are centered in northern Brazil while non-spherical smoke layers (b) are located in eastern Brazil and northern Argentina.

Figure 11 .
Figure 11.National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis 1 soil moisture values over South America during the 2016 burning season (a) and daily Visible Infrared Imaging Radiometer Suite high confidence detected fires throughout the burning season over South America during the Cloud Aerosol Transport System (CATS) operational period (b).Low values of soil moisture align with observations of non-spherical smoke layers identified by CATS, while higher values are regions of CATS observed spherical smoke layers.Highest fire counts are observed over Brazil and Bolivia, where average daily fires are >150.NCEP/NCAR Reanalysis 1 mean surface (c) and 700 mb (d) winds during the 2016 burning season indicate no transport of desert dust into the analysis domain where observations of non-spherical smoke were frequent.