Survey of Transverse Range Fire Scars in 10 Years of UAVSAR Polarimetry

Abstract Because cross‐polarized radar returns are highly associated with volume scatter, radar polarimetry returns tend to show strong evidence of wildfire scars and recovery in forest and chaparral. We focus on the polarimetry images from UAVSAR (PolSAR) line SanAnd_08525, which covers a roughly 20 km wide swath over the Transverse Range including parts of the Santa Monica, San Gabriel and San Bernardino Mountains. We select images from four acquisition dates from October 2009 to September 2020, very roughly 4 years apart. These are compared to fire perimeters from the national Geospatial Multi‐Agency Coordination and NIFC databases for years 2003–2020, which shows the areas affected by the major fires (west to east) Springs2013, Woolsey2018, Topanga2005, LaTuna2017, Station2009, BlueCut2016, Pilot2016, Slide2007, Butler2007, and many smaller fires. PolSAR images are shown to be helpful in identifying types and boundaries of fire, 50‐meter scale details of vegetation loss, and variability of vegetation recovery in post‐fire years.

• The urban-facing part of the Transverse Range is imaged by 27 Uninhabited Aerial Vehicle Synthetic Aperture Radar visits from 2009 to 2020, including locations of over 20 wildfires • PolSAR images identify types and boundaries of fire, 50 m scale details of vegetation loss, and variability of post-fire recovery • Observed signs of biomass recovery in gap areas between recent fires provide clues to the locations and extents of near-future fires commonly brushy or leafy biomass (Flores-Anderson et al., 2019). Fires remove some or all of the material that creates volume scatter, so it is instructive to compare fire perimeters with non-green areas surrounded by dominantly green areas, particularly if we can identify a change from a greener area in pre-fire interferograms.
Therefore, PolSAR products are valuable for distinguishing fire types and estimating biomass loss. There is also potential for using a post-fire sequence of these products to evaluate the recovery of the fire area through forest succession (or recovery in non-forested natural areas, such as chapparal). A basic description of the process may be found in (Horn, 1975). There the recovery process is described as a regeneration of an environment similar to the pre-fire environment, through a succession of stages characterized by varying mixtures of trees and plants; the plant varieties and ecological balance of one stage is replaced by a new balance and mix of species in the next.
NASA airborne synthetic aperture radar polarimetry dates back several decades, and among the first reported merits is the identification of HV cross-polarized returns with volume scatter for discriminating vegetation features (Evans et al., 1986) The first study of fire scars with UAVSAR products describes PolSAR image changes in successive visits bracketing the 2014 Colby fire (Glasscoe et al., 2016). There it is pointed out that because radar imaging penetrates clouds and smoke, potential rapid products should become an important complimentary source of information for fire management along with current use of Landsat and other optical mapping products. This potential is demonstrated therein by display of the PolSAR map and fire perimeter for January 17, 2014, during the active phase of the Colby Fire. Another feature of radar images is that they reflect volume scatter, not the presenting surface imaged by optical means.
An integrated approach to the multihazard of wildfire followed by rain-induced debris flow is demonstrated (Donnellan et al., 2018) for the 2017 Thomas fire and subsequent 2018 debris flow causing widespread damage in the community of Montecito, California and severing United States Highway 101, an essential traffic corridor. The georeferenced information is provided by satellite optical images, the PolSAR product and repeat-pass interferogram products. The polarimetry product provided a reliable image of the burn area, while the repeat-pass coherence and unwrapped interferogram contributed critical information for a holistic assessment of the fire and debris flow.
In this work, we compare the PolSAR product for a fixed UAVSAR flight line to GeoMAC-archived fire perimeters that intersect this product for a variety of times from 2009 to 2018. For selected cases, we compare the UAVSAR image to the corresponding Google Earth optical image to show the extent that UAVSAR-determined areas of fire damage match the optical evidence.

Polarimetry Data Sources
PolSAR images are openly shared through the UAVSAR Data Search page https://uavsar.jpl.nasa.gov/cgibin/data.pl which allows search for products by flight line number. Product pages list binary PolSAR products including metadata, binary slant range products, binary orthorectified products, a combined KMZ orthorectified image, and ancillary products defining the geometry and topography of the imaging. The links on the UAVSAR product page provide access to the data products which are stored at the Alaska Satellite Facility, and require users to register and log in. PolSAR products for flight line 08525 are available for flight dates listed in page link titled "High Resolution KMZ file" at product pages listed in Table 2. In Table 3, we list the selected PolSAR products used in this work and associate lists of substantial fires in flight-prior years.

Fire Perimeters
GeoMAC fire perimeters intersecting the PolSAR image from flight line 08525 are presented overlaying the PolSAR image from September 18, 2020 in Figure 1. Perimeters are obtained from the GeoMAC and NIFC archives, and their shape files are converted to KML using the public-domain GDAL translator library utility ogr2ogr (GDAL/OGR contributors, 2021). Figure Figure 5 shows the September 18, 2020 PolSAR image with immediately prior fires from 2016 to 2020 outlined in magenta and earlier fires in muted colors.

Discussion
PolSAR images from the UAVSAR airborne radar show many features indicating leafy biomass density along with some radar artifacts. Water, highways, and other smooth level surfaces tend to be displayed as black, because the radar is illuminating downward and to the side, so that horizonal smooth high-dielectric-constant surfaces reflect the radar signal away from the receiver in the aircraft. Urban areas are highly spotty in shades of pink and magenta, combining all three polarization signal colors but with red predominating. This is consistent with the red color assigned for HH polarization, which is the component most enhanced in areas where vertical structures predominate. The southernmost part of the mountain ranges contains patches and streaks of bright colors including white. These areas contain steep south-facing slopes ideal for radar specular reflection back to the UAVSAR that is (in line 08525) sending radar signals to its north side, from the south. Therefore, exceptionally high radar returns are expected in these areas. Generally, topography results in overall signal brightness variation, such that south-facing slopes are brighter (whether green or green-depleted) than north-facing slopes; in the most rugged terrain, the north facing slopes are in radar shadow, and so return no useful signal. Aside from these features, green indicates an abundance of leafy biomass because the sparse arrangement of moisture-containing biomass displays volume scatter, associated with strong HV cross-polarized return (which the PolSAR images color green) at L PARKER ET AL.   band. Biomass-depleted regions are usually purple, a combination of the HH (red) and VV (blue) returns. Manual spot checks of isolated slivers of preserved biomass (confirmed with Google Earth images) indicate that the spatial resolution is <50 m.
Figures 2-5 demonstrate that each PolSAR kmz image product is directly useful for indicating presence of biomass, depletion of biomass after fires, and recovery of biomass over time. More complete characterization of biomass changes is possible by creating differences of matching polarimetry channels for successive visits, which would require resampling one of the images to match the georeferenced pixels of the other. Most UAVSAR strips have an overlapping mapped strip from the conjugate direction, and neighboring lines usually have some area of overlap as well; combining these optimally could result in equalizing some of the variations of brightness response mentioned above, and fill in some of the image missing from high slopes facing away from one of the radar observations. It is also likely that combinations of channels (perhaps a ratio of HV with HH + VV) produce better discriminants than the image colors used in this work. Because we assume the evidence of fires and multi-year recovery are large signals and should usually predominate, we have not looked for seasonal or post-rainfall effects, which certainly cause further variations in the biomass response. We have not exploited the UAVSAR binary data image file products, which would provide expanded options for optimal resolution. These options are left for future work; this work serves to show the immediate utility of the PolSAR kmz image products, which are smaller than the binary files and so easier to download and view. These kmz image products have usefully high resolution.
Mentioned above are means of fruitfully combining PolSAR maps with repeat-pass UAVSAR products and optical images for assessment of damage and risk of debris flows (Donnellan et al., 2018). Public web-based tools for combining diverse mapped data types have great potential for research, disaster management and public awareness (Glasscoe et al., 2015;Donnellan et al., 2021).
It is well known that successive fires in southern California tend to create a mosaic of interlocking pieces, and therefore a patchwork of biomass maturity (e.g., Minnich, 1983); less established is the relation of suppression strategy (including controlled burns) on the severity of fires in this mosaic (e.g., Foster et al., 2017). This pattern is evident in Figure 1   cycle in most of the fires outlined by the fire perimeters. In the first interferogram shown, from 2009, we see areas in deep green, which are subsequently overtaken by fires in the next years; other areas in deep green (including a large region north of Glendora around Mount San Antonio, and the area north of San Gorgonio Mountain and around Big Bear Lake) that have had no fires in at least 18 years. These may be due for fires in the coming years or may have advantages of terrain, fire-resistant vegetation, or high seasonal moisture that hinder large fires. We note alpine areas more purple than green (more parallel-polarized return than cross-polarized) that correspond to perimeters of fires from 2003 to the image time in 2009. We note below two cases of pre-2003 fires that also contribute to the green-depleted areas in the San Gabriel Canyon area, but we have not attempted to find complete data before the GeoMAC files from 2003. GeoMAC perimeters are available in the span 2000-2002 but are in an obsolete database format which we have not attempted to map.
The perimeter history shown in Figures 1-5 generally conforms to the mosaic of patches pattern: new fires burn right up to the boundary of older fires, and there stop, presumably for lack of fuel due to an incomplete recovery. Examination of the figures suggests that this protective factor lasts for at least two decades, but there are exceptions: note from Figure  image in Figure 4 is missing that western portion of the image. Details of moisture history, winds during the later fires, and partial recovery of forest succession species may explain these two exceptions, but these are beyond the scope of this work.
Complementing the history of fire intensity and advantages (or not) of particular areas due to such factors as climate, soil type, and biomass flammability, the tendency of a sequence of fires to form a mosaic pattern stresses the importance of gaps between recent fires. The PolSAR imaging of these gap areas supplies clues for the location and extent of future fires. Figures 1 and 5 show large gaps colored in biomass-rich green color in large areas, notably in the Santa Monica mountains west of Thousand Oaks and northeast of Pacific Palisades, the San Gabriel Mountains between San Dimas and Wrightwood, and in the eastern San Bernardino mountains between Forest Falls and Big Bear Lake. While there may be fire-retarding factors in these areas, it is possible they are due for large-area fires in the near future.
The degree of biomass depletion varies considerably from fire to fire, and within individual fire domains. The Colby Fire 2014 in Figure 4c and the Pilot Fire 2016 in Figure 5d are examples of regions where fires removed nearly all of the leafy biomass. From Figures 5b and 5c, note that the Bobcat Fire 2020 domain displays much more deep green in the southern part of the range where it abuts the cities Monrovia and Duarte than it does farther north: we judge the fire depleted far more of the biomass in the remote regions north of the West Fork San Gabriel River than in the more accessible front range.
Careful examination of the green to non-green indicator of biomass loss for many fires shows minor deviations of the burned regions compared to the final GeoMAC perimeters. See, for example, the western boundary of the Cabin Fire 2014 in Figure 4c, and the area between the Morris Fire 2009, the Fish Fire 2006, and the eastern Bobcat Fire 2020, in Figure 5c. We deduce that the biomass depletion in the latter case (compare to Figure 4c)   managing fires in progress or recently contained, deviations are not surprising. We suggest that when Pol-SAR images are available, fire perimeters may be drawn with greater accuracy, particularly in areas remote from road access. PolSAR images render competent images of biomass depletion through cloud and smoke during fires as demonstrated for the Colby Fire 2014 (Glasscoe et al., 2016). Processing and quality control of images took weeks for the UAVSAR flight that imaged the Colby Fire in progress, a latency too great for real-time fire support. But latency will be reduced as airborne radar missions reach the maturity of standard operations, so support for fire suppression is a viable goal in the near future.
The NASA Disasters Program recently activated in response to the Bobcat Fire and UAVSAR data were collected in support of California response efforts. The Disasters program provided these data to California agencies including: The California National Guard, the California Geological Survey, the California Office of Emergency Services, and CalFIRE. The data were also provided to federal agencies including the US Geological Survey and FEMA Region IX. In particular, the California National Guard used these data to identify burn severity and location of burn scars. They also added the data to a web map that was shared with the Joint Operations Center. The State Guard used the data for fire assessment during the event and will in the also use it for post fire debris flow assessment. Their feedback was quite positive, though they did identify the need for more training on the use of the data, what resolutions are available, how the data are interpreted, data latency, overlap of previous fires, and other related agencies with which to coordinate (Phil Beilin, personal communication).
We present two case studies in more detail below, one in the San Bernardino Mountains and one in the San Gabriel range.
PARKER ET AL.

Case Study: Old Fire, 2003 (Northern Part)
The Old Fire of 2003, named for the Old Waterman Canyon Road above San Bernardino where the blaze originated, burned a largely forested area to the east of Lake Arrowhead, California. The alpine forest is diverse, characterized by firs, pines, and oaks. It contains numerous youth camps and residential structures. The range covered by this image is moderately rugged, cut with the deep ravines of Deep Creek, Little Bear Creek, and Hook Creek and much of it is above 1.5 km elevation. The Old Fire was contained by November 2, 2003, while we are using PolSAR images beginning February -2009, after 64 months of biomass recovery. Details of that recovery are evident in Figure 6, which are enlarged areas taken from Figures 2d, 3d, 4d, and 5d. By 2009 the recovery is far from complete, since the area inside the fire perimeter of 2003 is still generally less green than the surrounding terrain, and we have noted that green indicates the HV return from multiple scattering, which here may be easily associated with the density of the fir-pine forest and undergrowth. It is also notable that the coloration of the Old Fire territory is uneven: the southern part is greener than the north and has less contrast than the north with the unburned area to the west, north and south of Lake Arrowhead. Riparian areas around the major waterways (Deep, Little Bear, and Hook creeks) are also greener than their surroundings. Gradual recovery is evident in Figures 6b and 6d, but even the latest image (September 18, 2020) does not show full recovery, and the fire zone coloration of the 2016 image is ambiguous.

Case Study: Pre-2003 Fires, Williams 2012, Cabin 2014, and Eastern Bobcat 2020
The images in Figure

Conclusions
We show strong evidence of wildfire scars and recovery in forest and chaparral. Using the PolSAR images from UAVSAR line SanAnd_08525, we discern patterns of biomass variability due to wildfires and recovery in the Santa Monica, San Gabriel and San Bernardino Mountains. Systematic patterns in time and space are documented in images from four acquisition dates from October 2009 to September 2020, very roughly 4 years apart. These are compared to fire perimeters from the national GeoMAC database from years PARKER ET AL.
10.1029/2021EA001644 9 of 11 2003-2018, which shows the areas affected by the major fires (west to east) Springs2013, Woolsey2018, Topanga2005, LaTuna2017, Station2009, Bobcat2020, BlueCut2016, Pilot2016, Slide2007, Butler2007 along with many smaller fires. PolSAR images are shown to be helpful in identifying types and boundaries of fire, 50-m scale details of vegetation loss, and variability of vegetation recovery in post-fire years. In this region of the Transverse Range, where fire perimeters gradually fill in a mosaic pattern, using PolSAR to monitor recovery in gaps between recent fires may help to forecast the locations and extents of future fires.