Using Remote Sensing to Assess Burn Severity, Angora Fire,
South Lake Tahoe, CA
by: Jesus Alvarez
Remote Sensing, Emporia State University
AbstractRemotely sensed multi-spectral data collected by Landsat 7 Enhanced Thematic Mapper Plus (ETM+) are used to create several color composites of the South Lake Tahoe Basin before and after to the Angora fire, June 24, 2007. The severity of the burned landcover was assessed by creating a Differenced Normalized Burn Ratio (DNBR) used by the USGS National Park Service to assess fire severity from radiometric values. Landsat 7 ETM near-infrared and mid-infrared bands 4 and 7 are used to create Normalized Burn Ratio (NBR) for two different Landsat 7 datasets (before/after fire) with temporal scale of 32 days. Color composites and DNBR results are combined with a Digital Terrain Model (DTM) created from microwave Interferometric Synthetic Aperture Radar (SAR) datasets from Shuttle Radar Topography Mission (SRTM) (2000) to generate an orthographic perspectives of the color composites pre and post fire. South Lake Tahoe Vegetation The Sierra Nevada's vegetation near the South Tahoe region changes with increasing elevation. The lower elevations in the Tahoe basin have mixed pockets of conifer zones and white and red fir. The vegetation near the Angora lakes region have scattered pockets of white and red fir and grasslands near the camping areas and along the slopes of the Tahoe basin east of Fallen Leaf Lake. The sub-alpine elevations have pockets of hemlock, western juniper, lodgepole pine, and other species of pine, which become more scattered with increasing elevations. Fires in the Tahoe Region Increasing drought conditions in northern California have increased the intensity and frequency of fires in woodland areas. The Coast Ranges north of San Francisco and the Sierra Nevada to the east receive plenty of precipitation under normal conditions. Winds in northern California’s Sacramento Valley blow northeast towards the higher elevations from the lowlands of Sacramento River delta and westerlies are constant year round on both the Coast Ranges and Sierra Nevada west facing slopes (windward sides). In resent years precipitation has decreased in northern California igniting a state of drought condition in many counties, while demand for camping grounds in forested areas remains high increasing the probability of fires on both the Coast Ranges and inland mountains. A combination of low precipitation, constant winds, and campfires were the main factors in one of the most recent fires in northern California in the South Lake Tahoe region, the Angora fire.
The Angora fire took place on the west slopes of the South Lake Tahoe basin in the National Forest System (NFS), east of Fallen Leaf Lake, just east of Angora Lakes (see Figure 1). According to the Lake Tahoe Management Unit (LTBMU) (2007), the human-caused fire burned over 3,100 acres within the Wildland Urban Interface (WUI) and destroyed or damaged more than 250 structures. The fire started on June 24, 2007 in a camp ground east of Angora Lakes by an out of control campfire (location noted on Figure 1) and spread to the north along the west slopes of the basin. The fire was contained 100% a week after on July 2nd.
Figure 1. Fire location map used by operations planning during the Angora fire (LTBMU, 2007).
Burn Severity Analysis of the Angora Fire
Landsat 7 ETM+ datasets for the South Lake Tahoe area were downloaded from the USGS, Earth Resources Observation and Science (EROS) satellite data download site. The datasets were downloaded for June 20, 2007 (pre-fire) and July 22, 2007 (post-fire). Both datasets for the Tahoe region (path 43 and row 33) contain 0.09% cloud cover (see Figure 2. for spectral and spatial resolutions). The datasets have Scan Line Corrector (SLC) off, which may affect the DNBR values in the areas away from the center of the scene. The SLC, which compensated forward motion of Landsat 7 failed on May 31, 2003. Without an SLC the ETM+ line of sight now traces a zig-zag pattern along the satellite ground track (USGS, 2003).
Figure 2. Landsat band properties. The images show the bands for path 43, row 33 (Lake Tahoe Region). Both datasets used on the project have 0.09% cloud cover. Lake Tahoe is shown near the center of the bands and the Angora region is located below on the south sores of the lake.
To assess the geography of the Tahoe basin, a smaller area concentrating around the Fallen Leaf Lake was cropped from all bands and several composite image scenes were created by combining the Landsat ETM+ bands of the June 20 dataset (pre-fire). The color composites show the vegetated areas, snow-covered mountains, bare rocks and water features prior to the Angora fire (see Figures 3 - 7). All image composites, NBR and DNBR analysis and calculations were generated with image processing software IDRISI Taiga (2009).
Figure 3. Pre-fire, natural-color composite. Snow-covered mountains and bare rocks on the Sierra Nevada standout on the left side of the scene shown with bright gray-white colors.
Figure 4. Pre-fire, color composite. Vegetated areas are shown in green bright tones, water is dark blue, snow cover is cyan, bare rocks and soils are light pink to brown colors, light gray tones show roads and paved airport ramps south of Lake Tahoe.
Figure 5. Pre-fire, color-infrared composite. Vegetated areas are show with bright red tones, snow is bright white, rocks and bare soils are light gray, and water features are dark blue.
Figure 6. SRTM 2000, interferometric SAR. 1-arc second (30m) resolution Digital Terrain Model (DTM) of the Tahoe basin enhanced with shading.
Figure 7. Pre-fire, orthographic color-infrared view from the southwest. Both the SAR and Landsat datasets have a 30 meter spatial resolution. The SAR image was resampled to match the same number of pixel rows and columns of the Landsat bands in order to drape the color composite over the SAR. The orthographic perspective was created with a vertical exaggeration of 2x the Tahoe basin is clearly visible between the two snow cover mountain ranges.
The fire in the Tahoe basin started in the afternoon of June 24, 2007 and lasted approximately 8-days. Landsat 7 has a 16-day revisit interval, the next available Landsat dataset with the same amount of cloud cover (0.09%) was obtained from July 22, 2007. The same bands were used to produce color composites of the same geographic area after the burn. The amount of energy reflected and absorbed by the spectral windows in all bands changed dramatically for the Angora burn area (see Figures 8 - 10).
Figure 8. Post-fire color composite. The burned area is visible on orange-red colors, also noticeable is the lighter green tones of the vegetation and the reduced amount of snow cover shown on cyan color compared to the pre-fire composite (Figure 4).
Figure 9. Post-fire, color composite. The Angora burned area is shown with dark gray color. The red tones shown the vegetated areas are now lighter compared to those of Figure 5. Bare rocks and soils are shown with light gray tones and snow cover is no longer visible as on Figure 5.
Figure 10. Close look of burn area from southwest of Fallen Leaf Lake. The color-infrared composite was draped over SAR DTM with vertical exaggeration of 2x. The fire is clearly visible along the east side of the drainage divide east of Fallen Leaf Lake.
The Differenced (delta) Normalized Burn Ratio (ΔNBR) was generated for the Angora burn area from pre-fire and post-fire datasets bands 4 and band 7 spectral resolutions: band 4 (.76 - .90 µm) and band 7 (2.08 - 2.35 µm). To minimize the effects of the SLC-off pixels and the snow cover difference observed between the pre and post fire scenes, the analysis area was cropped to an area around the Angora fire. The SLC affected pixels do not coincide between the two Landsat datasets (pre and post fire), however, they only cover about 4% of the new cropped area. Prior to generating the ΔNBR, the NBR was calculated, scaled, and adjusted for SLC pixels for both the pre-fire and post-fire datasets. To avoid any obscurity of normalized values a mask was created for each dataset from the pixels containing SLC values, a mask was generated for each dataset and was applied to the unscaled NBR results prior to calculating the ΔNBR. The results of the NBR calculations after applying the mask were scale by 103 to have a quantitative measurable scale to be classified.
Calculations used to generate NBR for pre-fire and posr-fire and ΔNBR:
NBR = (band 4 - band 7)/(band 4 + band 7)
pre-fire NBR scaled = (NBR*(mask pre-fire))*( 103)
post-fire NBR scaled = (NBR*(mask post-fire))*( 103)ΔNBR = (NBR pre-fire - NBR post-fire)
The ΔNBR calculation resulted in values ranging between -1298 to +1222, which isolated
the unburned or active vegetation from the burned area on the resulting
image (see Figure 11). The mask applied was able to neutralize the values by
assigning each pixel a value of 0 before the ΔNBR
calculation. Based on the area of the SLC pixels only about 4% of the ΔNBR
image area is affected by SLC pixels. The ΔNBR results were classified into six different classes
based on the natural breaks observed between the burn and unburned values (see
Figure 11). A histogram of the burn values was created to observe the frequency
of distribution of burn pixel values (ranges between +100 to + 1222) (see Figure
12) and the areas were calculated for each of the class created (see Figure
The results produced by ΔNBR isolated the burn areas from the unburned areas. The affected area delineated by the burn pixels seems to correlate with that of the map used during the fire fighting operations. To better understand the burned pixels boundary it was necessary to drape the ΔNBR scene with the SAR DTM (see figure 14). The orthographic view of the ΔNBR scene gives a better perspective of the burn areas. The pixels with severe burn values are concentrated on the sloped pockets (facing Nevada) formed by the geography of the mountains east of Fallen Leaf Lake (see Figure 14).
Conclusion With the use of remotely sensed datasets from Landsat 7 ETM+ it is possible to analyze the geography of an area affected by a fire and assess the burn severity. The wide spectral range offered by the visible, near-infrared, and middle-infrared bands allow the composition of color composites that can help assess the health and presence of vegetated areas, identify bare soils and rocks and delineate water features. The changes of reflected and/or absorbed energy recorded by the Landsat sensors allows the creation of radiometric values by obtaining the difference of NBR's of pre and post fire datasets. Small burn areas in the center of the Landsat bands are not affected by the SLC-off and those that are affected by the SLC-off pixels can and should be minimized by applying a mask prior to generating the ΔNBR. Depending on where the burn occurred in the Landsat scene, it is possible to neutralized or minimize the effects of SLC-off pixels. If the burn occured near the edge of the Landsat bands it may be necessary to generalize the SLC-off pixels by resampling to the nearest neighbor pixel value. The classification of the ΔNBR values should be based on the natural breaks observed on frequency of the burn values. The classification used in the Angora fire produced a boundary almost identical to that digitized by the fire crews and the areas calculated from the ΔNBR's were within 9% of the actual burned areas. The temporal scale (16 days revisit) of the Landsat datasets allows for generations of ΔNBR's of burned areas under good atmospheric conditions (e.g. less than 10% cloud cover). Care must be taken when producing ΔNBR's since active vegetation, snow, and/or other atmospheric effects in mountainous regions can skew the results of ΔNBR's. The burn ratios should be produced from datasets of approximately the same season. To obtain sound values from ΔNBR's Landsat datasets with minimal cloud cover should be used and under the same or approximate water conditions (drought or normal) since bands used (4 and 7) are affected by water content in vegetation. Interferometric SAR datasets that share the same spatial resolution of Landsat ETM+ bands can be used generate orthographic perspective views of burned areas. Using DTM from SAR helped understand the geography, delineation of burned areas and the nature of their shapes formed. In addition, ΔNBR results draped over a DTM can be a valuable tool to predict mudslides and or soil erosion when fires occur in the slopes of mountainous regions.