Abstract
The purpose of this paper is to investigate the ecological damage and healing progress of the 1989
fire scars in Yellowstone National Park. Using datasets captured by both Landsat and ASTER, this
data was processed using ENVI software to determine the change over time. Composite color
images, reflectance plots and density slices were among some of the chief tools used to make these
determinations. Additionally, two unsupervised classifications were run and cross references to
SWIR TM data to better help understand the boundaries of the fire scars. It was determined that
the boundaries of the scars shrunk over time, highlighting new vegetation growth and ecosystem
healing. TIR data from ASTER was also used to create a density slice to determine the hottest
pixel in a given region. Using a slice based on the hottest recorded temperature in the image, it
was determined that the hot pixels were located on the left-hand side (this can be observed in figure
7 below).
Introduction
In 1988, wildfires swept through almost 800,000 km2 of Yellowstone National Park, leaving large
fire scars. Using images captured by the Landsat Thematic Mapper (TM) and the Advanced
Spaceborne thermal Emissions and Reflection Radiometer (ASTER), these fire scars have been
able to be analyzed over time. Analyzing TM imagery from September of 1990 2000, and 2010,
respectively, the progress of the recovery was able to be seen. Although not complete in the 2010
TM data, the scarring had improved greatly. This was able to be accomplished by using a series of
image enhancements and spectral analyses in both VNIR and SWIR. In addition to this method,
several unsupervised classifications were employed to help determine specific classes to aid in
determining which regions were scarred. The classification methods and class meanings will be
discussed in greater detail below.
Methods
Initial Analysis
This analysis was performed using ENVI software. A true color composite image based on the
respective bands available was loaded into the software for preliminary analysis. Several stretches
were run on the image. Subsequently, the metadata for the TM bands was examined to confirm the
USGS projection, sensor, pixel units, number of bands, along with a number of other important
metrics. In addition, the spatial resolution of the image was found to be 30 meters (TM resolution
is 30 meters, with exception of its TIR band, which is 120 meters).
Spectral and Temporal Analysis
The converted reflectance bands were loaded in R, G, B using bands 7, 2, and 4, respectively.
Additionally, the same band combination was loaded using the original TM bands. These screens
were then linked, and their spectra were examined in both standard TM data and reflectance data.
Four distinct pixels were located and analyzed in this fashion; subsequently, surface composition
was assumed based on the respective spectra. The change over the 10-year period from 1990 to
2000 using TM and ASTER imagery, respectively. Two color composite images in R, G, B were
loaded and observed to look for changes in the scar boundaries after a decade. Both the TM and
ASTER images were loaded from converted SWIR reflectance bands.
Image Classification in SWIR and TIR
Two unsupervised classifications, along with a density slice were generated to detect general
surface groupings, along with heat signatures. The first classification generated was ISODATA,
set to have 5 classes as a maximum. The newly formed ISODATA band was then loaded in,
revealing 3 discernable classes – blue green and red. These classes were algorithmically
determined to have statistical significance. The ISODATA figure will be included below (see
figure 1). Additionally, K – mean was run set to have a maximum of 5 classes. These classes were
also algorithmically determined based on the center of each class and its statistical significance.
This figure will also be included below (see figure 6). TIR data was also used to generate a density
slice. Band 14 of the TIR file was used and the assumed emissivity was set to 0.99. These settings
were saved and then used to generate a density slice to determine the temperature in each pixel
(recorded in kelvin). The spectral plot for this pixel can be seen below (figure 2).
Results and Analysis
After performing several stretches on the image, it was determined that the image equalization was
the most effective stretch for determining specific features in an image. This stretch made it much
easier to distinguish finer geological features such as rivers and ridges. Linking the reflectance and
standard TM data screens revealed that different scales are used for both reflectance and standard
TM on the spectral plots. Reflectance is plotted against the percent of each wavelength that is
reflected, while standard TM is plotted against DN values (both having wavelength on the x-axis).
An example plot for both reflectance and standard TM will be included below (see figures 4 and
5, respectively). Using both the absorbance and reflectance plot in unison gave the unique
advantage of confirming one against the other. Observing the 10-year change from the 1990 TM
and 2020 ASTER imagery revealed that much of the damage done by the fire scars had been
healed. While the growth was not absolute, it could clearly be seen that the boundaries of the scars
had receded. Additionally, the ISODATA classification proved to be more helpful than K-mean in
determining where the fire scarring occurred. Cross referencing the ISODATA classification with
SWIR TM data it seemed that the red areas were indicative of the scarring. K-mean data was left
as unconclusive as to which parts were representing the scarring. The TIR density slice was split
into several bulk temperature classifications. After reducing the slice to only one classification
based on the hottest temperatures recorded in the image, the hottest pixel was located toward the
left-hand side. This final image will be included below (see figure 7).
Conclusions
Using remote sensing can be a very effective way to monitor ecological damage over time. Using
the methods outlined in this paper, such as unsupervised classifications, density slices, and
reflectance bands, the progress of ecological healing was able to be seen. The Yellowstone fire
scar boundaries had shrunken since the initial burn. ISODATA statistics made it possible to cross
reference the TM imagery in SWIR with the generated classifications to better determine fire
scarring. Additionally, density slices could be used in the future to model and determine the spread
of wildfires in order to help fight them more effectively.
