along-track direction at 170-meter intervals from a 600-km
altitude orbit. Global Land Surface Altimetry (GLA-14) prod-
uct provides surface elevations for land. It includes the laser
footprint geolocation and reflectance (Kääb
et al.
, 2012). The
National Snow and Ice Data Center (
NSIDC
) distributes the
ICESaT GLAS
data products. Also, we used a 90 m spatial reso-
lution elevation model from the Shuttle Radar Topographic
Mission (
SRTM
) - version 4 for the year 2000.
Moreover, we used Landsat ETM+ statellite data acquired
close to the
SRTM
acquisition date. Table 1 provides acquisi-
tion dates of these scenes. We used these multispectral images
to classify the spectral signatures into different land cover
classes. We used glacier boundary shapefiles from the Global
Land Ice Measurements from Space (
GLIMS
) and Randolph
Glacier Inventory 5.0 (
RGI
) to identify glacial boundaries.
Methods
Figure 3 shows a flowchart of the methodology used to ana-
lyze the glacial change in the Andes. The
ICESaT
elevation data
was provided in Topex /Poseidon ellipsoid, and the
SRTM
data
was in
EGM96
Geoid; therefore, we geo-referenced both
ICESaT
and
SRTM
data to the
WGS84
ellipsoid before the analysis. We
created a 5-km buffer around the glacier boundaries. Based
on the location of the
ICESaT
footprint, we categorized
ICESaT
footprints as “on-glacier” and “off-glacier” points (Figure 3).
We used off-glacier footprints in the calculation of uncertain-
ty in the estimation of glacier elevation differences and mass
balance. We excluded the
ICESaT
footprints that fall within the
SRTM
void mask to avoid possible bias in glacier mass estima-
tion (Kääb
et al.
, 2012).
We first georectified and mosaicked Landsat images. Then,
we classified the images into five land cover classes, Clean
Ice, Snow, Debris, Water, and Others (Figure 3), using mul-
tiple band ratios and classification methods. The high albedo
in the snow and clean ice helps to separate snow from the
encompassing territory (Racoviteanu
et al.
, 2008). We used
the visual and near infrared (
VIN
) bands of Landsat to cre-
ate a band ratio of Band 3/Band 5. We applied a threshold
of 2.05 to this ratio to separate the clean ice and snow areas
from the rest of the land cover. Although, individual Landsat
image had different threshold, the value 2.05 was the high-
est value among all the images that separated ice and snow
from other land cover class. Then, we created a band ratio
of (Band 4 × Band 2)/ Band 5 (Kääb
et al.
, 2012). This ratio
T
able
1. L
ist
of
L
andsat
(ETM+) S
cenes
U
sed
in
this
S
tudy
Landsat
path/row
Scene
date
Cloud
(%)
Landsat
path/row
Scene
date
Cloud
(%)
231/095 10/14/01 2
232/084 1/20/00
0
232/090 12/8/01 0
232/085
2/7/01
0
232/091 12/8/01 0
232/086
2/7/01
0
232/092 3/11/01 0
232/087
2/7/01
0
232/093 3/11/01 8
232/088 12/8/01
0
232/089 12/8/01 0
233/085 1/29/01
1
233/083 12/26/99 0
233/086 1/27/00
0
226/099 12/14/01 18
233/087 11/29/01
0
227/098 2/7/02 12
232/086
2/7/01
0
229/096 2/21/02 6
233/082 2/28/00
1
229/098 3/19/00 32
232/094 5/14/01
10
230/096 10/2/99 7
232/083 12/5/00
0
230/097 8/4/01
9
233/080 11/26/00
0
231/094 10/27/00 10
233/081 3/21/02
0
Figure 3. Flowchart summarizing the methodology
814
October 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
