IEEE Geoscience and Remote Sensing Magazine - June 2022 - 52

Reference [149] employed the maximum-likelihood
classification and principal component analysis to classify
the first-year ice (FYI) and MY ice (MYI) in the Arctic region
using SCATSAT-1 sigma-0, gamma-0, and BT at dual polarization
(HH and VV), as shown in Figure 5(b).
As compared to sea ice type products [European OrganTHE
CRYOSPHERE, WHICH
COMPRISES POLAR SEA ICE,
ICE SHEETS, ICE SHELVES,
SNOW, GLACIERS, AND
FROZEN GROUND, PLAYS A
VITAL ROLE IN REGULATING
EARTH'S CLIMATE SYSTEM.
isation for the Exploitation of Meteorological Satellites (EUMETSAT)],
the classified maps
have achieved 92% and 88%
accuracies for the FYI and
MYI, respectively. The fall in
accuracy was observed from
March to May. The scatterometer
has already been proven
to be useful in the monitoring
and mapping of the snowmelt
extent at northern latitudes
[11], [150]. Moreover,
the
observation of surface melt
conditions is important to
evaluate the stability of Antarctic
ice shelves [151] because longer melt seasons and the
existence of melt ponds are linked to shelf breakup [152].
In the past few decades, various algorithms have been
developed and summarized to retrieve the surface melt information
using a scatterometer [153], [154]. Reference [20]
investigated the surface melting over the Amery ice shelf
(the third-largest embayed shelf in Antarctica) using QuikSCAT
for the period from 2000 to 2009 and found that
very prominent dips in the backscatter values in January are
due to the physical processes of snow/ice surface melt, and
these start rising in February because of the freezing phenomenon.
The results-derived melt index (MI) was found
to be in good correlation with the passive microwave-based
MI [155] and air temperature determined by the Australian
Antarctic Station " Davis. " Reference [25] implemented the
MI to study the shelfwise variability in the melt over different
areas using SCATSAT-1, as shown in Figure 5(c). The
melting grid is computed using predefined criteria based on
the austral mean and SD as follows:
logic " 1 " or " true " for ^^ 2- v# HHSDmaxhh
logic " 0 " or " false " for ^^2- v# HHSDmaxhh (4)
vv1 HHmean
vv1 HHmean
n
n
00 0
HH
00 0
HH
.
In addition, an adaptive threshold has also been utilized
to acquire the backscatter of a specific grid, and the MI can
be computed to analyze the shelfwise changes as follows:
n
= Ai/^h
#
MI MD ,
i=1
(5)
where A and MDi are the melt days (MD) with respect to the
total number of grids ,n respectively. Moreover, the concept
of temporal reduction in backscatter has also been defined
to derive the average MI using SCATSAT-1 level 4 BT products
over Antarctic ice shelves, and lower average MIs were
observed over the Ronne and Ross shelves as compared to
52
other shelves, i.e., Amery, Larsen-C, West, and Shackleton
(See Figure 5d) [25]. The average MI is computed as follows:
average MI=
^h/(BT
BTsi
MD
wH
,
(6)
where BTwH and BTsi are the average of the horizontally
polarized BT and its value on ith day, respectively. These results
have been well correlated with earlier studies based on
backscatter as well as passive microwave data [156], [157].
The SWE is also one of the important snowpack measurements
used by hydrologists and water researchers. The
SWE, as a direct integration of snow density over snow
depth [5], defines the amount of liquid water stored within
a snowpack [157]. Therefore, it is necessary to estimate the
SWE in a timely and accurate manner for freshwater supply,
agriculture irrigation requirements, disaster mitigation, water
management, and climate monitoring [158]. Due to the
inhomogeneous distribution of the snow cover, it is challenging
to accurately measure and model, especially over
large and complex terrain, like the Himalayas, due to fast
freezing and melting processes [159]. The Ku-band scatterometers
have already been proven their potential to retrieve
SWE measurements with ±0.3-dB relative accuracy
at three SDs for NSCAT [160] and ±0.3-dB relative accuracy
for SeaWinds [161].
Reference [25] demonstrated the change detection technique
to retrieve SWE temporal variations using SCATSAT-1
over the Himalayas, as shown in Figure 5(e) [162]. To characterize
the cross section of the snowpack, the radiative
transfer model can be utilized as follows [72]:
vv t h,
0 AA exp Cdg00
=- --^
h
where v and A0 are the backscattering coefficient and volv
ume
scattering from the snowpack, respectively; C and g
represent the normalized attenuation parameter and scattering
from the ground surface/short vegetation under the
snowpack, respectively; and the ,t snow density, and
d ,
snow depth, parameters indicate the SWE. The outcomes of
[25] showed a good correlation with the calibrated variable
infiltration capacity model SWE [see Figure 5(e)] [163], [164].
Reference [133] demonstrated the estimation of snow
conditions and global snow cover monitoring with NSCAT
backscatter. In the case of dry snow, the microwaves at the
Ku band can penetrate deeper into the snowpack and, as
the result, there is high backscatter due to the volume scattering
[165]. In this regard, [22] estimated the snow cover
area using SCATSAT-1 over the Himalayas with a machine
learning-based algorithm [see Figure 5(f)] and achieved
95.22% accuracy with an artificial neural network postclassification
comparison (PCC) and 93.26% accuracy with
linear mixture model (LMM) PCC. In addition, the seasonal
snow cover trends have also been demonstrated with
83.01-95.33% accuracy. Moreover, [27] performed a comparative
analysis of different classifiers-i.e., KMC [166],
support vector machine (SVM) [167], and LMM [168]-to
generate snow cover maps using SCATSAT-1. From the
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2022
^
.
(7)
IEEE Geoscience and Remote Sensing Magazine - June 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2022
