IEEE Geoscience and Remote Sensing Magazine - June 2022 - 49

new GMF (CMOD7) to achieve better performance on all
relevant quality parameters as compared to CMOD5 [109].
As far as validations are concerned, the NSCAT windretrieval
information was tested against the in situ observations
and numerical weather prediction (NWP) models
and delivered above-expectation outcomes [110]-[112]. After
the premature failure of the NSCAT and SeaWinds scatterometers,
QuikSCAT offered the wind measurements with
an RMS difference of 0.7 m/s and 13° [113]. As compared
to ASCAT for different seasons, OSCAT (recognized as the
continued mission of QuikSCAT) has shown a correlation of
.,
2075 bias of 1- to 1 m/s, and RMS difference of125 . m/s
[114]. With the recent advancements in computing capacity,
deep learning-based weather prediction is expected to be a
strong supplement to enrich the conventional theory-driven
method for forecasting certain meteorological attributes or
weather phenomena. Reference [115] reported the inclusion
of data-driven deep learning into weather forecasting models
and, also, achieved the preliminary outcomes.
With advancements in the wind-retrieval algorithm,
better performance can be achieved in meteorological and
oceanographic applications. To overcome the directional
ambiguity and improve the wind fields derived from a scatterometer,
two wind-retrieval models-i.e., the normalized
standard deviation (NSD) [see Figure 4(b)] and directional
stability and conservation of scattering (DiSCS) algorithms
[see Figure 4(c)]-have also been developed. Reference [116]
developed the NSD algorithm for QuikSCAT, which can be
expressed as follows:
NSD
where WQ
Q=-Q
mk
k
=
=
(wind direction), Wmk
W M WW
11
kM
Q
/^h (1)
1
2
,
(wind speed), and NSDQ
(cost function) corresponding to the minima of the NSD
along the trial directions represent the vector solutions.
The result of the NSD algorithm has been compared with
the National Centers for Environmental Prediction (NCEP)
and ECMWF model winds. From the results, an RMS difference
of 0.8 m/s in wind speed and mean absolute difference
in wind direction of 8.56° have been observed.
Reference [118] developed the DiSCS algorithm to improve
the retrieval of wind vector fields by reducing the
directional noise while retaining the natural wind flow.
Ambiguity removal is performed for the DiS of winds
and invoking the CS. The DiS S jk
cell (WVC) in the jth row and kth column is computed
as follows:
jk
Z
[
\
]
]
S = ]]
where jk
M
R
T
S
S
SS
nj N
=+
nj
=
// ^ QQ hmod^180ch
-
ik N
ik N
==
=+
ni
^ N+
Q and ni
21h2
2
jk
M
2
V
X
W
W
WW
-2
_
`
a
b
b
bb
,
(2)
Q are the circular median direction and
direction falling within N neighboring cells, respectively.
JUNE 2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
The results of the DiSCs performance were compared with
NCEP and ECMWF using that QuikSCAT data set and found
to be closer to the model winds as compared to those in the
QuikSCAT winds [118]. From the experimental outcomes,
the RMS difference in directions is lower for the ISRO
winds as compared to the NASA winds (up to a 15-m/s
wind speed) but marginally
higher at a 20-m/s wind speed.
An improved model based
on a multiple-solution scheme
(MSS) in conjunction with a 2D
variational ambiguity removal
(2DVAR) method has been developed
by the KNMI [120]. In
this model [see Figure 4(d)], the
backscattered measurements
are grouped, including metadata
in each WVC for wind retrieval.
The Bayesian approach-based
maximum-likelihood estimation
(MLE) method can be utilized to obtain the high-quality
winds derived from backscatter measurements using a GMF
[116], [124]. An MLE function is computed as follows:
MLE= N i=1
/
a set of 0
1 ii^ ^^
N
vv
Kp
obs
obs
,
GMF
0 hhh2
-
.
(3)
The MLE is defined as the measure of distance between
v and v0
GMF backscattered stimulated by the GMF
for a given trail value of wind speed and direction [105]. N
and Kp are defined as the number of available measurements
and variance of the measurement error, respectively.
The MLE is computed via wind speed (),
WITH ADVANCEMENTS IN
THE WIND-RETRIEVAL
ALGORITHM, BETTER
PERFORMANCE CAN BE
ACHIEVED IN
METEOROLOGICAL AND
OCEANOGRAPHIC
APPLICATIONS.
150m/s and the
speed with minimum MLE acts as a function of wind direction,
called the MLE cost function [120]. The MSS retains all
of the solutions and, the probability of being correct is given
via the empirical probability function (depends on the
scatterometer instrument and measurement noise). At last,
the 2DVAR approach is implemented to select the specific
ambiguous wind solutions as the observed wind for every
WVC [120]. From the comparative analysis, it has been
found that wind speed derived through the MSS is noisy for
low ()
14m/s and high wind speeds (),
for a given wind vector
217m/s while the
wind direction represented the lower bias and stability for
the wind speed even beyond 17 m/s.
NWP computes the current weather conditions via mathematical
models of the atmosphere and oceans to predict the
weather. In NWP [see Figure 4(e)], the measurement of wind
vector data are one of the critical driving forces for better understanding
and monitoring air-sea interactions, extreme
weather conditions, and other processes [67], [121], [126].
The scatterometer provides the high-resolution wind vector
information with frequent temporal availability at a global
level [127]. Reference [128] highlighted the significance of
SeaWinds in NWP, with the effects propagating well away
from the swath of the instrument within the first 6-18 h of
49
IEEE Geoscience and Remote Sensing Magazine - June 2022

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