4. Study
Area
between 67° 25ʹ 41.62″ E, 24° 14ʹ 47.87″ N
and 67° 30ʹ 0.88″ E, 24° 12ʹ 30.24″ N
Keti Bander taluka
of Thatta district of
Sindh Province.
Mangroves, Prosopis, s
pp, marine algae, salt
bushes, and grasses for
this ecosystem
5. Data Acquisition
WorldView-2
captured on
10 February 2010
3002×2027 pixels
of multispectral
(1.85 m spatial
resolution)
12005×8105 pixels
of panchromatic
(0.46 m spatial
resolution)
UTM zone 42 N
with World Geodetic
System 84
6. GroundTruthing
Field photographs: (a) close canopy (Avicenna marina),(b) open canopy mangroves,
(c) salt bushes, and (d) typifying example of diverse land cover features in the area.
120 GCPs by using Garmin 76CSx
for closed and open canopy mangroves,
salt bushes, algal mat, and water, etc.
7. METHODS
Edge matching of pan and multis
pectral images
01 02
The two images, i.e. LRMI and HRPI sho
uld only be valuable for high resolution m
ultispectral
image (HRMI) if these have the features
with
same edges.
To achieve this, the AutoSync tool of the i
mage processing software (i.e. ERDAS I
magine) was
Image fusion
In this study, the candidate pan-sharpenin
g
algorithms were(1) ET, (2) MIHS (3) WT
(4) OHPFA and (5) SRM. These algorith
ms wereapplied to the WorldView-2 imag
e of
study area using ERDAS Imagine 2013 a
nd
ENVI 5.1 software.
8. 03
02
Delineation of reference polygons
of the tree crown
High-resolution satellite image of World
View-2
was used to delineate the boundaries of di
fferent canopies of mangrove trees on the
basis of
visual analysis and field survey measure
ments.These delineated polygons were used a
s a
reference for the evaluation of the spati
al
accuracy of image segments for each p
an-
sharpening technique.39 reference polygons scattered thro
ughout the area were delineated whic
h comprised mangroves of different
crown canopy sizes.
9. Correlation coefficient (r)
The correlation coefficient between the
HRMI and HRPI illustrates the degree of similarity of
their spatial gradients.
A
B
C
Calculating the spectral RMSE
Calculating the spatial Sobel filter-based RMSE
for spatial values of pixels
D
Image segmentation and thematic layer
generation
Using Definiens Developer 7®. The most suitable scale
parameter for each pan-sharpened layer was chosen
using the estimation of scale parameter (ESP) tool.
04 Performancemeasure and
quality assessment
Ashraf, Brabyn, and Hicks (2012):
Pradhan et al. (2006):
10. Clinton et al. (2010) :
Johnson,
Tateishi, & Hoan (2012) :
Class hierarchy :General classification schema of image
segment for over and under segmentation :
E.Spatial accuracy of
segments: calculation of
over segmentation and
under segmentation
F. Spectral accuracy of
segments: calculation of
RMSE
G.Defining class
hierarchy and applying
classification schema
05
Accuracy assessment
The overall, user and producer accuracies were determined using a confusion matrix.
Kappa coefficient (κ), standard error, and weighted error, with a 95% confidence interval for kappa were also calc
11. RESULTS &
DISCUSSION1. Pan-sharpening results
Original low resolution multispectral image (LRMI) with R
GB as 752.
(a)
High resolution multispectral image (HRMI) by Ehler’s Tra
nsformation (ET).
(b)
Fused HRMI by Modified Intensity Hue Saturation (MIH
S).
(c)
Wavelet Transformation (WT).
(d)
Fused image from Optimized High Pass Filter Addition (O
HPFA).
(e)
Fused image of Subtractive Resolution Merge (SRM).
(f)
12. SRM provided better results for Green, Red,
and NIR band.
2. Spectral and spatial
evaluation
A graphical representation for the average
values of r, spectral, and spatial RMSE.
r and spectral RMSE showed quite different
results highlighting OHPFA, SRM, and ET in
the queue for further assessment.
OHPFA preserved spectral values for the
blue band only.
OHPFA showed maximum correlation for
Red band, respectively, leaving behind SRM
for blue and green.
SRM is provided good results for the other
bands.
It clearly shows that the MIHS approach
provided minimum correlation and highlights
the second highest RMSE when compared
with WT and ET.
Considering the highest correlation and the
minimum RMSE, it can be deduced from
figure that OHPFA and SRM provided relatively
better results than the other techniques.
13. 3. Spatial and spectral accuracy of
image segments
High accuracy of SRM (Table 2) and OHPFA (Table 3)
in NIR band was further evaluated by generating
thematic layers and performing accuracy assessment.
14. From this table, it can be seen that the highest
accuracy of 92.3 and κ of 0.875 was attained by SRM
technique when compared with accuracy of other
techniques.
4. Thematic layer ge
neration
5. Accuracy assessme
nt of LCs
15. CONCLUSION
Mangroves ecosystem
mapping requires the s
election of suitable
sensor, which is high
quality data is more
important for
information
extraction & analysis.
Pan-sharpening
methods i.e. ET, MIHS,
WT, OHPFA, & SRM
qualitative and
quantitative analyses
were done evaluated
for mapping mangroves
ecosystem.
In the light of GEOBIA
framework, this hybrid
approach found SRM to
be the most suitable
technique for mapping
& assessing mangroves
ecosystem. OHPFA was
found to be the second
best pan-sharpening.
The outcomes will be very helpful f
or various applications
among which the accurate
calculation of crown projection
area using high resolution
satellite data is particularly helpful
in estimating the blue carbon in
this ecosystem.