Comparative analysis of Worldview-2 and Landsat 8 for coastal

saltmarsh mapping accuracy assessment

Sikdar M.M. Rasel*a, Hsing-Chung Changa, Israt Jahan Ditib, Tim Ralpha, Neil Saintilana
a
Department of Environmental Sciences, Macquarie University, North Ryde, NSW-2109, Australia.
b
Department of Soil Science, Faculty of Agriculture, Rajshahi University, Rajshahi-6000,
Bangladesh.

ABSTRACT

Coastal saltmarsh and their constituent components and processes are of an interest scientifically due to their ecological
function and services. However, heterogeneity and seasonal dynamic of the coastal wetland system makes it challenging
to map saltmarshes with remotely sensed data. This study selected four important saltmarsh species Pragmitis australis,
Sporobolus virginicus, Ficiona nodosa and Schoeloplectus sp. as well as a Mangrove and Pine tree species, Avecinia and
Casuarina sp respectively. High Spatial Resolution Worldview-2 data and Coarse Spatial resolution Landsat 8 imagery
were selected in this study. Among the selected vegetation types some patches ware fragmented and close to the spatial
resolution of Worldview-2 data while and some patch were larger than the 30 meter resolution of Landsat 8 data. This
study aims to test the effectiveness of different classifier for the imagery with various spatial and spectral resolutions.
Three different classification algorithm, Maximum Likelihood Classifier (MLC), Support Vector Machine (SVM) and
Artificial Neural Network (ANN) were tested and compared with their mapping accuracy of the results derived from
both satellite imagery. For Worldview-2 data SVM was giving the higher overall accuracy (92.12%, kappa =0.90)
followed by ANN (90.82%, Kappa 0.89) and MLC (90.55%, kappa = 0.88). For Landsat 8 data, MLC (82.04%) showed
the highest classification accuracy comparing to SVM (77.31%) and ANN (75.23%). The producer accuracy of the
classification results were also presented in the paper.
.
Keywords: Saltmarsh, Worldview-2, Landsat 8 OLI, Classification, Assessment, MLC, SVM, ANN.

1. INTRODUCTION
Saltmarsh is an intertidal community and treated as Ecological Endangered Community (EEC) in Australia [1]. For
these reasons, monitoring and dynamic change analysis of saltmarsh is a pressing issue and scientists are much more
dependent on high quality remote sensing data for mapping and monitoring of saltmarsh and their proactive management
[2-4]. However, this community usually lives with dominant mangroves. This causes a problem in selection of scale and
radiometric properties of the remotely sensed data when performing classifications. The scale or spatial resolution is the
projected area on the ground associated with the radiance measurement of the sensors. It is the smallest distinguishable
spatial unit (ground resolution element) recorded in a remotely sensed data. For example, the spatial resolution is 2 m for
Worldview-2 data and 30m for Landsat 8 OLI or EO-1 Hyperion data. Sometimes the spatial resolution denotes the
ground sampling distance in an image after image re-sampling that can be varied from the actual spatial resolution
recoded by the sensor. Spectral resolution means the range over the electromagnetic spectrum the energy is measured
and recorded by the sensor. The use of higher spectral resolution images usually improve the capability of detecting
spectral variability within a land cover by increasing the discrimination capacity [5]. But spatial resolution may be more
important than spectral resolution when spatial extent of land cover types is an issue. However, both spatial and spectral
resolution are two fundamental characteristics of a remotely sensed image for any application, like as classification or
feature extraction.
Remote sensing image classification accuracy is affected by two factors: firstly, influence of boundary pixels and
secondly, finer spatial resolution that increases the spectral radiometric variation of land cover types [6]. Optimum
spectral and spatial resolution determination for vegetation mapping has been an ongoing area of research in remote
sensing [7]. It becomes more challenging when dominant and fragmented plant species patches are distributed together
in a same community, similarity to our study site. Our study area supports an extensive temperate saline coastal wetland
system comprising mangrove, saltmarsh and mixed mangrovesaltmarsh habitats [8]. About 4,257 ha of these wetlands

Sensing for Agriculture and Food Quality and Safety VIII, edited by Moon S. Kim, Kuanglin Chao, Bryan A. Chin,
Proc. of SPIE Vol. 9864, 986409 2016 SPIE CCC code: 0277-786X/16/$18 doi: 10.1117/12.2222960

Proc. of SPIE Vol. 9864 986409-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 are formally protected within the Hunter Wetlands National Park and this includes over 2,926 ha listed as internationally
important under the Ramsar Convention in 1984. However to facilitate agriculture and industrial production, a network
of 176 levees, culverts and floodgates that were constructed in the period 19501980 under the Hunter Valley Flood
Mitigation Scheme [9]. As a result wetland areas behind levee banks have transitioned from saltmarsh to brackish reed
swamps dominated by Schoenoplectus subulatus,Typha orientalis and Phragmites australis [9]. Due to this, the whole
wetland and surrounding part is a place of scientific research based on different objectives. Our interest was to know the
current status of theses degraded saltmarshes with respect of the other co-habitats, for example mangrove and Casuarina
sp and to find out optimum mapping accuracy for each saltmarsh species from remote sensing data.
Some research for wetland and saltmarsh classification has been done in Australia based on the high resolution data. In
the Northern Territory, Australia, a study [10] compared aerial photograph with SPOT multispectral (XS) and Landsat
for classification where aerial photography was showing more accurate result compare to SPOT XS and Landsat TM.
Similarly, another study [11] demonstrated higher classification accuracy (81%) from digital aerial photography in
contrast to SPOT-4 (63%) and Landsat ETM+ (53%) when the authors classified wetlands of the Hunter Region in
Australia. Most of the results discussed above are based on wetlands mapping at regional scale where species
classification was not the major concern. Species classification based on spectral properties of different vegetation
classes within the wetland environment is a challenging task. Saltmarsh vegetation types may possess similar spectral
signature in remotely sensed image, and spatial and spectral resolution of data may be insufficient to detect saltmarsh
class. That is why spectral and spatial resolution are the current limitation of monitoring in this environments and
researcher is therefore focused on improving sensor specifications.
In this study we used data collected by two different sensors: High Resolution Satellite Imagery (HRSI) Worldview-2
and recently launched coarse spatial resolution Landsat 8 OLI . Both sensors have some similarities in their spectral
coverage, however their spatial resolution (pixel) is very different. This paper aims to examine the effects of sensor
spatial and spectral resolution in a complex classification scene.

2. DATASET AND METHODS

Study site
The study area is located (figure 1 ) in Tomago, Australia which is approximately 10 km north of Newcastle (NSW)
(Figure 1). The topography is generally flat, low lying and subject to periodical flooding. A series of drainage channels
and levee banks dissect the study area from sea.
Input Data

Two different platform satellite imagery of Worldview-2 and Landsat 8 OLI were used for this research. Worldview-2
has a pixel resolution of 0.46m in the panchromatic mode and 1.84 m resolution in the multispectral mode whereas
recently launched Landsat 8 OLI consist of nine spectral bands with a spatial resolution of 30 meters for Bands 1 to 7
and 9. New band 1 (ultra-blue) is useful for coastal and aerosol studies. Another new band 9 is useful for cirrus cloud
detection. The resolution of Band 8 (panchromatic) is 15 meters. Thermal bands 10 and 11 are useful in providing more
accurate surface temperatures and are collected at 100 meters [12]. Approximate scene size is 170 km north-south by 183
km east-west (106 mi by 114 mi). In our study we did not use thermal and cirrus bands (table 1)

Proc. of SPIE Vol. 9864 986409-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 380000 384000
380000 384000

380000 384000

Figure 1. Location of the study area on World Imagery (Google earth map) (Left) and Landsat 8 RGB (Right).

Table 1. Specification of Worldview-2 and Landsat 8 OLI data.

Worldview-2 Landsat OLI imagery
Bands Central Band with Bands Central Wavelength (nm)
Wavelength (nm) Waveleng range
(nm) th (nm)

Coastal 427 396 - 458 Coastal 443 433 - 453

Blue 478 442 - 515 Blue 482 450 - 515
Green 546 506 - 586 Green 562 525 - 600
Yellow 608 584 - 632 Red 655 630 680
Red 659 624 - 694 NIR 865 845 -885
Red Edge 724 699 - 749 SWIR 1 1610 1560 -1660
NIR1 833 765 - 901 SWIR 2 2200 2100 - 2300
NIR2 949 856 - 1043 Pan 600 500 - 680
Pan 627 447- 808 CIRRUS 1370 1360-1390

Image acquisition and Image processing

For WorldView-2 imagery, the raw Digital Number (DN) has been converted at sensor radiance using Worldview-2
calibration utility, available in ENVI V4.6 and greater [13]. The image was acquired in May 2015 that is the transition of
season from summer to winter in NSW, Australia. Based on the initial analysis of the DN value of the image we found
that coastal band (lower spectral region) is affected with atmospheric scattering. To compare the effect of atmospheric
correction of image spectrum we applied atmospheric correction method, The Fast Line-of-sight Atmospheric Analysis
of Spectral Hypercube (FLAASH). Afterwards, WorldView-2 imagery was registered to Map Grid of Australia
(MGA94) Zone 56 using topographic map of local council, Land and Property information, NSW, Australia
(https://maps.six.nsw.gov.au/). A first order polynomial transformation was applied and RMSE values estimated for this
transformation was 0.32 pixels. The first order polynomial was selected because this wetland area has a generally flat
terrain. However, the RMSE was considered well enough considering the nominal GPS measurement error of 3-4 m that
we obtained in the field.

Proc. of SPIE Vol. 9864 986409-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Similarly, the DNs of the Landsat OLI imagery were first converted to at-sensor radiance by using the radiometric
calibration parameters. The FLAASH algorithm was then used to convert radiance to reflectance. Georeferenced
Worldview-2 imagery was used for rectification of Landsat OLI imagery. A first order polynomial transformation with a
nearest neighborhood resampling was applied and RMSE values estimated for image transformation were about 0.87
pixels.
Processing of field data
Randomly collected ground reference sample from different strata were divided into two groups, calibration and
validation set. This is the basic criteria to remove any possible bias that could be caused by using the same set of pixels
to calibrate and to validate the classifiers. It is known that Landsat OLI data were hampered by the low spectral
dimensionality and spectral resolution [14]. Due to this, maximum seven endmembers that can be extracted from
Landsat OLI images based on theoretical approach. However maximum of four endmembers is usually determined due
to strong correlations among the first four visible bands [15]. In this context, seven endmember, may not be fully
satisfied in consideration of the complex land cover system in the Hunter Wetland National Park. In this circumstances,
to deal with medium resolution (30 m) Landsat OLI data we prepared two data set for classification purposes. One
dataset is prepared based on the overlaying of field sample vector file (stratified random sampling) over Landsat OLI.
There is another way to prepare dataset based on Principal Component Analysis (PCA). Because PCA used the highly
uncorrelated bands in a 2-dimensional scatterplot (figure 2) and could be used to identify the individual endmembers of
multiple surface components [16-17]. According to [16] and [17], in a mixture of three substances, the first two
components produced a triangle (Figure 2) in a scatterplot to locate the endmembers at the corner of triangle. Hence we
used both method to make our training and test pixels unbiased as much as possible.
For Worldview-2 data, total 2228 training and 1853 test data were used for eight different classes. On the other hand 638
pixels were used based principal component analysis to calibrate the model for Landsat 8 OLI and field based 529 pixels
were used to validate the result. It was difficult to select pixels for Ficiona nodosa from 30X30m pixel due to their
narrow patch in against 30x30 m pixels. For Worldview-2 data, total 2228 training and 1853 test data were used for eight
different classes.

PC Bane 2 (AILI. yer1B9ill.df):P2A3.tif

i
n

8
2
-0.113
0.1 5
PC Iknd 1 (AILLarraEL1111.50:PcA-s.trf

-a:os
PC Bond 2 r.C43.tif
'0i.;,13_8',.;,),' a.@ ois

Figure 2: Principal Components Analysis (PCA) based endmember selection for classification of Landsat 8 OLI data.
Scatterplot of the three PCs from Landsat OLI data to select seven endmember: (Clockwise) PCA 1vs PCA2 : Mangrove,
Grass, Phragmitis, Water; PCA 1 vs PCA3: Casuarina, Sporobolus, Water ; PCA 2 vs PCA3: Phragmitis.

Proc. of SPIE Vol. 9864 986409-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 On the other hand 638 pixels were used based principal component analysis to calibrate the model for Landsat 8 OLI and
field based 529 pixels were used to validate the result. It was difficult to select pixels for Ficiona nodosa from 30X30m
pixel due to their narrow patch in against 30x30 m pixels. In addition, it is imperative to keep maximum class number 7
for Landsat 8 OLI data when we deal with only 7 bands (table1). Moreover our study area scene was completely cloud
free and it was unnecessary to add cirrus band (1360 1380 nm). We did not deal with the effect of temperature on
vegetation and discard thermal infra-red bands TIRS 1 and TIRS 2 (1060 to 1119 and 1150 to 1251 nm). Rest seven
multispectral bands (Table 1) were used in the classification process.

Selected classifier
Maximum Likelihood Classifier (MLC), Support Vector Machine (SVM) and Artificial Neural Network (ANN) have
been selected for classification purposes. MLC is one of the statistical classifiers. It is based on that depends on the idea
that the distribution of the data in each class are normally distributed and called as parametric classifier. While the SVMs
identify the optimal hyperplane that creates the largest distance, or margin, between the vectors for the two classes. In
situations where the training sets for the classes are not linearly separable, the classification is performed in the mapped
space (higher dimensional space) instead of the input feature space [18]. ANN works based on the processing node that
corresponds to the neuron of the human brain where each node receives and sums a set of input values and passes this
sum through an activation function providing the output value of the node [19]. Here input layers represents for spectral
bands of the data and output layers are the number of classes for the network.
Accuracy Assessment and kappa analysis
Accuracy assessment results were discussed based on the confusion error matrix, overall map accuracy and kappa values.
For individual class, producer and user accuracy were computed based on the dominant class in each reference plot [20-
22].
Mapping Accuracy (MA)
MA [23] was computed based on the following formula

3. RESULTS
Principal Component Analysis
The PCA eigenvalue showed that the first three PCA components accounted for almost 99.50% of the total variance for
Worlview-2 data and 99.76% for of the total variance for Landsat 8 OLI data. However, if we consider the pixel
resolution of Worldview-2 data then rest 0.50% variance is really important for 2X2 m pixels. But we can ignore rest
0.25% variance (Table 2) based on 30x30 m pixels resolution for Landsat OLI data to select independent endmember
dataset for classification. In this perspective our calibration data from Landsat OLI extracted maximum number of
endmember (for Landsat 8 it is 7) from uncorrelated PCA bands. Existing studies suggest that the pixels located on the
corner of the scatterplot (Figure 2) can be treated as endmember [24].

Proc. of SPIE Vol. 9864 986409-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Table 2: Percentage depiction of gain in variance with increase in PCs in both dataset
Wprldview-2 Landsat OLI
Eigen % Cumulative Eigen % variation Cumulative
PCs Variation percentage
value value Percentage
1 0.063954 96.9822 96.982288 0.023215 93.9802445 93.9802445
2 0.001278 1.9380 98.920296 0.001256 5.08460853 99.0648530
3 0.000392 0.5944 99.514739 0.000174 0.70439640 99.7692494
4 0.000195 0.2957 99.810445 0.000029 0.11739940 99.8866488
5 0.00005 0.0758 99.886267 0.00002 0.08096510 99.9676139
6 0.000032 0.0485 99.934793 0.000007 0.02833778 99.9959517
7 0.000024 0.0363 99.971187 0.000001 0.0040482 100
8 0.000019 0.0288 100

When the spectral curve of these selected endmember were compared with the original image spectra, it was also proved
that maximum of seven endmember (Figure 3) could be extracted from Landsat OLI data. Based spectral curve error
bar it is very clear that all saltmarsh species are overlapped with each other within all bands of the spectrum. But
Mangrove and Casuarina tree species are clearly separable from each other up to bands 5. Similarly, band 1-4 are clear
indication to separate Marshy wetland from water.
Spectral profile Analysis:
On the other hand for Worldview-2 data (Figure 4), reflectance properties of water in bands 6-8 (Red Edge, NIR1 and
NIR 2) and reflectance of marshy wetland in band 7 are significantly different from other classes.

0.4

0.35
Mangrove
0.3 Casuarina
0.25 Phragmitis
% Reflectance

Sporobolus
0.2 PerrenialGrass
0.15 Water
MarshyWetland
0.1

0.05

0
1 2 3 4 5 6 7
-0.05
Different bands of Landsat 8 OLI data

Figure 3: Different endmember (class) extracted from 7 bands of Landsat 8 OLI data

Proc. of SPIE Vol. 9864 986409-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Perennial grass are unique in band 3(Green) and 6 (Red Edge) and Ficiona nodosa in band 8 (NIR2). However in most
of the bands, the error bars of Mangrove, Casuarina sp, Phragmitis australis, Schoenoplectus sp and Sporobolus
virginicus overlapped to each other. Among these species, the spectral separation is not very clear as there are overlaps
in the reflectance region.
So based on spectral properties analysis, it demonstrated how challenge it was to separate saltmarsh species from each
from a multispectral data, such as Worldview-2 or Landsat 8 OLI. However, it would be useful to investigate why
Mangrove and Casuarina species were clearly separable in the result derived from Landsat 8 OLI but not in the result
derived from Worldview-2 data.

0.7

0.6

0.5
t Mangrove
a) 0.4 Casuarina
"
C -II- Water
co
0.3 tPhragmitis
T.5
a) --Sporobolus
-411- MarshyWetland
= 0.2 PerrenialGrass
0.1

0
1 2 3 4 5 6 7 8
-0.1
Different bands of Worldview 2 data

Figure 4 : Different endmember (class) extracted from Worldview-2 data

Classification Result: Different classification algorithm for Worldview-2 data

The confusion matrix (table 3) of MLC classifier indicates that there is a high incidence of misclassification of
Mangrove and Casuarina sp from Worldview-2 data. This misclassification mostly comes from the similarity of the
spectral properties of this tree species.

Proc. of SPIE Vol. 9864 986409-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Table 3 : Different Classification algorithm for Worldview-2 data
Maximum Likelihood Classifier (MLC)
Class A B C D E F G H I Total EC UA
A 366 0 0 0 1 3 0 0 3 373 1.88 98.1
B 2 130 0 0 0 0 0 0 0 132 1.52 98.4
C 0 0 286 2 0 6 0 0 0 294 2.72 97.2
D 0 0 8 129 0 0 2 0 2 141 8.51 91.4
E 0 0 0 4 248 0 0 3 58 313 20.7 79.2
F 1 0 4 0 0 128 0 0 0 133 3.76 96.2
G 0 0 6 5 1 1 8 1 0 22 63.6 36.3
H 0 0 1 0 0 0 0 127 0 133 0.78 99.2
I 0 1 0 0 58 1 0 1 256 317 19.2 80.76
Total 369 131 305 140 308 139 10 132 319 1853
EO 0.81 0.76 6.23 7.86 19.48 7.91 20 3.79 19.75 OA=90.55%
PA 99.19 99.24 93.77 92.14 80.52 92.09 80 96.21 80.25 Kappa =0.88
Support Vector Machine (SVM)
Class A B C D E F G H I Total EC UA
A 369 0 0 0 1 7 0 0 4 298 3.15 96.85
B 0 131 1 0 0 0 0 0 0 332 0.76 99.24
C 0 0 298 3 0 1 1 0 0 303 1.65 98.35
D 0 0 4 131 1 0 6 0 1 143 8.39 91.61
E 0 0 0 4 247 0 1 2 44 298 17.11 82.89
F 0 0 2 0 0 131 1 0 0 134 2.24 97.76
G 0 0 0 0 0 0 0 0 0 0 0 100
H 0 0 0 0 0 0 0 130 0 130 0 100
I 0 0 0 2 59 0 1 0 270 332 18.67 81.33
Total 369 131 305 140 308 139 10 132 319 1853
EO 0 0 2.30 6.43 19.81 5.76 100 1.52 15.36 OA=92.12%
PA 100 100 97.70 93.57 80.19 94.24 0 98.48 84.64 Kappa =0.91

Table 4: Artificial Neural network (ANN) algorithm for Worldview-2 data

Class A B C D E F G H I Total EC UA
A 369 0 0 0 1 5 0 0 3 373 1.88 98.1
B 0 131 0 0 0 0 0 0 0 132 1.52 98.4
C 0 0 296 5 0 0 1 0 0 294 2.72 97.2
D 0 0 0 125 1 0 4 0 1 141 8.51 91.4
E 0 0 0 7 235 0 1 1 50 313 20.7 79.2
F 0 0 0 0 0 132 0 0 0 133 3.76 96.2
G 0 0 6 0 1 1 0 0 0 22 63.6 36.3
H 0 0 1 0 0 0 0 130 0 133 0.78 99.2
I 0 0 2 3 71 1 4 1 265 317 19.2 80.76
Total 369 131 305 140 308 139 10 132 319 1853
EO 0 0 2.95 10.71 23.70 5.03 100 1.5 16.92 OA=90.83%
PA 100 100 97.04 89.28 76.29 94.96 0 98.48 83.07 Kappa =0.89
Class key: A= Water, B = Marshy Wetland, C = Phragmitis australis, D = Sporobolus virginicus, E= Mangrove, F = Ficiona
nodosa, G = Schoenoplectus sp., H = Perennial Grass, I = Casuarina sp; EO = Error of Omission, EC= Error of Commission,
UA= User Accuracy, PA= Producer Accuracy, OA = Overall Accuracy.

Another sources of classification error comes from the similarity of spectral profile of Sporobolus virginicus with
Schoenoplectus sp. similar trend was also observed in other two classifier, SVM and ANN Moreover Schoenoplectus
was the most difficult classes for which to collect ground reference data, due to relatively small size of the parcels. Based
on these three classifier, SVM showed the maximum overall accuracy among these three.

Proc. of SPIE Vol. 9864 986409-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Different classification algorithm for Landsat 8 OLI data:
On the other hand, the confusion matrix (table 5) of MLC classifier with Landsat 8 OLI data indicates that the producer
and user accuracy for Mangrove, Casuarina and Grass species are higher than others species. Although overall
classification accuracy (table 5) is lower for all classifier when it was compared with that of Worldview-2 data. But the
main findings is that the two dominant species of Mangrove and Casuarina are well classified from the Landsat 8 image.

Table 5: Artificial Neural network (ANN) algorithm for Landsat 8 data

MLC
Class A B C D E F G Total EC UA
A 54 1 0 0 0 0 7 56 1.78 98.21
B 15 65 0 9 0 0 2 91 9.89 90.10
C 0 0 52 11 0 0 0 70 15.71 84.28
D 0 0 14 39 0 0 5 58 22.62 77.37
E 1 0 0 0 81 5 0 87 5 95
F 3 1 0 0 4 79 1 87 6.15 93.85
G 0 0 8 7 0 2 63 80 10 90
Total 73 67 74 66 85 86 78 529
EO 20.54 20.37 23.52 21.60 4.70 5.81 7.69 OA=82. 04%
PA 79.45 79.62 76.47 78.39 95.29 94.18 92.30 Kappa =0.79
SVM
A 54 1 0 0 0 0 7 56 1.78 98.21
B 15 65 0 9 0 0 2 91 9.89 90.10
C 0 0 52 11 0 0 0 70 15.71 84.28
D 0 0 14 39 0 0 5 58 22.62 77.37
E 1 0 0 0 81 5 0 87 5 95
F 3 1 0 0 4 79 1 87 6.15 93.85
G 0 0 8 7 0 2 63 80 10 90
Total 73 67 74 66 85 86 78 529
EO 20.54 20.37 23.52 21.60 4.70 5.81 7.69 OA=77.33%
PA 79.45 79.62 76.47 78.39 95.29 94.18 92.30 Kappa =0.74
ANN
A 54 1 0 0 0 0 7 56 1.78 98.21
B 15 65 0 9 0 0 2 91 9.89 90.10
C 0 0 52 11 0 0 0 70 15.71 84.28
D 0 0 14 39 0 0 5 58 22.62 77.37
E 1 0 0 0 81 5 0 87 5 95
F 3 1 0 0 4 79 1 87 6.15 93.85
G 0 0 8 7 0 2 63 80 10 90
Total 73 67 74 66 85 86 78 529
EO 20.54 20.37 23.52 21.60 4.70 5.81 7.69 OA=75.23%
PA 79.45 79.62 76.47 78.39 95.29 94.18 92.30 Kappa =0.72
Class key: A= Water, B = Marshy Wetland, C = Phragmitis australis, D = Sporobolus virginicus, E= Mangrove, F = Casuarina,
G = Perennial Grass, EO = Error of Omission, EC= Error of Commission, UA= User Accuracy, PA= Producer Accuracy, OA =
Overall Accuracy

To test individual species accuracy, Mapping Accuracy (MA %) result is shown on table 7 to compare the result from
two different platform sensors.

Proc. of SPIE Vol. 9864 986409-9

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Table 6: Accuracy Assessment and Kappa statistics
Worldview-2 data MLC SVM ANN
Overall accuracy 90.55 92.12 90.83
Kappa coefficient 0.89 0.91 0.89

Landsat 8 OLI MLC SVM ANN

Overall Accuracy 82.04 77.31 75.23
Kappa coefficient 0.79 0.74 0.72

Table 7 : Mapping Accuracy assessment

Class MLC (%) SVM (%) ANN (%)
Worldview-2 data
Mangrove 66.48 67.28 64.03
Casuarina 67.36 67.64 66.41
Water 97.34 97.10 97.61
MarshyWetland 98.48 99.24 99.24
Phragmitis 91.37 94.56 95.12
Sporobolus 84.86 82.58 82.23
Scoeloplectus sp 42.10 0 0
Grass 99.21 98.48 98.48
Landsat 8 OLI data
Mangrove 89.01 79 74.76
Casuarina 84..04 70.29 81.48
Water 74.32 63.93 67.07
MarshyWetland 69.89 51.06 56.04
Phragmitis 56.52 57.77 51.61
Sporobolus 45.88 52.94 50.56
Grass 66.31 65.55 55.20

4. DISCUSSION
Our finding from parametric (MLC) with non-parametric classifiers (SVM and ANN) in Worldview-2 data are
comparable to the results given the previous studies. For example, [25] found a better performance of using SVM when
it was compared against ANN using SPOT-5. One of the limitation of our finding with SVM is 0 with Schoeloplectus sp
where MLC provided the better accuracy to delineate the narrow patch and some fragmented location of this species.
This is might be due to the training sample size. Based on [26] classification accuracy is affected by training sample size

Proc. of SPIE Vol. 9864 986409-10

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 and number of bands used as an input. Because SVM defines decision boundaries between classes using support vector
rather than statistical attributes which are sample size oriented. As a result a larger training data set has a better chance of
including support vectors that determine the actual decision boundaries and also provide higher accuracies that is barely
fulfilled by the training and validation sample size of Schoeloplectus sp.
With Worldview-2 data, MLC produces lower MA for classes those have overlapping information in the electromagnetic
spectrum, like as Mangrove and Casuarina sp. Similar trend was observed for Phragmitis australis, Sporobolus
virginicus and Ficiona nodosa. But MLC produces higher mapping accuracy for classes those have fairly homogenous
composition of the target classes at the 2X2 m spatial resolution. Because homogenous distribution results relatively
normal distribution of the reflectance values of that class in each of the spectral bands. For instance water and grass were
clearly separable and based on probabilistic parametric classifier MLC was giving the highest MA for these two classes.
Similar studies in wetland areas have also delineated water body from other wetland features [27].
On the other hand, MLC provided the maximum Overall Accuracy (82.04 %) for Landsat 8 OLI followed by SVM
(77.31 %) and ANN (75.23 %). For all classifier, Producer Accuracy and User Accuracy were higher for Mangrove,
Casuarina and grass due to their homogenous patch in the study site. These characteristics (higher spatial extent) played
an important rule to improve Producer Accuracy and User Accuracy for these three classes. The main reason is that with
coarse spatial resolution, dominant species with high spatial extent become easier to be defined. When we calculated the
mean and standard deviation from different homogenous pixels for these two species we found a very low standard
deviation. Due to this the error bar was not overlapped within first 5 bands for these two species. Contrary Worldview-2
(2m spatial resolution) data try to find out more chemical variation within visible to NIR due to more number of bands
within the same spectrum. In some extent, within 2 m spatial resolution their leaf pigment or vigour might be similar and
creates a confusion to classify it at this scale. But when it calculated the mean value within 30 m scale, it was clearly
separable by Landsat OLI data. For this reason the MA was lower for these two tree species. These findings also
supported by [28] as they got higher OA for some species when spatial resolution increased from 12 m to 20m. Our
findings from Landsat 8 OLI are also comparable with [29] when they found a positive correlation between NIR and
green bands with the spatial resolution of image. They found that when the resolution become coarser, the correlation
coefficient between two bands is supposed to increase. In similar way Herold and Roberts [30] compared simulated
coarse resolution broadband IKONOS and AVIRIS data for classifying land cover and found that IKONOS was more
appropriate than AVIIRS.

5. CONCLUSION AND RECOMMENDATION

This study examined the distribution of fragmented saltmarsh in the context of two other dominant species, Mangrove
and Casuarina. The effect of various spatial resolution to the accuracy of classifying the dominant vegetation species was
investigated. The major findings are summarized as follows:
1. Highs spatial resolution Worldview-2 data with 8 spectral bands are useful to make a map for fragmented
saltmarsh where the minimum patch size is at least 2m by 2m for each species.
2. Coarse resolution Landsat 8 OLI data works well for classifying the dominant species here when they spread
over a wide extent. This is a good indication to map mono-species mangrove at community or regional level
with Landsat 8 OLI data.
3. SVM and ANN did not work well to classify Casuarina tree species when they are in mixed pixel with grass and
Sporobolus. This was the main cause to reduce the overall accuracy for SVM and ANN.
Based on these findings several future research are very clear. Firstly, mixed pixels could be resolved with spectral
unmixing model. This will show the actual percentage of Grass, Sporobolus and Casuarina when they are in a same
pixel. Laboratory analysis can be carried for leaf pigments, Chlorophyll and other nutrients that are related with plant
reflectance. So that maximum information can be extracted from more bands of Worldview-2 data to remove
overlapping information between two species.

Proc. of SPIE Vol. 9864 986409-11

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 REFERENCES

[1] Daly, T., Prime fact: Coastal Saltmarsh,

<http://www.dpi.nsw.gov.au/__data/assets/pdf_file/0007/459628/Coastal-Saltmars Primefact.pdf> ( 17 September
2015) (2013).
[2] Zomer, R. J., Trabucco, A., Ustin, S. L., Building spectral libraries for wetlands land cover classification and
hyperspectral remote sensing, Journal of Environmental Management. 90(7), 2170-2177 (2009).
[3] Belluco, E., Camuffo, M., Ferrari, S., Modenese, L., Silvestri, S., Marani, A., & Marani, M. Mapping salt-marsh
vegetation by multispectral and hyperspectral remote sensing, Remote sensing of environment, 105(1), 54-67 (2006).
[4] Hunter, E. L., and C. H. Power. An Assessment of Two Classification Methods for Mapping Thames Estuary
Intertidal Habitats Using CASI Data, International Journal of Remote Sensing. 23 (15): 29893008, (2002).
[5] Almeida, T., & Filho, D. S. Principal component analysis applied to feature-oriented band ratios of hyperspectral
data: a tool for vegetation studies, International Journal of Remote Sensing, 25(22), 5005-5023 (2004).
[6] Markham, B., & Townshend, J. Land cover classification accuracy as a function of sensor spatial resolution,
(1981)
[7] Curran, P. J., & Atkinson, P. M. Issues of scale and optimal pixel size Spatial statistics for remote sensing,
Springer , pp. 115-133 (1999).
[8] Rogers, K., Saintilan, N., & Copeland, C. Managed retreat of saline coastal wetlands: challenges and opportunities
identified from the Hunter River Estuary, Australia, Estuaries and Coasts, 37(1), 67-78 (2014).
[9] Winning, G., & Saintilan, N. Vegetation changes in Hexham Swamp, Hunter River, New South Wales, since the
construction of floodgates in 1971, Cunninghamia, 11, 185-194 (2009).
[10] Harvey, K., & Hill, G. Vegetation mapping of a tropical freshwater swamp in the Northern Territory, Australia: a
comparison of aerial photography, Landsat TM and SPOT satellite imagery, International Journal of Remote Sensing,
22(15), 2911-2925 (2001).
[11] Yang, X. Integrated use of remote sensing and geographic information systems in riparian vegetation delineation
and mapping, International Journal of Remote Sensing, 28(2), 353-370 (2007).
[12] USGS. Landsat Mission. Retrieved from http://landsat.usgs.gov//index.php ( 8 December 2015).
[13] Exilis visual information Solution Retrieved from
http://www.exelisvis.com/IntelliEarthSolutions/GeospatialProducts/ENVI.aspx ( 15 January 2016).
[14] Xiao, J., & Moody, A. A comparison of methods for estimating fractional green vegetation cover within a desert-
to-upland transition zone in central New Mexico, USA, Remote sensing of environment, 98(2), 237-250 (2005).
[15] Ferreira, M., Ferreira, L., Sano, E., & Shimabukuro, Y. Spectral linear mixture modelling approaches for land
cover mapping of tropical savanna areas in Brazil, International Journal of Remote Sensing, 28(2), 413-429 (2007).
[16] Johnson, P.E., Smith, M.O and Adams, J.B., Simple algorithms for remote determination for mineral abundances
and particle sizes from reflectance spectra, Journal of Geophysical Research, 97, E2,2649-2657(1992).
[17] Smith, M. O., Johnson, P. E., & Adams, J. B. Quantitative determination of mineral types and abundances from
reflectance spectra using principal components analysis, Journal of Geophysical Research: Solid Earth, 90(S02)
(1985).
[18] Pal, M., & Mather, P. Support vector machines for classification in remote sensing, International Journal of
Remote Sensing, 26(5), 1007-1011 (2005).
[19] Kavzoglu, T., & Reis, S. Performance analysis of maximum likelihood and artificial neural network classifiers for
training sets with mixed pixels, GIScience & Remote Sensing, 45(3), 330-342 (2008).
[20] Story, M., & Congalton, R. G. Accuracy assessment-A user\'s perspective. Photogrammetric engineering and
remote sensing, 52(3), 397-399 (1986).
[21] Stehman, S. V. Selecting and interpreting measures of thematic classification accuracy, Remote sensing of
environment, 62(1), 77-89 (1997).
[22] Richards, J. A. Classifier performance and map accuracy, Remote sensing of environment, 57(3), 161-166 (1996).
[23] Congalton, R. G., & Green, K. Assessing the accuracy of remotely sensed data: principles and practices, CRC
press (2008).
[24] Qu, L., Han, W., Lin, H., Zhu, Y., & Zhang, L. Estimating vegetation fraction using hyperspectral pixel unmixing
method: a case study of a karst area in China. Selected Topics in Applied Earth Observations and Remote Sensing,
IEEE Journal of, 7(11), 4559-4565(2014).

Proc. of SPIE Vol. 9864 986409-12

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 [25] Song, X., Duan, Z., & Jiang, X. Comparison of artificial neural networks and support vector machine classifiers
for land cover classification in Northern China using a SPOT-5 HRG image, International Journal of Remote Sensing,
33(10), 3301-3320 (2012).
[26] Huang, C., Davis, L., & Townshend, J. An assessment of support vector machines for land cover classification,
International Journal of Remote Sensing, 23(4), 725-749 (2002).
[27] Davranche, A., Lefebvre, G., & Poulin, B. Wetland monitoring using classification trees and SPOT-5 seasonal
time series, Remote sensing of environment, 114(3), 552-562 (2010).
[28] Roth, K. L., Roberts, D. A., Dennison, P. E., Peterson, S. H., & Alonzo, M. The impact of spatial resolution on
the classification of plant species and functional types within imaging spectrometer data, Remote sensing of
environment, 171, 45-57(2015)
[29] Chen*, D., Stow, D., & Gong, P. Examining the effect of spatial resolution and texture window size on
classification accuracy: an urban environment case, International Journal of Remote Sensing, 25(11), 2177-2192
(2004).
[30] Herold, M., and Roberts, D. A. Multispectral Satellites-Imaging Spectrometry-LiDAR: Spatial-Spectral Tradeoffs
in Urban Mapping, International Journal of Geoinformatics, 2(1), 1-13 (2006).

Proc. of SPIE Vol. 9864 986409-13

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/12/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx