Available online at www.sciencedirect.

com

Advances in Space Research 48 (2011) 799805

www.elsevier.com/locate/asr

Field-based landcover classication using TerraSAR-X texture analysis

Ali Mahmoud a,, Samy Elbialy a, Biswajeet Pradhan a,b,1, Manfred Buchroithner a
a

Institute for Cartography, Faculty of Forestry, Geo and Hydro-Science, Dresden University of Technology, 01062 Dresden, Germany
b
Institute of Advanced Technology, University Putra Malaysia, 43400 Serdang, Malaysia
Received 20 December 2010; received in revised form 5 April 2011; accepted 6 April 2011
Available online 12 April 2011

Abstract
The present study aims to evaluate the eld-based approach for the classication of landcover using high-resolution SAR data.
TerraSAR-X (TSX) strip mode imagery, coupled with digital ortho-photos (DOPs) with 20 cm spatial resolution was used for landcover
classication and parcel mapping respectively. Dierent ltering and analysis techniques were applied to extract textural information
from the TSX image in order to assess the enhancement of the classication accuracy. Several attributes of parcels were derived from
the available TSX images in order to dene the most suitable parameters discriminating between dierent landcover types. Then, these
attributes were further statistically analysed in order to dene separability and thresholds between dierent landcover types. The results
showed that textural analysis resulted in high classication accuracy. Hence, this paper conrms that integrated landcover classication
using the textural information of TerraSAR-X has a high potential for landcover mapping.
2011 COSPAR. Published by Elsevier Ltd. All rights reserved.
Keywords: Landcover classication; TerraSAR-X; Field-based; Texture analysis; Remote sensing

1. Introduction
Landcover classication is probably among the most
prominent applications of remote sensing (cf. i.a. Soergel,
2010). Crop mapping at a specic time or growth stage is
of high importance for agricultural and economic applications. In some cases it is important to survey the existent
crops in order to manage possible irrigation requirements.
In other cases it is inevitable for crop-yield estimation for
cash crops (Ren et al., 2008) or for subsidies control (Blaes
et al., 2005). Remotely sensed (RS) data plays an important
role in retrieving landcover classes and to discriminate
between dierent types of crops. In recent years, the
object-based image analysis (OBIA) proved to be more ecient than pixel based classication mainly due to the availability of the high spatial resolution RS data (Al Fugara
et al., 2009; Blaschke et al., 2008).
Corresponding author. Tel.: +49 351 463 34809; fax: +49 351 463
37028.
E-mail address: ali.gis.rs@gmail.com (A. Mahmoud).
1
Tel.: +60 3 8946 8466; fax: +60 3 8656 6061.

Historically, Synthetic Aperture Radar (SAR) data

was made available for landcover classication much
later than optical RS data. As a result, more research
has been conducted and accumulated in the extraction
of features from optical data than from SAR images.
However, SAR has many advantages over the optical
data because of its ability to penetrate cloud cover and
its night sensing capabilities. In some cases the information of interest is better visible at microwave frequencies
rather than at optical ones (DellAcqua and Gamba,
2010). Radar interacts very dierently with surface features than optical data, providing information more
related to shape and structure than surface composition
(Herold and Haack, 2004). Thus, for many applications
such as disaster management or when data have to be
acquired at specic dates within a short period of
time radar systems are more valuable (Pradhan, 2010;
Pradhan and Shae, 2009).
Recently, eld-based landcover classication using different data sources as an OBIA approach has been
applied and improved the classication accuracy (Lu and
Weng, 2007). Dean and Smith (2003) used an Airborne

0273-1177/$36.00 2011 COSPAR. Published by Elsevier Ltd. All rights reserved.

doi:10.1016/j.asr.2011.04.005

800

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

Fig. 1. (a) Location map of the study area, (b) TSX image of the study area acquired 31.05.2010 with parcel boundaries, and (c) TSX image acquired
17.06.2010 with parcel boundaries.

Table 1
TerraSAR-X data applied.
Date

Sensor

Polarisation

Pass direction

Incidence angle range

Look direction

Resolution (m)

31.05.2010
17.06.2010

StripMap
StripMap

HH
HH

Ascending
Ascending

41.7643.89
29.6632.42

Right
Right

3
3

Table 2
Object features used for B and J calculations.
Mean
Standard deviation
GLCM homogeneity
GLCM contrast
GLCM dissimilarity
GLCM entropy
GLCM ang. 2nd moment

GLCM mean
GLCM standard deviation
GLCM correlation
GLDV ang. 2nd moment
GLDV entropy
GLDV mean
GLDV contrast

Thematic Mapper (ATM) imagery with 1.25 m spatial resolution and found that the parcel-based representation was
shown to be most appropriate for mapping agricultural
landcover in comparison to semi-natural areas because
agricultural landscapes have an inherent parcel structure
(Dean and Smith, 2003). There is, however, a limitation
to this eld-based approach as it is heavily dependent on
the eld boundaries which need to be acquired prior to
starting such classication. Some studies used the existing
parcels-data (digital or hard copies) or generated it for current and further studies (Wu et al., 2007) using dierent
techniques such as segmentation.
Improvement of classication accuracy has been
achieved for both multispectral optical data (Chauhan
et al., 2008) and radar imagery (Blaes et al., 2005; Waske
and Braun, 2009). More recently, TerraSAR-X (TSX)
images have been used in various studies for landcover
classication (Baghdadi et al., 2009; Breidenbach et al.,
2010; Burini et al., 2008; Mroz and Mleczko, 2008). In a
recent paper, Breidenbach et al. (2010) stated that the use

of textural parameters (Haralick et al., 1973; Liang, 2008;

Lloyd et al., 2004; Tso and Mather, 2009), object-based
classication approaches and multi-temporal data can signicantly improve the classication result of the TSX
images. In summary, the above body of literature indicates
that high-resolution TSX imagery has not been fully
exploited for landuse/cover classication yet.
Thus, the present study aims to test the eld-based
approach for classifying landcover using the TSX data by
employing texture analysis and various ltering methods
in order to maximise the extracted information from the
SAR images. First, the classication was applied to two
TSX scenes using each SAR image as a single image then
both of the two available images were used jointly. The
advantage of the analysis of single SAR images (besides
their costs) is the necessity of rapid mapping, for instance
in the case of time critical events (Soergel, 2010). In the
present study, eld boundaries were digitised using the digital ortho-photos (DOPs) with 20 cm spatial resolution.
Several attributes of parcels were derived from the available TSX imagery, then the separability and threshold
(SEaTH) method (Nussbaum and Menz, 2008) was applied
in order to dene the most suitable attributes that discriminate between dierent landcover types. Finally, these attributes were used in the classication tree.
2. Study area
The study area is located near Pirna, Saxony, Germany
(Fig. 1a). In order to apply the proposed methodology

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

801

Table 3
Separability and thresholds.
Object class combination

Separability

Omen

Threshold

1. Cereals from grass

Gama map 7_17.06.10
Mean 7_17.06.10

GLCM ang. 2nd moment (45)

GLCM ang. 2nd moment (45)

1.25
1.24

Great
Great

0.00241
0.00246

2. Cereals from maize

TSX 31.05.10
TSX 31.05.10

GLCM contrast (all dir.)

Mean

1.99
1.92

Small
Small

351.74
165.94

3. Cereals from orchards

TSX 170610
Local region 7_17.06.10
Mean3_17.06.10

GLCM dissimilarity (0)

GLCM homogeneity (all dir.)
Mean

1.94
1.95
1.87

Small
Great
Small

12.85
0.145
160.01

4. Cereals from rape

TSX 17.06.10
Frost 3_17.06.10

GLCM dissimilarity (0)

GLCM homogeneity (0)

1.98
1.98

Small
Great

13.25
0.103

5. Grass from maize

TSX 31.05.10
TSX 31.05.10
TSX 31.05.10

GLCM contrast (90)

Mean
GLCM homogeneity (135)

1.99
1.96
2.00

Small
Small
Great

259.66
164.16
0.0603

6. Grass from orchards

TSX 31.05.10
TSX 31.05.10

GLDV ang. 2nd moment (90)

GLCM homogeneity (135)

1.96
1.99

Small
Great

0.0398
0.0620

7. Grass from rape

TSX 31.05.10
TSX 31.05.10
Adaptive median 7_17.06.10

Mean
GLCM dissimilarity (all dir.)
GLCM homogeneity (0)

1.99
1.99
1.77

Great
Small
Great

167.49
13.65
0.1686

8. Maize from orchards

Median 3_31.05.10
Median 3_31.05.10
TSX 31.05.10 TSX 17.06.10

GLCM ang. 2nd moment (0)

GLCM entropy (0)
Mean

1.84
1.92
1.99

Small
Great
Great

0.000679
7.699
160.80

9. Maize from rape

Adaptive median 3_17.06.10
Local region 3_17.06.10
TSX31.05.10 TSX17.06.10

GLCM homogeneity (45)

GLDV ang. 2nd moment (45)
Mean

1.99
1.90
2.00

Small
Great
Great

0.0855
0.0313
153.77

10. Orchards from rape

Median 7_31.05.10
Median 7_31.05.10
TSX 31.05.10

GLCM dissimilarity (0)

GLDV ang. 2nd moment (0)
Mean

1.65
1.64
1.35

Small
Great
Small

3.146
0.1336
178.28

eciently for landcover (crop) mapping, an agricultural

area was chosen (Fig. 1b and c, demarcating the parcels
in red). During the acquisition of the TerraSAR-X data
the following landcover classes were mapped in the eld:
crops, orchards and grass.
3. Methodology
3.1. Data preparation and segmentation
TerraSAR-X images acquired on 31.05.2010 and
17.06.2010 (Table 1) were imported into ERDAS Imagine
and subsequently lter types with three dierent kernel
sizes (3  3, 5  5 and 7  7) were applied. On the other
hand, the eld boundaries were digitised from the digital
ortho-photos (2005) and updated from the TerraSAR
images to match the current landcover boundaries. Deniens 7 Software supports dierent segmentation algorithms

which are used to subdivide the entire image represented by

the pixel level domain or specic image objects from other
domains into smaller image objects. In the current study,
the multi-resolution segmentation algorithms were applied
using the eld-boundaries thematic layer coupled with the
used images. The image layers were weighted dierently
depending on their importance or suitability for the segmentation result. The higher the weight, the more of its
information is used during the segmentation process.
Therefore, the thematic layer weight was assigned to 1
while the image weight was assigned to zero (Deniens,
2007).
3.2. Feature extraction and analysis
Representative samples for each landcover type were
selected. Then a total of 62 texture feature values (Table 2)
for each image were exported as .csv les. This step was

802

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

applied to the original and the ltered TerraSAR-X

images. Additionally, an output layer was created by subtracting both images acquired on 31.05.2010 and
17.06.2010, and then used as additional layer to separate
maize from other crops. In order to dene the proper features that discriminate between dierent landcover types,
the separability was calculated for each class-pair of landcover types. The Bhattacharya Distance B (Eq. (1)) and
the JeriesMatusita Distance J (Eq. (2)) were calculated
(Nussbaum and Menz, 2008). Then the threshold T for
each pair of the classes was calculated using the Eq. (3).
An excel model was built to calculate these parameters
automatically from the *.csv les using Visual Basic Application (VBA), in order to nd the best features showing the
highest J value.
 2

1
2
1
r1 r22
2
B m1  m2 2
ln
;
1
8
2r1 r2
r1 r21 2
Fig. 2. Methodological owchart adopted in this study.

(a )

J 21  eB ;

GLDV Ang. 2nd moment (0)

2

TSX_310510

Separability

1.5

Adaptive Median 3
1

Local Region 3
Local Region 7

0.5

Gamma Map7
Median 7

(b )

GLCM Homogeneity (0)

2

Separability

1.5

TSX image
Adaptive Median 3
Adaptive Median 7

Frost 3
0.5

Frost 5
Local Region 7

Fig. 3. Eect of lter type on separability value: (a) image acquired 31.05.2010; (b) image acquired 17.06.2010.

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

803

Fig. 4. Landcover map of the study area.

x12

m2 r21  m1 r22  r1 r2



r 1 n2
A log

:
r 2 n1

q
m1  m2 2 2Ar21  r22

r21  r22

Table 4
Overall classication accuracy statistics.

Classied data

3
4

3.3. Classication
Depending on the J values (Table 3) the proper images
and relative texture features for classication were dened
for each class. Then the class description was assigned
using the membership functions that oer a transparent
relationship between feature values and the degree of
membership to a class. The process tree employed as a
class-by-class analysis that enabled to exclude the classied
objects from the classication process. Generally speaking
the classication algorithm uses class descriptions to classify the image objects by evaluating the class description
and determines whether an image object can be a member
of this class or not. Moreover, it allows a fuzzy-logic
combination of dierent features (Deniens, 2007). Fig. 2
shows the methodology applied in this study in the form
of a ow-chart.

Reference data
Cereals

Maize

Rape

Grass

Orchards

Sum.

Cereals
Maize
Rape
Grass
Orchards
Sum.

30
0
0
0
1
31

0
5
0
0
0
5

0
0
5
0
0
5

1
0
0
4
0
5

1
0
0
0
20
21

32
5
5
4
21
67

Accuracy
Producer
User
Kappa
khat
Overall accuracy

97%
94%
88%
93.27%
95.52%

100%
100%
100%

100%
100%
100%

80%
100%
100%

95%
95%
93%

4. Results and discussion

The results show that some texture features of the original TerraSAR-X images (without ltering) produced high
values of separability (J) for some class-pairs such as
cerealsmaize and grassmaize. However, specic lters
have improved the separability for other classes such as

804

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

cerealsgrass. The best features with highest J value are

shown in Table 3.
Two texture features, namely GLCM Homogeneity (0)
for the image acquired on 17.06.2010 and GLDV Angular
2nd moment (0) for the image from 31.05.2010 were chosen as an example to explain the variation in separability
with reference to the lter type (Fig. 3).
As shown in Fig. 3(a) the median seven lter improved
the separability value for the class-pairs; cereals-rape and
orchards-rape. On the other hand, Fig. 3(b) shows that,
the adaptive median seven improved the separability of
the class-pair grass-rape.
The classication result shows that the test site is covered by 5 landcover types as depicted in Fig. 4. Some classes can be separated and classied accurately using one
image of the used TSX images only, while other classes
require a combination of both images to increase the
classication accuracy. Moreover, the separability values
of maize-orchards and maize-rape were signicantly improved by using the output layer of subtracting both images
acquired on 31.05.2010 and 17.06.2010 (Table 3).
Further, an accuracy assessment after (Jensen, 2005)
was applied. For that purpose, the study area was investigated during the acquisition time/periods where samples
were selected for classication and reference elds for accuracy assessment. As shown in Table 4 a total number of 67
reference elds were used for accuracy assessment which
resulted in an overall accuracy of 95.52%. Moreover, the
kappa coecient of agreement (Khat) was calculated which
is a measure of agreement or accuracy between the remote
sensing-derived classication map and the reference data as
indicated by (a) the major diagonal and (b) the chance
agreement, which is indicated by the row and column
totals. Finally kappa coecient of 93.27% was achieved.
5. Concluding remarks
For the investigated agricultural region the eld-based
landcover classication method proved high potential of
using single TerraSAR-X image for some specic landcover
types, while the combination of images of two acquisition
dates resulted in a landcover classication with even higher
accuracy. Applying a texture-based analysis an overall
accuracy of 95% with kappa coecient 93% could be
reached. The proposed approach allows to study larger
areas with only a few number of images and within a short
period of time. Moreover, a combination of dierent lter
types improved the separability value for some class-pairs,
thus leading to higher classication accuracy. The ndings
of the study serve as a testimony of the applicability of high
resolution TerraSAR-X data for landcover mapping at a
detailed scale.
Acknowledgements
The German Aerospace Center (DLR) provided TerraSAR-X data under the Science proposal ID: HYD0326

(Th. Hahmann, M.F. Buchroithner). Ali Mahmoud and

Samy Elbialy thank the Cultural Aairs & Mission Sector,
Ministry of Higher Education, Egypt, for awarding PhD
scholarships.
References
Al Fugara, A.M., Pradhan, B., Mohamed, T.A. Improvement of land-use
classication using object-oriented and fuzzy logic approach. Appl.
Geomat. 1, 111120, 2009.
Baghdadi, N., Boyer, N., Todoro, P., El Hajj, M., Begue, A. Potential of
SAR sensors TerraSAR-X, ASAR/ENVISAT and PALSAR/ALOS
for monitoring sugarcane crops on Reunion Island. Remote Sens.
Environ. 113 (8), 17241738, 2009.
Blaes, X., Vanhalle, L., Defourny, P. Eciency of crop identication
based on optical and SAR image time series. Remote Sens. Environ. 96
(34), 352365, 2005.
Blaschke, T., Lang, S., Hay, G.J. (Eds.). Object-Based Image Analysis
Spatial Concepts for Knowledge-Driven Remote Sensing Applications, rst ed Springer, New York, 2008.
Breidenbach, J., Ortiz, S.M., Reich, M. Forest monitoring with TerraSAR-X: rst results. Eur. J. Forest Res. 129 (5), 813823, 2010.
Burini, A., Putignano, C., Del Frate, F., Licciardi, G., Pratola, C.,
Schiavon, G., Solimini, D. TerraSAR-X/SPOT-5 fused images for
supervised land cover classication. Int. Geosci. Remote Sens. 5, V373V-376, 2008.
Chauhan, H.J., Arora, M.K., Agarwal, A. Estimating land cover
class area from remote sensing classication. J. Appl. Remote Sens.
2, 2008.
Dean, A.M., Smith, G.M. An evaluation of per-parcel land cover mapping
using maximum likelihood class probabilities. Int. J. Remote Sens. 24
(14), 29052920, 2003.
Deniens, A.G. Developer 7 User guide. URL: <www.deniens.com>.
DellAcqua, F., Gamba, P. Rapid mapping using airborne and satellite
SAR images, in: Soergel, U. (Ed.), Radar Remote Sensing of Urban
Areas. Springer, New York, 2010.
Haralick, R.M., Shanmuga, K., Dinstein, I. Textural features for
image classication. IEEE Trans. Syst. Man Cyb. SMC3 (6), 610
621, 1973.
Herold, N.D., Haack, B.N. Solomon Elizabeth an evaluation of radar
texture for land use/cover extraction in varied landscapes. Int. J. Appl.
Earth Obs. 5, 113128, 2004.
Jensen, J.R. (Ed.). Introductory Digital Image Processing: A Remote
Sensing Perspective, third ed Prentice Hall, Upper Saddle River, NJ,
2005.
Liang, S. (Ed.). Advances in Land Remote Sensing System. Modeling
Inversion and Application. Springer, Dordrecht, 2008.
Lloyd, C.D., Berberoglu, S., Curran, P.J., Atkinson, P.M. A comparison
of texture measures for the per-eld classication of Mediterranean
land cover. Int. J. Remote Sens. 25 (19), 39433965, 2004.
Lu, D., Weng, Q. A survey of image classication methods and techniques
for improving classication performance. Int. J. Remote Sens. 28 (5),
823870, 2007.
Mroz, M., Mleczko, M. Potential of Terrasar-X stripmap data in early
and rapid agricultural crops mapping, in: TerraSAR-X Science Team
Meeting, DLR Oberpfaenhofen, Germany, 2008.
Nussbaum, S., Menz, G. (Eds.), Object-Based Image Analysis and Treaty
Verication: New Approaches in Remote Sensing Applied to
Nuclear Facilities in Iran. Springer Science+Business Media B.V.,
Dordrecht, 2008.
Pradhan, B. Flood susceptible mapping and risk area estimation using
logistic regression, GIS and remote sensing. J. Spat. Hydrol. 9 (2),
118, 2010.
Pradhan, B., Shae, M. Flood hazard assessment for cloud prone
rainy areas in a typical tropical environment. Disaster Adv. 2 (2), 7
15, 2009.

A. Mahmoud et al. / Advances in Space Research 48 (2011) 799805

Ren, J.Q., Chen, Z.X., Zhou, Q.B., Tang, H.J. Regional yield estimation
for winter wheat with MODIS-NDVI data in Shandong, China. Int. J.
Appl. Earth Obs. 10 (4), 403413, 2008.
Soergel, U. Review of radar remote sensing on Urban Areas, in: Radar
Remote Sensing of Urban Areas. Springer, New York, 2010.
Tso, B., Mather, P.M. (Eds.), Classication Methods for Remotely Sensed
Data, second ed., CRC Press, Boca Raton, 2009.

805

Waske, B., Braun, M. Classier ensembles for land cover mapping using
multitemporal SAR imagery. ISPRS J. Photogram 64 (5), 450457,
2009.
Wu, S., Silvan-Cardenas, J., Wang, L. Per-eld urban land use classication based on tax parcel boundaries. Int. J. Remote Sens. 28 (12),
27772800, 2007.