Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/301199496

Applicability and Limitations of Land Use / Land Cover Classification Using


High Resolution Satellite Imagery in Arid and Semi-Arid Areas of the
Northern Kordofan State (Sudan)

Conference Paper · September 2005

CITATIONS READS

2 430

1 author:

Mohamed Salih Dafalla Mohamed


University of Khartoum
56 PUBLICATIONS   70 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Soil fertility View project

‫ ﺗﻘﻴﻴﻢ اﻻﺳﺘﺰراع اﻟﺴﻤﻜﻰ واﻟﻘﻴﻤﺔ اﻟﻐﺬاﺋﻴﺔ ﻟﻠﺴﻤﻚ اﻟﻤﻨﺘﺞ ﻓﻰ وﻻﻳﺔ اﻟﺨﺮﻃﻮم‬View project

All content following this page was uploaded by Mohamed Salih Dafalla Mohamed on 11 April 2016.

The user has requested enhancement of the downloaded file.


Applicability and Limitations of Land Use / Land Cover Classification
Using High Resolution Satellite Imagery in Arid and Semi-Arid Areas of
the Northern Kordofan State (Sudan)
Mohamed Salih Dafalla
University of Khartoum, Department of Soil Science and Environmental Studies, presently at
Department of Geosciences, University of Dresden
Elmar Csaplovics
Department of Geosciences, University of Dresden

Abstract
Sudan as the largest African country extends over a variety of eco-climatic zones, ranging from desert at north to the
savanna zone in the south. Being one of the Sahelian countries it faced numerous drought periods. Semi-arid regions of
the Sudan are thus heavily attacked by severe desertification, which is driven by climatic as well as human impacts.
Spatial data on dynamics of land use and land cover is poor and thus insufficient. Nevertheless extended knowledge on
state and changes of land use and land cover is needed in order to support the implementation of sustainable strategies
of regional (re)development. The main objective of this study is to discuss the applicability of sources and methods of
remote sensing for reliable classification of land use / land cover (LUC) in terms of the specifications set up by FAO,
UNCCD and others. The area of the case study is located in the Sahelian eco-climatic zone in Northern Kordofan State,
Sudan. A geometrically and radiometrically corrected sub-scene of Landsat ETM+ imagery (174/51) of Nov.27th 1999
was used in this study. Moreover, ancillary data and interviews recorded during field work were used to support image
classification. Applying image analysis and statistical software allowed for producing colour composites used for on-
screen and in-situ mapping as well as for calculating vegetation indices and other band ratios and finally for classifying
imagery by means of clustering algorithms. The study area was divided into five regions; unsupervised classification
was carried out for these regions to produce 11 classes for each region, only 9 signatures were selected. These
signatures were labeled to their corresponding land use / land cover types. This study supports the increasing activities
in integrating systematical and periodical monitoring of regional dynamics of land use and land cover by means of
operationalised low-cost and easy-to-handle tools of remote sensing and geoinformation analysis on national and
regional levels in the Sudan.

Keywords: Sudan, Sahelian zone, Northern Kordofan State, Land use / land cover, Unsupervised / supervised
classification.

Introduction
Sudan is the largest African country covering an area of approximately 2.6 million km2 and inhabited by a population
of about 29.4 millions. The country extends over a variety of eco-climatic zones, ranging from desert at north with nil
annual rainfall to the savanna zone in the south with up to 1200 mm annual rainfall (Danida, 1989). Being one of the
Sahelian countries it faced numerous drought periods, especially during the 1960s and 1980s. Severe famine and large
immigration movements occurred. Semi-arid regions of the Sudan are thus heavily attacked by severe desertification,
which is driven by climatic as well as human impacts (UNSO, 1992).
Sudan is considered one of the poorest countries in world (Hinderson, 2004). Besides all kinds of data supply,
especially spatial data on dynamics of land use and land cover is poor and thus insufficient. Nevertheless extended
knowledge on state and changes of land use and land cover is needed in order to support the implementation of
sustainable strategies of regional (re)development (Hielkema et. al., 1986; IFAD, 2004).
The difficulties concerning land use / land cover classification in semi-arid regions by means of remote sensing are well
known. Since vegetation-soil-patterns in arid and semi-arid zones respectively are characterized by a sparse distribution
of often non-photo-synthesising vegetation (NPV) its spectral behaviour often interferes with spectral signatures of bare
soil patterns (Schmidt and Karnieli, 2000). Moreover, the spatial heterogeneity at pixel level is strongly affecting
systematic separation between dominant land uses. Therefore, there are many studies recommending subpixel unmixing
analysis as suitable method to overcome such constrains (Elmore, et al., 2000), but still there are many difficulties to be
overcome such as unavailability of spectral libraries for dominant plant species and soil types.

1
Objectives
The main objective of this study is to discuss the applicability of sources and methods of remote sensing for reliable
classification of land use / land cover (LUC) in terms of the specifications set up by FAO, UNCCD and others.

Methodology
The area of the case study is located in the Sahelian eco-climatic zone in Northern Kordofan State, Sudan. These
regions generally provide reasonable harvests of rainfed crops as sesame, millet and hibiscus, of gum arabic (Acacia
Senegal) and plenty of livestock such as sheep and camels. Severe constraints for the development of medium- to long-
term strategies of sustainable land management are raised by temporal variations of impacts of drought and
desertification during the last decennia (Hielkema et. al., 1986).
A geometrically and radiometrically corrected sub-scene of Landsat ETM+ imagery (174/51) of Nov.27th 1999 was
used in this study (Fig. 1). Moreover, ancillary data of the relevant topographic map sheets 1:250000, a land use map, a
vegetation map and interviews recorded during field work were used to support image classification.
Applying image analysis and statistical software allowed for producing colour composites used for on-screen and in-
situ mapping as well as for calculating vegetation indices and other band ratios and finally for classifying imagery by
means of clustering algorithms.
The study area was divided into five regions (R) (Fig. 1); unsupervised classification was carried out for these regions
to produce 11 classes based on previous study have been carried out by the Northern Kordofan Rangeland Development
Programme (NKRDP) (Personal communication) for each region using visible bands 1, 2, 3 in addition to NIR and
SWIR bands 4, 5, and 7. As result of thematic similarity among these 5x11 spectral classes, only 9 signatures were
selected after evaluation of the produced spectral classes selected by visual interpretation and by analysis of spectral
signatures and scatter plots (Appendices A, B, C, D, F). These 9 signatures were labeled to their corresponding land use
/ land cover types. Finally a supervised classification using Mahalanobis Distance Classifier was applied to produce the
classified image. This classification approach was adopted since it allowed for the detection of a variety of additional
spectral classes within the subset, thus for a better representation of its heterogeneity.

Fig. 1: ETM+ Subscene of the study area

R2

R1 R3 R5

R4

Result
Despite the above mentioned limitations the overall classification accuracy meets values of around 75%. The
dominating land use / land cover classes in the region of interest are small-scale traditional agriculture on sand (23.08),
open shrubland (19.27%), fallow land (14.75%), grassland (12.47%), open forest (11.61%), shrubland on clay

2
(10.25%), closed shrubland (4.15%), mixed forest and shrubland (3.99%) and natural water surfaces (0.39%) (Table 1,
Fig. 2 and Fig. 3). Apart statistical uncertainties full-area coverage of reliable classified data was mainly limited by the
structural heterogeneity of the areas, due to typical vegetation continua in grasslands of varying densities as well as by
the building materials in rural settlements, which comprise wood and grasses cut in the local environment. However,
the preliminary results of the study are promising as they evidently provide more detailed spatial information on
regional patterns of land use and land cover than up to now.

Table 1: Main dominant LULC classes

Class Name Area (Ha) %


Traditional Agriculture on Sand 285240 23.0883
Open Shrubland 238090.5 19.27186
FallowLand 182292.8 14.7554
Grassland 154052.8 12.46956
Open Forest 143441.2 11.61062
Shrubland on Clay 126691.2 10.25482
Closed Shrubland 51329.33 4.154771
Mixed Forest and Shrubland 49351.9 3.994712
Natural Water Bodies 4941.242 0.399961

Fig 2: Main dominant LULC classes

3
300000 25

250000
20

200000

15
Area (ha)

150000

%
10

100000

5
50000

0 0
Traditional Open FallowLand Grassland Open Forest Shrubland on Closed Mixed Forest Natural Water
Agriculture on Shrubland Clay Shrubland and Shrubland Bodies
Sand
Class name

Area (Ha) %

Fig. 3: Main dominant LULC classes areas and percentages.

Conclusion
This study, as part of comprehensive LULC change research in Northern Kordofan State, supports the increasing
activities in integrating systematical and periodical monitoring of regional dynamics of land use and land cover by
means of operationalised low-cost and easy-to-handle tools of remote sensing and geoinformation analysis on national
and regional levels in the Sudan.
Reference
Elmore, A.J., J.F. Mustard, S. Manning, and D. Lobell. 2000. Quantifying percent live cover in multitemporal data of a
semi-arid region: Comparison between spectral mixture analysis and NDVI. Remote Sensing of Environment 73,
87-102.
Danida, 1989. Enivronmental Profile: Sudan. Danida, Ministry of Foreign Affairs. Sect1: National Overview, 8-15.
Hielkema, J.U.; S.D. Prince and W.L. Astle, 1986. Rainfall and vegetation monitoring in the savanna zone of the
Democratic Republic of Sudan using the NOAA Advanced Very High Resolution Radiometer. Int. J. Reomte
Sensing, 1986, Vol. 7, No. 11, 1499-1513.
Hinderson, T. 2004. Analysing Environmental Change in Semi-arid Areas in Kordofan, Sudan. Seminar Series nr 109.
Geobiosphere Science Centre, Physical Geography and Ecosystem Analysis, Lund University.
International Fund for Agricultural Development (IFAD), 2004. Environmental Assessment Study. Main Report.
Republic of Sudan, Western Sudan Resource
Northern Kordofan Rangeland Development Programme (NKRDP). 2004. Unpublished Report. Personal
communication, Mr. Abdelrahman Ahmed Khatir. Agricultural Research Corportation, Elobeid, Sudan.
Schmidt, H. and A. Karnieli. 2000. Remote sensing of the seasonal variability of vegetation in a semi-arid environment.
Journal of Arid Environments (2000) 45: 43–59. Available online at http://www.idealibrary.com
Management Programme. Part 1: Greater Kordofan. Near East and North Africa Division Project Management
Department.
United Nations Sudano-Sahelian Office (UNSO), 1992. Assessment of Desertification and Drought in the Sudan-
Sahelian Region 1985-1991.

4
Appendices
Appendix A: Descriptive statistics of six bands

Mean Std. N
Deviation
MEAN1 60.2541 3.9586 55
MEAN2 59.4310 7.7080 55
MEAN3 80.3925 18.9569 55
MEAN4 76.8581 17.7228 55
MEAN5 100.6771 29.4468 55
MEAN6 76.8979 25.4968 55

Appendix B: Pearson correlation coefficients between means of six bands for 55 spectral classes

MEAN1 MEAN2 MEAN3 MEAN4 MEAN5 MEAN6


MEAN1 1.000 0.911 0.768 0.586 0.623 0.612
MEAN2 0.911 1.000 0.926 0.714 0.714 0.745
MEAN3 0.768 0.926 1.000 0.906 0.899 0.930
MEAN4 0.586 0.714 0.906 1.000 0.986 0.976
MEAN5 0.623 0.714 0.899 0.986 1.000 0.985
MEAN6 0.612 0.745 0.930 0.976 0.985 1.000
Appendix C: Scatter diagram for bands 1, 3 and 4 for 55 spectral classes

120

DN100

80
MEAN4
60

34
40

70 140
120
60 100 DN
80
50 60
MEAN1 DN 40
MEAN3

5
Appendix D: Scatter diagram for bands 1, 3 and 6 for 55 spectral classes

140

120

DN100

80
MEAN6
60

40

20

70 140
120
60 100 DN
80
50 60
DN 40
MEAN1 MEAN3

Appendix F: Frequency distribution of mean DN values of all bands for 55 classes.

20 12

10

8
Frequency

Frequency

10 6

Std. Dev = 3,96 2 Std. Dev = 7,71


Mean = 60,3 Mean = 59,4
0 N = 55,00 N = 55,00
0
46,0 50,0 54,0 58,0 62,0 66,0 37,5 42,5 47,5 52,5 57,5 62,5 67,5 72,5
48,0 52,0 56,0 60,0 64,0 68,0 40,0 45,0 50,0 55,0 60,0 65,0 70,0
DN DN
MEAN1 MEAN2

8 10

8
6
Frequency
Frequency

2
2
Std. Dev = 18,96
Std. Dev = 17.72
Mean = 80,4
Mean = 76.9
0 N = 55,00 N = 55.00
0
35,0 45,0 55,0 65,0 75,0 85,0 95,0 105,0 115,0 25.0 35.0 45.0 55.0 65.0 75.0 85.0 95.0 105.0
40,0 50,0 60,0 70,0
DN
80,0 90,0 100,0 110,0
6 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 110.0
DN
MEAN3 MEAN4
10 12

10
8

Frequency
Frequency

4
4

2
Std. Dev = 29,45 2 Std. Dev = 25,50
Mean = 100,7 Mean = 76,9

0 N = 55,00 0 N = 55,00
10,0 30,0 50,0 70,0 90,0 110,0 130,0 10,0 30,0 50,0 70,0 90,0 110,0
20,0 40,0 60,0 80,0 100,0 120,0 140,0 20,0 40,0 60,0 80,0 100,0 120,0
DN DN
MEAN5 MEAN6

MEAN4
160
MEAN5

MEAN1
140
MEAN6

120 MEAN1
MEAN5

100 MEAN1
MEAN4
DN

80 MEAN1
MEAN3
60
MEAN2
MEAN3
40
MEAN1
MEAN2
20
MEAN5
0 MEAN6
0 20 40 60 80 100 120 140 160
DN

View publication stats

You might also like