2004 
— 
cal and 
Saddle 
Sensing: 
M. Sc, 
» 97 pgs. 
LAND USE MAPPING USING ETM+ DATA 
(CASE STUDY:CHAMESTAN AREA, IRAN) 
Seyed Zeynalabedin Hossein', Sayed Jamaleddin Khajeddin”, 
Hossein Azarnivad', Seyed Ali Khalilpour' 
I-College of Natural Resources & Desert Studies, University of Yazd, P.O.Box 89195-741 Yazd, Iran 
TeleFax:+98351-8210312 , E-mail: hosseini_sz@yahoo.com 
2- College of Natural Resources, Isfahan University of Technology 
3- College of Natural Resources, University of Tehran 
4- Iranian Forests and Rangelands Organization 
KEYWORDS: ETM’, Land use, Mapping, Image enhancement, Best band set, Classification, Vegetation indices. 
e 
ABSTRACT: 
Land use maps are useful tools for agricultural and natural resources studies as a base data. Due to dynamism of natural resources, 
updating these maps is essential. Employing traditional methods through aerial photos interpretation to produce such maps are costly 
and time consuming. Satellite data is suitable for such purpose, as a consequence of its fast repeatability, wide and unique view and 
availability of data from most part of electromagnetic spectrum. The present study is conducted to investigate the capability of ETM' 
data on land use mapping of Chamestan region, Mazandaran, Iran. The studied area was 67000 ha. Image of 18th July, 2000 were 
registered to 1:25000 digital topographic maps. Images were enhanced using contrast enhancement, False Color Composite (FCC), 
Principal Component Analysis (PCA), Tasseled cap transformation and vegetation indices. The Optimized Index Factor (OIF) and 
correlation technique were employed to determine the best band sets for FCC and consequently for classification analysis. 
Unsupervised (clustering) and supervised (maximum likelihood, minimum distance and parallelepiped classifiers) classification 
methods were used. Finally Hierarchical method was applied to increase maps accuracies. The results showed that contrast 
enhancement, FCC, PCA and Tasseled cap have effective role in features enhancement. Using the best bands set (156H) caused to 
highest accuracy in classification. In supervise classification, overall accuracy and Kappa coefficient for maximum likelihood 
classifier were estimate 85,83% and 62,81% respectively, for minimum distance method 73,7794 and 47.12% and for parallelepiped 
34,27% and 19,03%. The highest overall accuracy and Kappa coefficient related Hierrarchical method is 94% and 84.89%. 
INTRODUCTION 
Land use maps are useful tools for agricultural and natural 
resources studies as a base data. Due to dynamism of natural 
resources, land use map updating is essential. Traditional 
methods utilization through aerial photos interpretation to 
produce such maps is costly and time consuming. Satellite data 
is suitable for such purpose, as a consequence of its fast 
repeatability, wide and unique view and availability of data 
from most part of electromagnetic spectrum. 
Land cover map of the tropical forest rehabilitation was 
produced applying NDVI to TM data (Apan et al 1997). Also 
land use map of Mouk area, Iran was produced using Landsat 
data and GIS (Alavipanah et al 2001). Gomarasca (1993) 
assessed land use changes in the metropolitan area of Milan 
(Italy) employing maximum likelihood classification of TM 
data, aerial photo and topographic maps. Zahedifard (2002) 
produced land use map of Bazoft area, Iran, by means of TM 
data. She used maximum likelihood, minimum distance and 
parallelepiped algorithms for supervised classification. In 
addition hybrid method and GIS was implemented to improve 
the classification accuracy. 
MATERIAL AND METHODS 
The studied area is located in Mazandaran province, northern 
Iran, between 36° 14 to 36° 30 Longitude and 52° 0 to 52° 15 
Latitude. Mean annual precipitation is 1200mm. The studied 
area was about 67000ha. Landsat ETM" images of 18th July, 
2000 were georeferenced to 1:25000 digital topographic maps 
by nearest neighbor resampling algorithm and the RMSe was 
39] 
less than one pixel. Images were enhanced using contrast 
enhancement, False Color Composite (FCC), Principal 
Component Analysis (PCA), Tasseled cap transformation and 
vegetation indices. The Optimized Index Factor (OIF) and 
correlation technique were employed to determine the best band 
sets for FCC image and consequently for classification analysis. 
Unsupervised (clustering) and supervised (maximum likelihood, 
minimum distance and parallelepiped classifiers) classification 
methods were used. Finally Hierarchical method was applied to 
increase maps accuracy. 
Hierarchical method: Assessing the supervised classification 
results has proved the high spectral similarities between 
rangelands and dry farmlands. Where the rangelands are located 
in south and the wheat dry farmlands are located in the north 
part of the studied areas. This similarity causes a low accuracy 
on classification results. The below mentioned hierarchical 
method was used to solve this problem. First of all, the 
rangelands and the dry farms were masked, and they were 
classified separately. Then the rest of the image was classified 
to the other features such as forest, rice fields, urban area and 
etc. Finally the whole classified land uses were integrated to 
develop the final map. Ground truth points were produced 
through field surveying with GPS as well as using topographic 
maps. 
RESULTS AND DISCUSSION 
The Landsat ETM" image was classified to the nine different 
land uses (Table 1).