M R Mustapha
Student
Universiti Sains Malaysia, Malaysia
[email protected]
H S Lim
Lecturer
Universiti Sains Malaysia, Malaysia
[email protected]
M Z Mat Jafri
Lecturer
Universiti Sains Malaysia, Malaysia
[email protected]
N Othman
Lecturer
Universiti Sains Malaysia, Malaysia
ABSTRACT
Land cover classification from remotely sensed image is an important research and widely used in remote sensing application. Satellite imagery from Landsat 7 ETM+ provide land cover information with high temporal frequency and spatial for land cover mapping over Makkah, Mina and Arafah, Saudi Arabia. This paper presents land cover mapping technique using remote sensing data for land cover features assessment of the study area. Three supervised classification techniques Maximum Likelihood, Minimum Distance-to-mean and Parallelepiped were applied to the imageries to extract the spectral information from acquired scene. Training sites were selected within each scene and land cover classes were assigned to each classifier. Accuracy assessment was performed in this study to determine the quality of the land cover map. The maximum likelihood classifier produced superior result based on its accuracy in this study. This study clearly classified the land cover features using the multispectral classification technique for urban planning and development purposes.
Introduction
The information classification and the mapping of the land use are very important. The land use has to be planned and determined by all levels of the urbanization for future planning. A major advantage of the remotely sensed satellite images is to monitor changes on the earth’s surface or update an existing data set. The increasing availability of remote sensing images, acquired periodically by satellite sensors on the same geographical area, makes it extremely interesting to develop the monitoring systems capable of automatically producing and regularly updating land cover maps of the considered site (Bruzzone, et al., 2002). Better assessment of the changes of land cover by using digital analysis of remote sensing satellite data can help decision makers to develop effective plans for the management of land (Gordon, 1980; Milington et al., 1986; Franchek and Biggam, 1992).
As Clawson and Stewart (1965) have stated: In this dynamic situation, accurate, meaningful, current data on land use are essential. If public agencies and private organizations are to know what is happening, and are to make sound plans for their own future action, then reliable information is critical. The variety of land use and land cover data needs is exceedingly broad. Many Federal agencies need current comprehensive inventories of existing activities on public lands combined with the existing and changing uses of adjacent private lands to improve the management of public lands. Federal agencies also need land use data to assess the environmental impact resulting from the development of energy resources, to make national summaries of land use patterns and changes for national policy formulation, and to prepare environmental impact statements and assess future impacts on environmental quality.
The objective of this study is to evaluate the accuracy of each classification technique to classify the land use during the years 2003 using three types of classification namely, Maximum Likelihood classification, Minimum Distance classification, and Parallelepiped classification. The observed land cover classification will be analyzed in this. Therefore, to maintain harmony among sustainable resources and socio-economic needs, land cover and land use studies should be dealt with care.
Case Study in the Kingdom of Saudi Arabia
The study’s site covered a rectangle of centered on Makkah, Mina and Arafah. Multi-spectral satellite imagery with 30m spatial resolution was used in this study. The studies were based respectively on Landsat (7) ETM+, 19 January 2003 satellite images taken of the study sites from Thematic Mapper sensor of Landsat satellite, topographical maps and ground observations in order to detect land use changes in the urban and sub-urban areas, by using remotely sensed data and geographic informationsystems technologies. Figure 1shows the Location of Makkah, Mina and Arafah The study area in the Arabian Peninsula is located on 21°N latitude and 39°E longitudes. These selected areas are covered by hill and desert terrain around them.
Methodology
One set of sub image was selected for analysis which centered at Makkah, Mina and Arafah. This was chosen on the basis of urbanization in that place. Enhanced Thematic Mapper data from January 2003 was taken as for being enhanced to get the land cover classification for that particular area. In addition, supervised classification has been used in this study. Supervised classification of multispectral remote sensing imagery is commonly used for land cover determination (Duda and Canty. 2002). As stated in the Introduction section of this paper, the research was designed to using three types of classification namely, Maximum Likelihood classification, Minimum Distance classification, and Parallelepiped classification. The research design attempted to test, as much as each method would give the best result of percentage of accuracy and kappa coefficient for classified area of Makkah, Mina and Arafah, the procedures so that results would enable a systematic, objective comparison among the methods. Due to the inherent differences among the four approaches, however, some variation in each of the methodologies was necessary. This became evident as the research progressed.
Maximum Likelihood Classification is a statistical decision criterion to assist in the classification of overlapping signatures; pixels are assigned to the class of highest probability. The Maximum Likelihood classifier is considered to give more accurate results than Parallelepiped classification however it is much slower due to extra computations.
Minimum distance classifies image data on a database file using a set of 256 possible class signature segments as specified by signature parameter. Each segment specified in signature, for example, stores signature data pertaining to a particular class. Only the mean vector in each class signature segment is used. Other data, such as standard deviations and covariance matrices, are ignored (though the Maximum Likelihood classifier uses this).The result of the classification is a theme map directed to a specified database image channel. A theme map encodes each class with a unique gray level. The gray-level value used to encode a class is specified when the class signature is created. If the theme map is later transferred to the display, then a pseudo-colour table should be loaded so that each class is represented by a different colour.
The parallelepiped classifier uses the class limits and stored in each class signature to determine if a given pixel falls within the class or not. The class limits specify the dimensions (in standard deviation units) of each side of a parallelepiped surrounding the mean of the class in feature space. If the pixel falls inside the parallelepiped, it is assigned to the class. However, if the pixel falls within more than one class, it is put in the overlap class (code 255). If the pixel does not fall inside any class, it is assigned to the null class (code 0).The parallelepiped classifier is typically used when speed is required. The draw back is (in many cases) poor accuracy and a large number of pixels classified as ties (or overlap, class 255). (Environmental Remote Sensing Courseware, Faculty of Science, Chulalongkorn University). Available online
: https://www.sc.chula.ac.th/courseware/2309507/Lecture/remote18.htm
Data Analysis and Results
The image satellite used in this study was captured on 19 January 2003. For the satellite scene, seven Landsat TM bands were used in the multispectral classification analysis using the classifiers mentioned earlier. The PCI Geomatica Version 10.1 digital image processing software was used in all image processing analyses. The raw image for the study area is shown in figure 2. Five land cover categories were recognised in the study area namely vegetation, hill, sand/land, urban and Al-Haram & Mina Tent. The images analysis involved three basic steps in supervised classification: the training stage, the classification stage and the output stage. In the training stage, the supervised classification required some training sites and the areas were established using polygon. A total of 40 sample areas were selected as a training site in this study. Selection of the training sites was based on the colour image.
In the classification stage, three supervised classification methods were selected to classify the images. The three methods are Maximum Likelihood, Minimum Distance-to-Mean and Parallelepiped was performed to the images. Accuracy assessment was carried out to compute the probability of error for the classified map. A total of 200 samples were chosen for accuracy assessment. Many methods of accuracy assessment have been discussed in the remote sensing literature (Aronoff, 1982). Three measures of accuracy were tested in this study namely overall accuracy, confusion or error matrix and kappa coefficient. Many measures of classification accuracy may be derived from a confusion matrix. Kappa coefficient were generated to describe the proportion of agreement between the classification result and the validation sites after random agreements by chance are removed from consideration these data (Thomas, et. al., 2002). In thematic mapping from remotely sensed data, the term accuracy is used typically to express the degree of ‘correctness’ of a map or classification (Foody, 2002).
In the output stage, the classification map was produced as a thematic map of land cover over Makkah, Mina and Arafah. Kappa coefficient and overall accuracy results of the three measures of accuracy are shown in the Table 1. The overall accuracy is expressed as a percentage of the test pixels successfully assigned to the correct classes.
Maximum Likelihood produced the highest accuracy with overall accuracy of 87%. Then followed by Minimum Distance-to-Mean gave the overall classification accuracy of 75.5% and Parallelepiped showed the overall classification accuracy of 10%. The confusion matrix results by using Maximum Likelihood classifier are shown in Table 2. Table 3 shows the results of producer accuracy and user accuracy by using Maximum Likelihood classifier. A classified image using Maximum Likelihood classifier is shown in Figure 3. The dominant feature from land cover type for Makkah, Mina and Arafah is hill which covered around 45.05%. Then, followed by sand/land and urban with the percentage of 24.9% and 24.16% respectively. 3.16% is vegetation area. Finally, the least is Al-Haram & Mina Tent area that covered 2.73% only.
Conclusion
The results presented in this study show efficiency of the three methods that have been used to classify the land cover map over Makkah, Mina and Arafah. From these three classifiers, Maximum Likelihood classification method produced the highest overall accuracy with 87% of accuracy. Besides, five features from the image have been classified. The features are hill, sand/land, urban, vegetation and Al-Haram & Mina Tent. A satellite image map at scale 1:113681 were produced in this study. The high resolution images gave more detail information of the classified map. The classified images could be used for township planning and development purposes in the future.
Acknowledgements
This project was carried out using the Hajj Research Cluster USM grants. We would like to thank the technical staff that participated in this project. Thanks are extended to USM for support and encouragement.
References
- Aronoff, S. (1982). Classification accuracy: a user approach. Photogram metric Engineering and Remote Sensing, 48, 1299-1307
- Bruzzone, L., Cossu, R. and Vernazza, G. (2002). Combining parametric and non-parametric algorithms for a partially unsupervised classification of multitemporal remote sensing images. Information Fusion 3, 289-297
- Clawson, Marion, and Stewart, Charles L., 1965, Land use information. A critical survey of U.S. statistics including possibilities for greater uniformity: Baltimore, Md., The Johns Hopkins Press for Resources for the Future, Inc., 402 p.
- Duda, T. and Cantry, M., 2002, Unsupervised classification of satellite imagery: choosing a good algorithms. International Journal of Remote Sensing, 23 (11), 2193-2212.
- Foody, G. M. 2002. Status of land cover classification accuracy assessment. Remote Sensing and Environment, 80, 185-201
- Franchek, R.J. and Biggam, P.F., 1992. Using USDA Soil Conservation Service County Soils Data with a Geographic Information System: Geographic Information Systems and Mapping Practices and Standards, ASTM STP 1126, A.I. Johnson, C.B. Pettersson and J. Fulton, Eds., American Society for Testing and Materials, Philadelphia, pp. 115-121.
- Gordon, S., 1980. Utilizing Landsat imagery to monitor land use change: A case study in Ohio. Remote Sensing of Environment, 9, pp. 189-196
- Milington, A.C., Jones, A.R., Quarmby, N.A. and Townsend, J.R.G., 1986. Monitoring geomorphological processes in desert marginal environments using multitemporal satellite imagery. Proceedings of the International Symposium on Remote Sensing for Resource Development and Environmental Management, A.A. Balkema.
- Thomas, V., Treitz, P., Jelinski, D., Miller, J., Lafleur, P. and McCaughey, J. H., 2002. Image Classification of a northern peatland complex using spectral and plant community data. Remote Sensing and Environment, 84, 83-99
