Citation
Al-Doski, Jwan M. Mohammed
(2013)
Use of hybrid classification algorithm for land use and land cover analysis in data scarce environment.
Masters thesis, Universiti Putra Malaysia.
Abstract
The technique of remote sensing satellite imaging has played a significant role in facilitating the study of land use/land cover changes (LULC). This is because the information that can be extracted from images constitutes a fundamental key in many diverse applications such as Environment, Planning and Monitoring programs and others. LULC changes are mainly the result of human intervention and natural phenomena such as population growth, urbanization, wars and other factors. During the 1980-1988 Iraq-Iran war, many cities and villages in the north of Iraq were shelled several times with chemical weapons that caused many changes in land covers. Among the cities seriously affected by these chemical weapons is Halabja City (the study area for this research), which was shelled on 16 March 1988, leaving approximately 5,000 people dead and 7,000 injured with long-term damage to their health. In this study, vegetation indices, tasseled cap transformation, hybrid classification as a combination of k-means and support vector machine algorithms,and post-classification comparison were respectively implemented to detect and assess LULC in Halabja. Two Landsat 5 (Thematic Mapper - TM) images obtained in 1986, 1990 with one Landsat 7 (Enhanced Thematic Mapper Plus - ETM+) image acquired in 2000 were used. All images were geometrically corrected and projected to UTM, Datum WGS_84 and Zone 38N using automatic image to image registration with polynomial transformation equations and a nearest neighbor re-sampling algorithm. The root mean square (RMS) error was less than 0.5 pixels. Subsequently,all images were atmospherically corrected by applying dark object subtraction and sub-setted to (1400) samples, (999) lines. The hybrid classifier with the aid of visual interpretation tools, knowledge-based assignment and other supplementary data like Google earth images and vegetation indices were run on subsets to classify images into five thematic classes based on the NLCD 92 classification system scheme (Water Bodies; Shrub Land; Cultivated/Planted Area; Low-Intensity Urban Area; and Bare Land). To assess classification accuracy, the classified images were randomly sampled to produce confusion matrix which provided LULCC maps with an average overall accuracy of 95% and 0.94 Kappa statistic that tendered them deal for further qualitative and quantitative analysis of land cover changes through a postclassification. Based on the overall accuracy and kappa statistics, hybrid classifier was found to be more preferred classification approach than k-means and SVM. A multi-date post-classification comparison algorithm was used to determine LULC changes in two intervals, 1986-1990, and 1990-2000. Change analysis during 1986 to 1990 revealed that all classes decreased and showed few changes except the bare land which showed an increase of about 30%. The Low intensity urban changed area was determined and overlaid with chemical weapons bombing location GPS points; roads with the aid of the NDBI index to locate low intensity urban area changes. It was noticed that bombed places are the same places where the urban area changed. During the 1990 to 2000 period, there were significant increases in low intensity residential and cultivated / plant areas. The low intensity residential area increased by 83%. Most of the increments of this class come from the conversion of 36% water bodies, 24% of shrub land, 14% of bare land, and 6% of low intensity residential areas. On the contrary, there was a significant decrease in water bodies by 55% overall and other class designations. In conclusion, hybrid classification as a combination of k-means and support vector machine algorithms and post-classification comparison change detection technique can be used to monitor land cover changes in Halabja city, Iraq.
Download File
Additional Metadata
Actions (login required)
|
View Item |