A. M. Dewan
Department of Geography and Environment, University of Dhaka, Ramna, Dhaka 1000, Bangladesh
Y. Yamaguchi
Graduate School of Environmental Studies, Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-8601, Japan
Dhaka Metropolitan . Land use/cover . Monitoring . Remote sensing . Change detection
Dhaka Metropolitan of Bangladesh
Knowledge Management
Remote sensing
Data acquisition and preparation The topographic maps of 1960 (scale 1:50,000) were obtained from Survey of Bangladesh (SOB) and used to generate the 1960 land use/cover types. In addition, Landsat 1, 4, 5 and 7, and IRS-1D data were acquired for this study. Four Landsat data comprised of MSS (27 March 1975), TM (03 February 1988; 01 February 1999) and ETM+ (04 January 2003) were obtained and employed in this study. Furthermore, one IRS-1D LISS III data (26 December 2005) was also acquired and used for the 2005 land use/cover categorization. Please note that the Landsat and IRS thermal bands were not included to detect land cover change. Both ArcGIS (ESRI 2005) and Erdas Imagine (Leica Geosystems 2006) software were used to derive land use/cover classification in a multi-temporal approach. First of all, the topographic maps of 1960 were scanned and rectified using a geometrically corrected image with the ArcGIS software. As unsystematic errors remain in commercially available remote sensing data, geometric correction was therefore needed to reduce the error. The images used in this study were geometrically corrected using a Landsat TM image of 1997 as a reference. At least 75 well distributed ground control points were used in the rectification process. The root mean square error (RMSE) varied from 0.25– 0.45 pixels. Finally, a first-order polynomial fit was applied and all the data were resampled to 30 m pixel size using the nearest neighbor method. The Bangladesh Transverse Mercator System (BTM) was used as the coordinate system which is an area-specific standard UTM projection system for Bangladesh. Besides, a number of geospatial data including municipal boundaries, road networks, geomorphic units and elevation units have been constructed as GIS layers from diverse sources. Land use/cover classification and change detection A modified version of the Anderson Scheme Level I (Anderson et al. 1976) was adopted to study the land use/cover change. Though the scheme was originally developed for the USA, it is the widely used land use/ cover classification system across the world (Mundia and Aniya 2006; Shalaby and Tateishi 2007; Yuan et al. 2005). The system proposed multilevel land use/cover classification of which level I classes can be mapped from Landsat data or from high-altitude airphoto/ imagery, whereas the extraction of information at levels II, III, and IV requires the use of high, medium, and low-altitude photographs, respectively. Six separable land use/cover types have been identified in this study as water bodies, wetland/lowlands, built-up, cultivated land, vegetation and bare soil/landfill. The 1960 land cover map has been developed from topographic maps which are created from aerial photographs taken in 1955 followed by extensive ground truthing in 1956. Two scanned topographic maps (sheet no. 79 I 5 and 6) were displayed on the computer screen. Using ArcGIS, the 1960 land cover map was digitized, edited, and leveled. Besides, a large scale map (1:20,000) of 1961 by Survey of Bangladesh (SOB) depicting the study area and its surroundings was employed to identify various land cover types in a GIS platform. The 1962 land use classification map with seven categories by Khan and Islam (1964) was also useful to distinguish the land use/cover information on topographic maps. Finally, vector land cover data were rasterized to perform raster-based change analysis. Assessing classification accuracy Classified land cover maps from satellite data were further used for validation using ground truth data obtained from a variety of sources. For the 1975, 1988, and 1999 land use/cover maps, a total of 125 pixels were generated using the stratified random sampling method. Then using the geographical locations of features available on the land use maps, high resolution images, and Survey of Bangladesh topographic maps, accuracy evaluation of the derived maps were performed. To assess the accuracy of 2003 and 2005 land use/cover maps, reference data obtained from the field were utilized. In doing so, 100 reference data for 2003 and 110 field data for 2005 were used to assess the classification accuracy. Finally, accuracy reports of each land cover data in terms of overall accuracy, producers/users accuracy, and kappa coefficient have been generated.
Environ Monit Assess (2009) 150:237–249
Journal