GRASS GIS for classification of Landsat TM images by maximum likelihood discriminant analysis: Tokyo area, Japan

Abstract

The presented paper is focused on satellite image analysis using GRASS GIS. The aim is to perform comparative analysis of the land cover changes in Tokyo metropolitan area through spatial analysis. Data include multi-temporal Landsat TM satellite images on 2002, 2006 and 2011. The images were captured from GloVis USGS service and imported to GRASS GIS via GDAL (utilities gdalwarp, r.in.gdal, gdalinfo). The methodology is based on GRASS GIS. The technique includes raster modules (d.rast, r.colors, g.region) and modules of image processing (i.landsat.rgb, i.class). Color composites were created by modules d.rgb, r.composite and auxiliary modules for visualization (d.rast, r.colors, etc). Spectral signatures were generated in an image using 'i.cluster' algorithm and 'i.group' for clustering data. The classification was done by Maximum Likelihhood classifier 'i.maxlik'. The results show variations in land cover types for 2001, 2006 and 2011, which also resulted in the automated grouping pixels into 7, 10 and 6 classes, respectively. The paper demonstrated technical functionality of the GRASS GIS applied for multi-temporal image processing aimed at land cover types / change analysis using shell scripting approach.