Data and Methods

Abstract

Landsat data with a resolution of 30 m at two or more points of time was used to assess land use/land cover change. Since supervised classification necessitates creating numerous training sets, data was collected using a handheld Global Positioning System. A combination of road sampling and transect sampling was followed. During road sampling, points were selected as far as possible at a distance of at least 100 m from the road. This was done to reduce the influence of the road and to avoid generating any bias in the sample that would have occurred had the points are collected close to roads. The maximum likelihood classifier (MLC) which is a widely used classification algorithm was used for the classification and the separability of classes was checked prior to running the classification. Standard accuracy assessment using measures such as user accuracy, producer accuracy and kappa accuracy were performed. Along with Landsat data, ASTER GDEMs and gridded population data from the Landscan dataset of the Oak Ridge National Laboratory for 2008 were used. These datasets were analyzed using Erdas Imagine 9.2 and ArcGIS 9.3 and the spatial statistics program Fragstats 4.1 was used to analyze landscape metrics.