A neural network model for land use change classification, with application to the Nile River Delta

Abstract

Detecting the monitoring changes in conditions at the Earth's surface are essential for understanding human impact on the environment and for assessing the sustainability of development. In the next decade, NASA will gather high-resolution multi-spectral and multi-temporal data, which could be used for analyzing long-term changes, provided that available methods can keep pace with the accelerating flow of information. This paper introduces an automated technique for change identification, based on the ARTMAP neural network. This system overcomes some of the limitations of traditional change detection methods, and also produces a measure of confidence in classification accuracy. Landsat thematic mapper (TM) imagery of the Nile River delta provides a testbed for land use change classification methods. This dataset consists of a sequence of ten images acquired between 1984 and 1993 at various times of year. Field observations and photo interpretations have identified 358 sites as belonging to eight classes, three of which represent changes in land use over the ten-year period. A particular challenge posed by this database is the unequal representation of various land use categories: three classes, urban, agriculture in delta, and other, comprise 95% of pixels in labeled sites. A two-step sampling method enables unbiased training of the neural network systems across sites.