Landslide Mapping by Textural Analysis of Atm Data

In this paper we evaluate two statistical approaches to semi-automated texture enhancement and discrimination for landslide mapping in semi-arid, sedimentary terrain from Daedalus ATM data. A supervised texture discrimination technique is applied, based on calculating similarity between a reference texture spectrum obtained from training samples and spectra from moving image windows. The results are compared with those from interpreting a set of popular texture measures from the literature, derived from grey level co-occurrence matrix statistics. In this comparison, interpretation is facilitated by statistical selection of the best combination of three measures using a sequential forward search algorithm. It is concluded that the texture spectrum based discrimination technique proves superior to using pre-defined sets of texture measures, since it is able to highlight areas on imagery which are often associated with disrupted, displaced land masses.