Linear discriminant and profile analysis An aid in remote sensing for geobotanical investigation

Abstract

Linear discriminant and profile analysis techniques have been applied using Airborne Thematic Mapper (ATM) data. Two sites were selected representing geochemically anomalous and background vegetation areas of part of the Assynt district in north-west Scotland. Linear discriminant analysis reveals that the vegetation over a geochemically anomalous area is distinguishable from background vegetation in the reflected and thermal infrared bands of ATM data. The correlation of individual bands with the linear discriminant functions demonstrates that the ATM infrared bands 6 (0·695–0·75 µm) and 7 (0·76–0·90 µm) contribute most to the separation of stressed and background vegetation. Band 1 (0·42–0·45 µm), band 9 (1·55–1·75 µm) and band 11 (8·5–13·0 µm) contribute moderately to the discrimination of two types of vegetation groups, background and stressed vegetation respectively. Profile analysis for different levels of a defined geochemically anomalous area suggests that a statistically significant difference exists in the mean response of bands from anomalous and background areas. The discriminant function is used for the classification of anomalous and background vegetation areas. The pixels classified as being in the anomalous vegetation group are classified 83·6 per cent correctly and 16·4 per cent incorrectly, while the pixels in the background vegetation group are classified 93·2 per cent correctly and 6·8 per cent incorrectly.