Use of remotely sensed data for forest type mapping and inventory in north east Tasmania, Australia

This thesis has developed a methodology for the extraction of landform and land cover information in complex terrain using Landsat Multispectral Scanner (MSS) digital data. As a result it has been possible to produce a forest inventory for the Scottsdale forestry district in north east Tasmania. Considerable research has been directed towards the application of Landsat multispectral scanner data in the forest environment. To date, most of this work has been done overseas, and relatively few studies have been done in Australian forests. This thesis has reviewed most of this work, highlighting major deficiencies and accomplishments. An important result in this study has been the extraction of useful information in complex terrain. Very few researchers have reported success in mapping land cover types in mountainous areas. This is mainly due to variations in facet slope and orientation which govern the radiant energy intercepted by individual pixels in the Landsat scene. As a result, cover types may have very similar spectral reflectances but quite different radiances due to the shading effect of topography. This makes the classification and labelling exercise difficult. Tasmania is a high relief area and an illumination model employing the logarithm of band ratios was used to account for topographic effects. Ancillary data comprising district and forest block boundaries were integrated with Landsat data in the classification and labelling of various forest types in the study area. A ground truth survey verified that the resulting land cover was being mapped within a satisfactory level of accuracy. The overall accuracy level was 89 percent, whilst for individual land cover types the accuracy level ranged between 37 to 96 percent. Two Landsat scenes (1980 and 1984) were classified and labelled separately. These two scenes were resampled to a common base grid with a two second resolution. Spectral change detection methods and change detection based on two dates classification were analyzed. Temporal changes in the major land cover types were obtained by differencing the two classifications. The extent of these changes were calculated not only for the district as a whole but also for each of the forest blocks separately. The methodology developed in this thesis is also applicable to satellite systems with increased spatial and spectral resolution. In particular, data from the Landsat Thematic Mapper, SPOT and MOS-1 create new and exciting possibilities for future work in remote sensing of forest resources. The increased spatial resolution of these satellites not only increases the potential for visual interpretation of the images, but as well should provide an improvement in the accuracy and detail of information provided by the classification techniques described in this thesis. This study has clearly demonstrated the value of merging MSS data with ancillary data such as digital terrain and different administrative boundaries. Using these methods, a conceptual information system for forest resources in Tasmania is also explored leading to specific recommendations for the form of an operational image base information system for forest resources.