The Role of Remotely Sensed Image Texture in Estimating Forest Stand Structural Complexity

Information on the structural complexity of forest stands is required to inform conservation priorities that guide the sustainable management of private native vegetation for multiple objectives of landowners. Whilst research into methods for the operational mapping of vegetation structure is expanding, the research is largely exploratory, and no methods have yet seen widespread acceptance. This research compares spectral (First principal component, Normalised Difference Vegetation Index, Enhanced Vegetation Index, Infrared Vegetation Index) and spatial transformations (Variance, Morans’ I, and G*) of Landsat remotely-sensed imagery in combination with environmental attributes (Topographic derivatives and Soil Fertility) for estimating within-stand variations in forest structural complexity using linear regression analysis. Dry sclerophyll forests on the Southern Tablelands of New South Wales, Australia were used as a case study. Predictions were obtained for all structural attributes with the exception of Vegetation Cover 0-0.5m, and Overstorey regeneration. Errors of up to 31 percent of the field measured range of attributes accompanied predictions. Environmental attributes were more commonly selected as explanatory variables than derivatives of Landsat imagery, with NDVI and spatial autocorrelation measures (Moran’s I and G*) the most commonly selected derivatives. Despite the moderate accuracy of predictions, the estimates of forest stand structural complexity are a useful information source for natural resource managers who require information on the relative structural complexity of native vegetation stands within the landscape. The utility of finer spatial resolution imagery is a key research priority since the moderate resolution of Landsat imagery limited the sensitivity of its derivatives in the current study.