Use of Quickbird Imagery to Map Vegetation Communities to extract Warkworth Sand Woodlands in the Hunter Valley in Australia

Abstract

The Warkworth Sands Woodlands (WSW) was listed as an Endangered Ecological Community in Australia in 2003 by the NSW Scientific Committee. The community has been circumscribed to occur on Aeolian sand deposits generally, resting on a river terrace with woodlands to low woodlands comprising 'Angophora floribunda', 'Banksia integrifolia', 'Acacia filicifolia', 'Pteridium esculentum', 'Imperata cylindrica', 'Brachyloma daphnoides' and 'Melaleuca thymifolia' and a variety of other woodland species that together amount to 42 species. WSW woodland can be difficult to define because the vegetation type has sub-elements (such as a wet spectrum dominated by M. 'thymifolia') and the subvegetation types often merge with other vegetation types found on shallower sandy profiles (such as 'Allocasuarina luehmannii' woodlands). The aim of this project was to determine if remote sensing tools and approaches could be used to capture a digital definition of WSW. Quickbird satellite imagery was analysed to distinguish the vegetation types in a 1395 hectare zone in the Warkworth area of the Hunter Valley. The image was rectified and a Root Mean Square (RMS) error of less than one pixel was obtained. The Maximum Likelihood classification technique was used to classify this image, originally into 18 classes, then down to 12 classes and then into 6 broad classes. The overall accuracy for the 14 classes was 88%, and for the 6 classes it was 83%. The results showed that even in such a complex ecosystem remotely sensed imagery could be used to classify the vegetation into communities with reasonably high accuracy levels.