Evaluation of Airborne Video Data for Land Cover Classification Accuracy Assessment

Abstract

This research was designed to evaluate the operational utility of airborne video for classification accuracy assessment of satellite imagery. There are a number of logistical advantages in utilizing airborne video data particularly in remote or large catchment regions where ground accessibility can be prohibitively expensive. In a study of land cover characteristics in the Barossa Valley, South Australia, comparisons of Landsat ETM+ classification accuracy assessments were derived from both airborne video and standard ground truthing point sampling techniques. The airborne video data was obtained by randomly frame grabbing 243 individual scenes along four predetermined flight lines and computing classification accuracies and error matrix statistics. These results were then compared with similar statistics derived from 180 randomly chosen ground sampling sites. The costs involved in obtaining both sets of data per sampling pont were also calculated. The results demonstrated that the airborne video data provided similar degrees of accuracy assessment at a cost of approximately one seventh that of the ground survey with the added advantages of increased data collection, overall improved site accessibility and the ability to continually store and review the data. The cost differentiation would become even more pronounced in more remote and inaccessible areas. In general the results of the unsupervised Landsat ETM+ classification were rather poor producing an overall general accuracy of 70% and a Kappa coefficient of 0.65, highlighting the classification difficulties encountered in using commercial satellite imagery for land cover assessments in an area with very heterogeneous fine grained land use patterns.