Remote Sensing of Lantana Camara in a Forest Environment

Abstract

Impacts of invasive plants have been well documented in terms of costs and effects upon the environment. The spread of invasives is one of the world's most serious conservation issues as it is a major contributing factor in the extinction of species and consequently, in management terms, advances global biodiversity loss resulting in the homogenisation of the world's flora and fauna. Lantana camara is a worldwide weed with international significance. Impacts of Lantana are numerous and far reaching. Current management strategies for Lantana include determining the extent and level of current infestations, the reporting of new infestations and the investigation of the dynamics of the spread of Lantana population. The purpose of this research was to investigate the effectiveness of high spatial satellite imagery in mapping Lantana in a forested environment. Multispectral Quickbird imagery was used to distinguish Lantana patches from surrounding forest vegetation. Lantana density, canopy cover and ground cover information was collected from 64 sample sites within Toonumbar National Park, Richmond Range National Park and Toonumbar State Forest, northern NSW, Australia. Four Lantana cover types, three forest cover types and cover types for water, road, pasture and shadow areas were selected for use in a supervised classification. Initial overall accuracy was 67%, with a kappa value of 35%. The satellite image was resampled to study the effect of different spatial resolutions on mapping accuracy. Application of a 5x5 majority filter to the image increased the classification accuracy to 75%, with a kappa value of 50%. It was considered appropriate to use majority filters as the positional accuracy obtained during the study was deemed too inaccurate to be convinced that the coordinates would match exactly with the pixel of interest. With GPS errors varying between a few metres to over 10m and an average RMSE value of 3.44 pixels (8.26m) recording only the cover within the one pixel area did not seem appropriate. The majority filters represented .this larger view of the area without averaging and losing information as would occur with an averaging filter. Our findings show that Quickbird imagery has the potential for mapping Lantana in forested environments. The high spatial resolution allows for detection of patches of Lantana smaller than previous medium resolution sensors could detect. Sources of error and methods to optimise accuracy are discussed in relation to the high spatial resolution imagery used for this study.