Using UAV imagery to perform fine-scale mapping of wetland vegetation

<p>Differentiating between species of plants in aerial imagery is often challenging and, in some cases, can be impossible without significant field data collection. However, remote sensing technology is developing to the point where it is increasingly possible to eliminate the need for extensive fieldwork entirely and conduct non-disruptive monitoring of fragile environments. The increasing availability of UAV platforms with integrated high-resolution cameras and low-cost image processing software is also making remote sensing operations accessible to those outside the scientific community with an interest in environmental monitoring. This project trialled an emerging set of image analysis techniques called ‘object-based image analysis’ to create fine scale maps of a recovering wetland area, based on aerial photographs collected using a consumer-grade UAV (unmanned aerial vehicle). The effects of including additional ancillary data (such as digital surface models (DSMs) and multispectral imagery) in the classification process were also assessed to compare the ability of a standard digital camera to produce high-accuracy classifications to that of a more specialised multispectral sensor. The inclusion of this extra information was found to significantly improve classification accuracy in almost all cases, making a strong argument for the inclusion of ancillary data whenever possible, especially when considering the ease with which ancillary datasets can be produced. The high-resolution (between 2 and 4cm/pixel) imagery provided sufficient detail to observe 28 distinct land cover classes in total, with around 20 classes per image. While the number of classes in the classification scheme may have imposed limits on the overall accuracy of the classified maps, several classes were classified with a high (70% or greater) level of accuracy, including two invasive species, showing that the object-based school of image classification has potential to be a powerful tool for detecting and tracking individual vegetation types.</p>