168
created from samples of the site mosaics which yielded 84% classification accuracy
in ERDAS
®
imagine v2016 (64bit). The ‘noisy’ mosaic dataset yielded a lower
overall classification accuracy of 64 %. The techniques demonstrated show the
potential for low-tech, UAV derived, multispectral imagery to aid land management. Spectral separability of rhododendron shows a high level of potential for
mapping its distribution using this method. However, more research should be conducted to streamline the process and reduce the potential sources of error.
Keywords Rhododendron • R.ponticum • Forestry management • Phytophthera •
UAV • Invasive species • Multispectral pixel-based classification
Introduction
Invasive species are currently classified, according to the International Union for the
Conservation of Nature (IUCN), as the second largest threat to biodiversity worldwide and by the UK Forestry Commission (FC) as a major obstruction to woodland
regeneration (Blackburn et al. 2014; Edwards 2006). Rhododendron ponticum is
one such species, which has become invasive across much of the northern hemisphere; in Europe, the UK and North America (Edwards 2006; Taylor et al. 2013).
Invasive species are traditionally surveyed manually from the ground (Lillesand
et al. 2014), with the surveyor undertaking species identification and estimating
percentage cover (Brinker and Minnick 2012). Typically, the scope of the exercise
would depend on the resources available to the organisation undertaking the survey
(Lillesand et al. 2014). Various studies attest to the spatial inaccuracy of manual
mapping compared with GIS spatial mapping. However, numerous studies also
demonstrate the inability of remote sensing to perform as well as a ground surveyor,
in accuracy and breadth of useable information (Powell et al. 2004; Burrough 1986).
This study is an example of how remote sensing can enhance the resources available
to land managers by providing an alternative means of surveying difficult to access
areas to help quantify removal costs. Therefore, one aim of this research was to use
low-cost commercially available UAV and cameras to mimic more advanced equipment, which is prohibitively expensive. A second aim was to use the output to identify understorey R. ponticum and assess the feasibility and accuracy of this equipment
for practical nature conservation purposes.
Species Characteristics and Invasion
R. ponticum has complex global spatial distribution patterns arising from farreaching radiations and advantageous hybridizations (Milne et al. 2003). The earliest fossilised evidence, found in North America, is carbon dated to around 68
A. Sanders
created from samples of the site mosaics which yielded 84% classification accuracy
in ERDAS
®
imagine v2016 (64bit). The ‘noisy’ mosaic dataset yielded a lower
overall classification accuracy of 64 %. The techniques demonstrated show the
potential for low-tech, UAV derived, multispectral imagery to aid land management. Spectral separability of rhododendron shows a high level of potential for
mapping its distribution using this method. However, more research should be conducted to streamline the process and reduce the potential sources of error.
Keywords Rhododendron • R.ponticum • Forestry management • Phytophthera •
UAV • Invasive species • Multispectral pixel-based classification
Introduction
Invasive species are currently classified, according to the International Union for the
Conservation of Nature (IUCN), as the second largest threat to biodiversity worldwide and by the UK Forestry Commission (FC) as a major obstruction to woodland
regeneration (Blackburn et al. 2014; Edwards 2006). Rhododendron ponticum is
one such species, which has become invasive across much of the northern hemisphere; in Europe, the UK and North America (Edwards 2006; Taylor et al. 2013).
Invasive species are traditionally surveyed manually from the ground (Lillesand
et al. 2014), with the surveyor undertaking species identification and estimating
percentage cover (Brinker and Minnick 2012). Typically, the scope of the exercise
would depend on the resources available to the organisation undertaking the survey
(Lillesand et al. 2014). Various studies attest to the spatial inaccuracy of manual
mapping compared with GIS spatial mapping. However, numerous studies also
demonstrate the inability of remote sensing to perform as well as a ground surveyor,
in accuracy and breadth of useable information (Powell et al. 2004; Burrough 1986).
This study is an example of how remote sensing can enhance the resources available
to land managers by providing an alternative means of surveying difficult to access
areas to help quantify removal costs. Therefore, one aim of this research was to use
low-cost commercially available UAV and cameras to mimic more advanced equipment, which is prohibitively expensive. A second aim was to use the output to identify understorey R. ponticum and assess the feasibility and accuracy of this equipment
for practical nature conservation purposes.
Species Characteristics and Invasion
R. ponticum has complex global spatial distribution patterns arising from farreaching radiations and advantageous hybridizations (Milne et al. 2003). The earliest fossilised evidence, found in North America, is carbon dated to around 68
A. Sanders