170
have sensors with numerous discrete bands, greater than just the visible region, such
as infrared, whereas hyper-spectral cameras have sensors with continuous sensitivity across the full optical spectrum (350–2500 nm). These types of sensor are available but expensive. Although likely to become a more affordable component of
UAV technology in the future, their cost is currently prohibitive. Traditional cameras can be adapted to sense IR radiation by removing or replacing the inbuilt IR
filters inside the internal sensor mechanism (Infragram 2016). Therefore, by using
multiple traditional cameras (RGB) and IR adapted cameras, and combining the
output, we should be able to produce a multispectral image capable of identifying
plants such as R. ponticum.
Taylor et al. (2013) showed, using lab-based radiometry of more than 500 leaves
of R. ponticum that the spectral reflectance of leaves from three non-target species;
(namely Beech (Fagus sylvatica), Holly (Ilex aquifolium) and Laurel (Prunus laurocerasus)), were significantly different (p < 0.05) from R. ponticum for all wavelengths except 450 and 460 nm, which yielded no significant difference (Taylor
et al. 2013). The bands used for statistical analysis were at 490, 550, 610, 1040 and
1490 nm, based on the absolute reflectance for the key wavelengths shown in previous studies. These reflectance wavelengths relate to the following specific pigments
and compounds: Chlorophyll a, phycoerythrin, phycocyanin, oils and cell sugars
respectively. The same study also found that the spectral characteristics of R. ponticum leaves differ in leaf size and spectral plasticity over four habitat types (garden,
oakwood, pinewood and lakeside), with these differences caused by variations in
the spectral intensity of specific leaf pigments.
Taylor et al.’s study demonstrates some of the major opportunities and pitfalls for
this technology; although differentiation between key non-target species is possible,
under controlled conditions, the plasticity of inter species variation could render the
information highly site specific. In our study, and contrary to the aims of many UAV
studies, site-specific technology would not make this tool less effective if the output
is sufficiently accurate to justify the processing time and costs.
The extent to which a remote sensing tool can be used to determine the coverage
and distribution of understorey R. ponticum is governed by the capacity of a sensor
in terms of spatial resolution, spectral sensitivity, extent covered, and temporal
frequency. Each remote sensing tool has benefits and limitations associated with
these four factors, often where a higher capacity in one factor is attained to the detriment of another attribute, for example; lower spatial resolution for greater coverage
in satellite imagery. Monitoring large-scale changes in forest cover, for example,
would not require a pixel size of 1 cm so using a lower resolution image would be
more appropriate: this highlights that there is no ‘one size fits all’ model for remotely
sensed data. Furthermore, although satellite imagery has the coverage, temporal
frequency and the spectral sensitivity to achieve the basic objective of ‘identifying
understorey R. ponticum from surrounding vegetation’, the spatial resolution will
not be sufficiently high to distinguish individuals between bare branches of the
upper canopy. Aerial imagery can overcome this but has a repeatedly high cost for
flights and the open-source imagery currently available for our study area lacks the
seasonal and temporal frequency (summer 2006 and 2014) to be used to identify
A. Sanders
have sensors with numerous discrete bands, greater than just the visible region, such
as infrared, whereas hyper-spectral cameras have sensors with continuous sensitivity across the full optical spectrum (350–2500 nm). These types of sensor are available but expensive. Although likely to become a more affordable component of
UAV technology in the future, their cost is currently prohibitive. Traditional cameras can be adapted to sense IR radiation by removing or replacing the inbuilt IR
filters inside the internal sensor mechanism (Infragram 2016). Therefore, by using
multiple traditional cameras (RGB) and IR adapted cameras, and combining the
output, we should be able to produce a multispectral image capable of identifying
plants such as R. ponticum.
Taylor et al. (2013) showed, using lab-based radiometry of more than 500 leaves
of R. ponticum that the spectral reflectance of leaves from three non-target species;
(namely Beech (Fagus sylvatica), Holly (Ilex aquifolium) and Laurel (Prunus laurocerasus)), were significantly different (p < 0.05) from R. ponticum for all wavelengths except 450 and 460 nm, which yielded no significant difference (Taylor
et al. 2013). The bands used for statistical analysis were at 490, 550, 610, 1040 and
1490 nm, based on the absolute reflectance for the key wavelengths shown in previous studies. These reflectance wavelengths relate to the following specific pigments
and compounds: Chlorophyll a, phycoerythrin, phycocyanin, oils and cell sugars
respectively. The same study also found that the spectral characteristics of R. ponticum leaves differ in leaf size and spectral plasticity over four habitat types (garden,
oakwood, pinewood and lakeside), with these differences caused by variations in
the spectral intensity of specific leaf pigments.
Taylor et al.’s study demonstrates some of the major opportunities and pitfalls for
this technology; although differentiation between key non-target species is possible,
under controlled conditions, the plasticity of inter species variation could render the
information highly site specific. In our study, and contrary to the aims of many UAV
studies, site-specific technology would not make this tool less effective if the output
is sufficiently accurate to justify the processing time and costs.
The extent to which a remote sensing tool can be used to determine the coverage
and distribution of understorey R. ponticum is governed by the capacity of a sensor
in terms of spatial resolution, spectral sensitivity, extent covered, and temporal
frequency. Each remote sensing tool has benefits and limitations associated with
these four factors, often where a higher capacity in one factor is attained to the detriment of another attribute, for example; lower spatial resolution for greater coverage
in satellite imagery. Monitoring large-scale changes in forest cover, for example,
would not require a pixel size of 1 cm so using a lower resolution image would be
more appropriate: this highlights that there is no ‘one size fits all’ model for remotely
sensed data. Furthermore, although satellite imagery has the coverage, temporal
frequency and the spectral sensitivity to achieve the basic objective of ‘identifying
understorey R. ponticum from surrounding vegetation’, the spatial resolution will
not be sufficiently high to distinguish individuals between bare branches of the
upper canopy. Aerial imagery can overcome this but has a repeatedly high cost for
flights and the open-source imagery currently available for our study area lacks the
seasonal and temporal frequency (summer 2006 and 2014) to be used to identify
A. Sanders