177
bandwidths (p > 0.01). The MANOVA between species’ full spectral signatures
were shown to all be universally significantly different at F(1, 42) = 6.795, p < 0.001.
This result confirms that the analysis using 7 band signatures is sufficient to distinguish species in principle.
Digital Classification
The spectral analysis checkerboard, as a tool in itself, provided a platform for standardising comparison of samples from multiple areas and was a valuable output
from this study as a stand-alone utensil (Fig. 5). Both supervised and unsupervised
classifications of the checkerboard displayed a clear visual distinction between
many of the samples, which was confirmed by the overall accuracy assessment:
52% unsupervised and 84% supervised (Fig. 5).
When a classification was conducted on 7-band test sections of the site the accuracy assessment yielded 62% overall classification accuracy for 100 pixel samples
using the ERDAS
®
imagine internal accuracy assessment feature. Cherry Laurel,
Beech and Ivy were all incorrectly identified as R. ponticum or vice versa. Beech
foliage would not be present at the optimal time of survey (midwinter) and Cherry
Laurel is an additional reservoir host for Phytophthora so both were considered as
target species. Ivy, however, was a non-target species of concern as it would be
expected to be present in most natural and semi-natural woodlands across the UK
and would confound results (Fig. 5). Sample sizes for the checkerboard were greater
than the coverage of Ivy (0.78 m) and this could have contributed to errors.
Discussion
Feasability Study
This study demonstrates the feasibility of creating multispectral mosaics from commonly sourced equipment. The separate bands of the intra-band variation were significantly different. Variability in significance between 540 nm and 550 nm showed
the importance of this region for the identification of Rhododendron from other
species, which confirmed other results in the literature (Taylor et al. 2013). This
information enables future studies to be targeted for more rapid optimisation of
tools for identification: a greater bandwidth in the 550 nm, for example, could have
made this region incorrectly significant due to overlap into the 540 nm bandwidth
(Fig. 6).
The accuracy levels seen in both the checkerboard and the mosaic demonstrate
the ability of a pixel-based supervised classification of 7-band composites to identify
groups of species, including the R. ponticum and Cherry Laurel group, both of
which are reservoir hosts of Phytophthora.
Mapping the Distribution of Understorey Rhododendron Ponticum Using…
bandwidths (p > 0.01). The MANOVA between species’ full spectral signatures
were shown to all be universally significantly different at F(1, 42) = 6.795, p < 0.001.
This result confirms that the analysis using 7 band signatures is sufficient to distinguish species in principle.
Digital Classification
The spectral analysis checkerboard, as a tool in itself, provided a platform for standardising comparison of samples from multiple areas and was a valuable output
from this study as a stand-alone utensil (Fig. 5). Both supervised and unsupervised
classifications of the checkerboard displayed a clear visual distinction between
many of the samples, which was confirmed by the overall accuracy assessment:
52% unsupervised and 84% supervised (Fig. 5).
When a classification was conducted on 7-band test sections of the site the accuracy assessment yielded 62% overall classification accuracy for 100 pixel samples
using the ERDAS
®
imagine internal accuracy assessment feature. Cherry Laurel,
Beech and Ivy were all incorrectly identified as R. ponticum or vice versa. Beech
foliage would not be present at the optimal time of survey (midwinter) and Cherry
Laurel is an additional reservoir host for Phytophthora so both were considered as
target species. Ivy, however, was a non-target species of concern as it would be
expected to be present in most natural and semi-natural woodlands across the UK
and would confound results (Fig. 5). Sample sizes for the checkerboard were greater
than the coverage of Ivy (0.78 m) and this could have contributed to errors.
Discussion
Feasability Study
This study demonstrates the feasibility of creating multispectral mosaics from commonly sourced equipment. The separate bands of the intra-band variation were significantly different. Variability in significance between 540 nm and 550 nm showed
the importance of this region for the identification of Rhododendron from other
species, which confirmed other results in the literature (Taylor et al. 2013). This
information enables future studies to be targeted for more rapid optimisation of
tools for identification: a greater bandwidth in the 550 nm, for example, could have
made this region incorrectly significant due to overlap into the 540 nm bandwidth
(Fig. 6).
The accuracy levels seen in both the checkerboard and the mosaic demonstrate
the ability of a pixel-based supervised classification of 7-band composites to identify
groups of species, including the R. ponticum and Cherry Laurel group, both of
which are reservoir hosts of Phytophthora.
Mapping the Distribution of Understorey Rhododendron Ponticum Using…