158
Arrasado. R. officinalis also shows such a broad spatial distribution, but it reached its
maximum cover in the Naves area. Meanwhile, S. genistoides colonised both the
Naves and Manto Arrasado zones but with a lower plant cover fraction, achieving a
maximum of 60% in the Naves. Finally, U. australis was distributed mostly across
intermediate altitude areas of the Manto Arrasado and reached a maximum fraction
cover of 60%.
We used the correlation coefficient and RMSE values between the resulting species fraction cover from the AHS imagery and the ground-truth fraction cover to
asses the overall accuracy. There was a strong correlation for E. scoparia with high
R
2
values (p < 0.05). H. halimifolium and R. officinalis had intermediate R
2
values
(p < 0.05). Finally, the correlation for U. australis and S. genistoides was very low
and with no statistical significance.
Discussion
The protocol presented in this study intended to incorporate the most robust and
widely used procedures in airborne imaging spectroscopy, field spectroscopy, and
spectral unmixing to establish a standard protocol for mapping plant species. The
spatially-explicit distribution maps of the plant species generated by the airborne
imaging spectroscopy increase the knowledge of the spatial spread of each species
and its relation with the ecological processes and the perturbation that takes place
within the ecosystem. Collaboration between the imaging sensor operator organisation and user organisation was fundamental for executing the protocol.
Although spectral unmixing enables plant species mapping, further work on
intra-species variability and similarity among species is required. It is crucial to
identify the time of the year with maximum separability among species. There is no
standard protocol for developing a spectral library for plant species, but the work by
Jiménez and Díaz-Delgado (2015) takes the first steps towards this.
Airborne imaging spectroscopy is currently the most important source of hyperspectral data, and has the best capacity to provide species mapping within and also
surrounding protected areas. Hyperspectral imagery acquired by manned aircraft
has less uncertainty in radiometric and geometric accuracy than RPAS or drones,
which, could represent a future systems for monitoring changes in plant species
distribution. In relation to the imaging spectrometer, the image spatial resolution
must be adapted to the size and cover for each vegetation type, taking into account
the number of flight lines needed to cover the study area. Furthermore, although the
VNIR region is the most important spectrum region for vegetation studies, an imaging spectrometer that records within the SWIR region is recommended to enhance
plant species discrimination and the spectral unmixing procedure.
Nowadays, MESMA is the most appropriate LSU algorithm to cope with plant
species mapping, primarily due to the incorporation of multiple endmembers and
algorithms for endmember optimization. However, at present, it is only implemented in the VIPER tool application (Roberts et al. 2007) and not used widely in
commercial remote sensing software.
M. Jiménez and R. Díaz-Delgado
Arrasado. R. officinalis also shows such a broad spatial distribution, but it reached its
maximum cover in the Naves area. Meanwhile, S. genistoides colonised both the
Naves and Manto Arrasado zones but with a lower plant cover fraction, achieving a
maximum of 60% in the Naves. Finally, U. australis was distributed mostly across
intermediate altitude areas of the Manto Arrasado and reached a maximum fraction
cover of 60%.
We used the correlation coefficient and RMSE values between the resulting species fraction cover from the AHS imagery and the ground-truth fraction cover to
asses the overall accuracy. There was a strong correlation for E. scoparia with high
R
2
values (p < 0.05). H. halimifolium and R. officinalis had intermediate R
2
values
(p < 0.05). Finally, the correlation for U. australis and S. genistoides was very low
and with no statistical significance.
Discussion
The protocol presented in this study intended to incorporate the most robust and
widely used procedures in airborne imaging spectroscopy, field spectroscopy, and
spectral unmixing to establish a standard protocol for mapping plant species. The
spatially-explicit distribution maps of the plant species generated by the airborne
imaging spectroscopy increase the knowledge of the spatial spread of each species
and its relation with the ecological processes and the perturbation that takes place
within the ecosystem. Collaboration between the imaging sensor operator organisation and user organisation was fundamental for executing the protocol.
Although spectral unmixing enables plant species mapping, further work on
intra-species variability and similarity among species is required. It is crucial to
identify the time of the year with maximum separability among species. There is no
standard protocol for developing a spectral library for plant species, but the work by
Jiménez and Díaz-Delgado (2015) takes the first steps towards this.
Airborne imaging spectroscopy is currently the most important source of hyperspectral data, and has the best capacity to provide species mapping within and also
surrounding protected areas. Hyperspectral imagery acquired by manned aircraft
has less uncertainty in radiometric and geometric accuracy than RPAS or drones,
which, could represent a future systems for monitoring changes in plant species
distribution. In relation to the imaging spectrometer, the image spatial resolution
must be adapted to the size and cover for each vegetation type, taking into account
the number of flight lines needed to cover the study area. Furthermore, although the
VNIR region is the most important spectrum region for vegetation studies, an imaging spectrometer that records within the SWIR region is recommended to enhance
plant species discrimination and the spectral unmixing procedure.
Nowadays, MESMA is the most appropriate LSU algorithm to cope with plant
species mapping, primarily due to the incorporation of multiple endmembers and
algorithms for endmember optimization. However, at present, it is only implemented in the VIPER tool application (Roberts et al. 2007) and not used widely in
commercial remote sensing software.
M. Jiménez and R. Díaz-Delgado