440
observations agree with a body of RS literature that show wide variation in patterns
of scale dependency for different landscapes using geostatistical approaches, with
the local variance a function of the relative sizes of the pixels and the discrete targets
themselves (e.g., Woodcock and Strahler 1987). For these reasons, the spatial aggregation approaches often used in other fields (e.g., plant productivity) may actually
confound the detection of optical diversity at certain scales. Consequently, optical
diversity (alpha or beta diversity) or variation in plant traits from RS requires particular attention to scaling methodologies (e.g., Asner et al. 2015).
By providing both fine-scale patterns (Fig. 16.3) and the larger context of landscape structure, imaging spectroscopy (Fig. 16.6) provides a valuable tool for further, more detailed studies, and can help define the proper scale at which alpha and
beta diversity can be most effectively sampled. Remotely sensed data can also be
used to define patterns of geodiversity, the physical template influencing biodiversity (Record et al., Chap. 10, this volume). Geostatistical approaches that examine
optical diversity as a function of distance, analogous to the use of semivariograms
(Curran 1988), can help design appropriate sampling methodologies by illustrating
the influence of landscape features (e.g., ecotones or different vegetation types) on
optical diversity (Fig. 16.6). Imaging spectrometry can also reveal temporal
dynamics and disturbance patterns that can provide additional context for understanding both the drivers and the consequences of biodiversity changes. Furthermore,
image spectrometry can be used to assess the relationships between optical diversity
and ecosystem function (e.g., productivity) over large areas (Wang et al. 2016;
Schweiger et al. 2018). For these reasons, satellite and airborne RS can be a powerful complement to local and regional field studies of biodiversity.
Other approaches to scaling involve the use of models for upscaling or downscaling, including radiative transfer models and statistical models (Knyazikhin et al.
2013; Malenovský et al. 2019; Verrelst et al. 2019). While potentially powerful,
such models are generally limited by the lack of suitable data, bringing us back to
the need for experimental studies incorporating empirical methods. Ultimately, a
combined empirical and theoretical framework that explicitly considers multiple
dimensions of scale is needed to advance our knowledge of biodiversity detection
with RS.
16.4 Designing a Scale-Aware Biodiversity Monitoring
System
To be truly robust, a global biodiversity monitoring system will need to consider
scale in all the dimensions discussed here. Biological concepts (e.g., grain size and
extent) often have a direct analog to RS concepts (e.g., pixel size and extent). Ideally,
we should design our instrument and optical sampling protocols to match our sampling scale to a particular organizational scale of biodiversity. While specific sampling rules have been proposed (e.g., Justice and Townshend 1981), the general rule
J. Gamon et al.
observations agree with a body of RS literature that show wide variation in patterns
of scale dependency for different landscapes using geostatistical approaches, with
the local variance a function of the relative sizes of the pixels and the discrete targets
themselves (e.g., Woodcock and Strahler 1987). For these reasons, the spatial aggregation approaches often used in other fields (e.g., plant productivity) may actually
confound the detection of optical diversity at certain scales. Consequently, optical
diversity (alpha or beta diversity) or variation in plant traits from RS requires particular attention to scaling methodologies (e.g., Asner et al. 2015).
By providing both fine-scale patterns (Fig. 16.3) and the larger context of landscape structure, imaging spectroscopy (Fig. 16.6) provides a valuable tool for further, more detailed studies, and can help define the proper scale at which alpha and
beta diversity can be most effectively sampled. Remotely sensed data can also be
used to define patterns of geodiversity, the physical template influencing biodiversity (Record et al., Chap. 10, this volume). Geostatistical approaches that examine
optical diversity as a function of distance, analogous to the use of semivariograms
(Curran 1988), can help design appropriate sampling methodologies by illustrating
the influence of landscape features (e.g., ecotones or different vegetation types) on
optical diversity (Fig. 16.6). Imaging spectrometry can also reveal temporal
dynamics and disturbance patterns that can provide additional context for understanding both the drivers and the consequences of biodiversity changes. Furthermore,
image spectrometry can be used to assess the relationships between optical diversity
and ecosystem function (e.g., productivity) over large areas (Wang et al. 2016;
Schweiger et al. 2018). For these reasons, satellite and airborne RS can be a powerful complement to local and regional field studies of biodiversity.
Other approaches to scaling involve the use of models for upscaling or downscaling, including radiative transfer models and statistical models (Knyazikhin et al.
2013; Malenovský et al. 2019; Verrelst et al. 2019). While potentially powerful,
such models are generally limited by the lack of suitable data, bringing us back to
the need for experimental studies incorporating empirical methods. Ultimately, a
combined empirical and theoretical framework that explicitly considers multiple
dimensions of scale is needed to advance our knowledge of biodiversity detection
with RS.
16.4 Designing a Scale-Aware Biodiversity Monitoring
System
To be truly robust, a global biodiversity monitoring system will need to consider
scale in all the dimensions discussed here. Biological concepts (e.g., grain size and
extent) often have a direct analog to RS concepts (e.g., pixel size and extent). Ideally,
we should design our instrument and optical sampling protocols to match our sampling scale to a particular organizational scale of biodiversity. While specific sampling rules have been proposed (e.g., Justice and Townshend 1981), the general rule
J. Gamon et al.