438
angular information may be useful in biodiversity detection. Sensor view angle
affects different spectral regions differently, illustrating an interaction between
angular scale and spectral scale (Gamon 2015). With the advent of lidar (Asner et al.
2012) and structure-from-motion (SFM; Wallace et al. 2016) to characterize plant
3-D structure, angular information can now be better understood and effectively
integrated with optical RS to improve biodiversity or plant trait detection.
16.3 Implementing Scaling Approaches
Properly addressing scale effects in biodiversity detection requires explicit attention
to scale in all its dimensions and designing a study approach that transcends scale
limitations. Scaling methods can be empirical or theoretical. Typical empirical
methods involve a multi-scale sampling strategy, often using multiple RS instruments operating at different scales, along with traditional field sampling for validation. Often, a goal of such field campaigns is to aggregate fine-scale data to be used
as validation for coarse-scale data (e.g., Cohen and Justice 1999; Wehlage et al.
2016), yet aggregation tends to obscure spectral variability at the scale of individual
leaves or plant canopies and undercuts the goal of detecting local (alpha) biodiversity with optical diversity. On the other hand, sampling at progressively larger scales
often involves transitions from local (alpha) to regional (beta) diversity, creating the
opposite effect of increasing optical diversity with increasing pixel size, and these
transitions can themselves be scale-dependent and vary with vegetation type
(Fig. 16.6).
The complexity of scaling effects involving the transition from alpha to beta
diversity is briefly illustrated in Fig. 16.6, which illustrates optical diversity (CV in
this case) calculated for different regions along a transect crossing several plant
communities, including woodland, grassland, and experimental grassland plots at
the Cedar Creek Ecosystem Science Reserve, Minnesota (USA). In this case, the
highest optical diversity values occur at edges, points of abrupt landscape transitions marking the boundaries between adjacent communities (i.e., ecotones), an
effect commonly seen in remotely sensed images (Paz-Kagan et al. 2017). In this
complex landscape, (and in contrast to the plot-level patterns shown in Fig. 16.3),
CV generally increases with spatial scale, reflecting a transition from alpha to beta
diversity with increasing spatial lag (pixel sizes). Interestingly, this transition occurs
at about 10 m × 10 m in the manipulated grassland (see arrow and blue curve,
Fig. 16.6), matching the plot sizes in this experiment, but occurs more gradually in
the natural grassland (black curve, Fig. 16.6). These patterns seem to contradict
findings of declining optical diversity in grasslands with increasing pixel size
obtained at finer scales (1 mm to 1 m in Wang et al. 2018a) (see also Fig. 16.3). This
comparison illustrates that the patterns of scale dependence are themselves scale
dependent and can differ within the same landscape or for different communities
depending upon both individual crown size and larger landscape structure. These
J. Gamon et al.
angular information may be useful in biodiversity detection. Sensor view angle
affects different spectral regions differently, illustrating an interaction between
angular scale and spectral scale (Gamon 2015). With the advent of lidar (Asner et al.
2012) and structure-from-motion (SFM; Wallace et al. 2016) to characterize plant
3-D structure, angular information can now be better understood and effectively
integrated with optical RS to improve biodiversity or plant trait detection.
16.3 Implementing Scaling Approaches
Properly addressing scale effects in biodiversity detection requires explicit attention
to scale in all its dimensions and designing a study approach that transcends scale
limitations. Scaling methods can be empirical or theoretical. Typical empirical
methods involve a multi-scale sampling strategy, often using multiple RS instruments operating at different scales, along with traditional field sampling for validation. Often, a goal of such field campaigns is to aggregate fine-scale data to be used
as validation for coarse-scale data (e.g., Cohen and Justice 1999; Wehlage et al.
2016), yet aggregation tends to obscure spectral variability at the scale of individual
leaves or plant canopies and undercuts the goal of detecting local (alpha) biodiversity with optical diversity. On the other hand, sampling at progressively larger scales
often involves transitions from local (alpha) to regional (beta) diversity, creating the
opposite effect of increasing optical diversity with increasing pixel size, and these
transitions can themselves be scale-dependent and vary with vegetation type
(Fig. 16.6).
The complexity of scaling effects involving the transition from alpha to beta
diversity is briefly illustrated in Fig. 16.6, which illustrates optical diversity (CV in
this case) calculated for different regions along a transect crossing several plant
communities, including woodland, grassland, and experimental grassland plots at
the Cedar Creek Ecosystem Science Reserve, Minnesota (USA). In this case, the
highest optical diversity values occur at edges, points of abrupt landscape transitions marking the boundaries between adjacent communities (i.e., ecotones), an
effect commonly seen in remotely sensed images (Paz-Kagan et al. 2017). In this
complex landscape, (and in contrast to the plot-level patterns shown in Fig. 16.3),
CV generally increases with spatial scale, reflecting a transition from alpha to beta
diversity with increasing spatial lag (pixel sizes). Interestingly, this transition occurs
at about 10 m × 10 m in the manipulated grassland (see arrow and blue curve,
Fig. 16.6), matching the plot sizes in this experiment, but occurs more gradually in
the natural grassland (black curve, Fig. 16.6). These patterns seem to contradict
findings of declining optical diversity in grasslands with increasing pixel size
obtained at finer scales (1 mm to 1 m in Wang et al. 2018a) (see also Fig. 16.3). This
comparison illustrates that the patterns of scale dependence are themselves scale
dependent and can differ within the same landscape or for different communities
depending upon both individual crown size and larger landscape structure. These
J. Gamon et al.