441
of thumb that the pixel size must be smaller than the size of the individual sampling
unit (e.g., typical plant canopy) is a good starting point, and this has been supported
by studies discussed above (see Fig. 16.3).
However, matching instrument to organizational scale remains a challenge; we
typically have a poor match between the pixel size and sampling extent and the
organism size and distribution on the ground, often due to the practical constraints
of field sampling (in the case of biological studies) and instrument design (in the
case of RS). Most remote sensing instruments are designed for a particular airborne
or spaceborne platform with physics and engineering requirements in mind. The
detector response is constrained by the amount of electromagnetic energy available,
which in turn determines the sensor design, pixel size, and spectral resolution
needed to achieve a given signal-to-noise ratio. Greater signal-to-noise ratios can be
attained by reducing the spectral resolution (combining narrow bands into broadbands, e.g., via spectral binning), or by reducing spatial resolution (e.g., pixel binning), but these choices limit the ability to distinguish individuals, species, and
functional types due to the degradation of spectral and spatial information. Orbital
and altitudinal considerations also determine the pixel size obtainable from a particular sensor platform. Together, these constraints often reduce the ability to
properly distinguish individual organisms or vegetation functional types. Adding to
this mismatch, field sampling (including plot size, transect size, and location) is
often limited by practical considerations of personnel, time, and budget and is rarely
designed with RS in mind.
Improvements in sensor design and novel sampling platforms can relax these
impediments, but with trade-offs. For example, flexible airborne platforms, emerging unmanned aerial vehicles (UAV) systems, or robotic ground-based systems
(Wang et al. 2018a, b) provide useful platforms for testing the effects of spatial
scale, but may not always have the temporal or spectral coverage desired for biodiversity detection (Fig. 16.5). Plans to deploy global satellite imaging spectrometers
with frequent repeat visits (Schimel et al., Chap. 19) offer new ways to explore the
temporal dimension with high spectral resolution and wide spectral range, although
at relatively coarse spatial resolution. Consequently, a key application of satellite
RS can be to provide the larger context within which effective sampling regimes can
then be defined at finer scales.
Due to these inherent limitations and trade-offs between scale dimensions, we
suggest that the ideal global monitoring system would be an integrated, multicomponent system, combining RS at different scales (satellite and aircraft sensors)
with proximal sensing (ground optical sensors and field sampling of biodiversity)
(Fig. 16.2). Such a system would operate within a clearly defined scaling framework, incorporating empirical and modeling approaches, with explicit attention to
sampling scale in each of the dimensions mentioned above. Although rarely used in
RS, explicit experimental approaches involving cross-scale and cross-instrument
comparisons should be a key capability of an effective global biodiversity monitoring system. With a multi-scale system, it would be possible to express results (e.g.,
spectral variability) as a function of sampling scale (pixel or grain size; see Figs. 16.3
16 Consideration of Scale in Remote Sensing of Biodiversity
of thumb that the pixel size must be smaller than the size of the individual sampling
unit (e.g., typical plant canopy) is a good starting point, and this has been supported
by studies discussed above (see Fig. 16.3).
However, matching instrument to organizational scale remains a challenge; we
typically have a poor match between the pixel size and sampling extent and the
organism size and distribution on the ground, often due to the practical constraints
of field sampling (in the case of biological studies) and instrument design (in the
case of RS). Most remote sensing instruments are designed for a particular airborne
or spaceborne platform with physics and engineering requirements in mind. The
detector response is constrained by the amount of electromagnetic energy available,
which in turn determines the sensor design, pixel size, and spectral resolution
needed to achieve a given signal-to-noise ratio. Greater signal-to-noise ratios can be
attained by reducing the spectral resolution (combining narrow bands into broadbands, e.g., via spectral binning), or by reducing spatial resolution (e.g., pixel binning), but these choices limit the ability to distinguish individuals, species, and
functional types due to the degradation of spectral and spatial information. Orbital
and altitudinal considerations also determine the pixel size obtainable from a particular sensor platform. Together, these constraints often reduce the ability to
properly distinguish individual organisms or vegetation functional types. Adding to
this mismatch, field sampling (including plot size, transect size, and location) is
often limited by practical considerations of personnel, time, and budget and is rarely
designed with RS in mind.
Improvements in sensor design and novel sampling platforms can relax these
impediments, but with trade-offs. For example, flexible airborne platforms, emerging unmanned aerial vehicles (UAV) systems, or robotic ground-based systems
(Wang et al. 2018a, b) provide useful platforms for testing the effects of spatial
scale, but may not always have the temporal or spectral coverage desired for biodiversity detection (Fig. 16.5). Plans to deploy global satellite imaging spectrometers
with frequent repeat visits (Schimel et al., Chap. 19) offer new ways to explore the
temporal dimension with high spectral resolution and wide spectral range, although
at relatively coarse spatial resolution. Consequently, a key application of satellite
RS can be to provide the larger context within which effective sampling regimes can
then be defined at finer scales.
Due to these inherent limitations and trade-offs between scale dimensions, we
suggest that the ideal global monitoring system would be an integrated, multicomponent system, combining RS at different scales (satellite and aircraft sensors)
with proximal sensing (ground optical sensors and field sampling of biodiversity)
(Fig. 16.2). Such a system would operate within a clearly defined scaling framework, incorporating empirical and modeling approaches, with explicit attention to
sampling scale in each of the dimensions mentioned above. Although rarely used in
RS, explicit experimental approaches involving cross-scale and cross-instrument
comparisons should be a key capability of an effective global biodiversity monitoring system. With a multi-scale system, it would be possible to express results (e.g.,
spectral variability) as a function of sampling scale (pixel or grain size; see Figs. 16.3
16 Consideration of Scale in Remote Sensing of Biodiversity