513
be achieved using stratification by a few biome types (grasslands/tundra, savannas,
shrublands/forests). Indeed, this approach has been adopted for many global RS
products, such as multiple MODIS variables (e.g., LAI, GPP). Much work is now
required to develop practical, broadly applicable yet acceptably accurate and wellcharacterized retrievals for global application. More specific and computational
methods, such as radiative transfer model inversion, may be used for local applications (Morsdorf et al., Chap. 4). The history of imaging spectroscopy has been of
relatively small data collections from aircraft and from space with Hyperion
(Pearlman et al. 2003), and as a result, investigators have not been greatly constrained by computation and have used algorithms that require a fair amount of
manual intervention. The likely volume of data from a global spectrometer is so
great that algorithms will have to run largely unattended and simple, robust
approaches will be needed!
19.6 Pixel Size
Much of the extant literature on IS and ecology has been collected using airborne
imagers flying at comparatively low altitudes, so that the pixels are quite small;
1–2 m
2
is common (Asner et al. 2012; Kampe et al. 2010), and even data collected
using NASA’s research aircraft, the ER-2, flying at >20,000 m, are often 15 m
2
,
smaller than the expected 30–45 m
2
spaceborne pixels. Data on most of the dimensions of biodiversity can be tied to measurements or samples from not only a particular species but often an individual of that species. Most field sampling involves
taxonomic identification followed by some sort of measurement or sampling process, increasingly tied to a geographic coordinate as well (Fig. 19.2).
By contrast, a 30 m pixel, 900 square meters, is not consistent with delineating a
single organism or species in most cases. In fact, not only is there a scale mismatch
between 30 m pixels and plants, but the magnitude of the mismatch depends on the
ecosystem (see Gamon et al., Chap. 16). However conceptualized, the reality is that
ecologists will have to cope with this range of mismatch and may only be able to
inject intermediate scales at specific field sites, efforts which are likely to prove
essential (Gamon et al., Chap. 16, Barnett et al. 2019). The range in size differences
between plants and pixels is extraordinary. For example, to take one extreme, there
is a factor of 10
10
difference in size between a single phytoplankter and a 30 m pixel.
A grassland represents an intermediate case, as a single grass plant (neglecting the
fact that large swards of genetically identical individuals are known) is about 10
−6
of
a 30 m pixel, a million times smaller, while a large tree may cover a tenth or more
of a 30 m pixel. There is thus not only a huge scale mismatch between plants and
pixels, but the degree of mismatch may vary by 10
9
between ecosystems.
The scale mismatch implies that the plant functional trait measurement from
space characterizes a collection of organisms, a huge number for phytoplankton,
and quite a large one for herbaceous plants, but possibly just a few individuals in a
forest. The interpretation of the functional trait measurements will thus depend
19 Prospects and Pitfalls for Spectroscopic Remote Sensing of Biodiversity…
be achieved using stratification by a few biome types (grasslands/tundra, savannas,
shrublands/forests). Indeed, this approach has been adopted for many global RS
products, such as multiple MODIS variables (e.g., LAI, GPP). Much work is now
required to develop practical, broadly applicable yet acceptably accurate and wellcharacterized retrievals for global application. More specific and computational
methods, such as radiative transfer model inversion, may be used for local applications (Morsdorf et al., Chap. 4). The history of imaging spectroscopy has been of
relatively small data collections from aircraft and from space with Hyperion
(Pearlman et al. 2003), and as a result, investigators have not been greatly constrained by computation and have used algorithms that require a fair amount of
manual intervention. The likely volume of data from a global spectrometer is so
great that algorithms will have to run largely unattended and simple, robust
approaches will be needed!
19.6 Pixel Size
Much of the extant literature on IS and ecology has been collected using airborne
imagers flying at comparatively low altitudes, so that the pixels are quite small;
1–2 m
2
is common (Asner et al. 2012; Kampe et al. 2010), and even data collected
using NASA’s research aircraft, the ER-2, flying at >20,000 m, are often 15 m
2
,
smaller than the expected 30–45 m
2
spaceborne pixels. Data on most of the dimensions of biodiversity can be tied to measurements or samples from not only a particular species but often an individual of that species. Most field sampling involves
taxonomic identification followed by some sort of measurement or sampling process, increasingly tied to a geographic coordinate as well (Fig. 19.2).
By contrast, a 30 m pixel, 900 square meters, is not consistent with delineating a
single organism or species in most cases. In fact, not only is there a scale mismatch
between 30 m pixels and plants, but the magnitude of the mismatch depends on the
ecosystem (see Gamon et al., Chap. 16). However conceptualized, the reality is that
ecologists will have to cope with this range of mismatch and may only be able to
inject intermediate scales at specific field sites, efforts which are likely to prove
essential (Gamon et al., Chap. 16, Barnett et al. 2019). The range in size differences
between plants and pixels is extraordinary. For example, to take one extreme, there
is a factor of 10
10
difference in size between a single phytoplankter and a 30 m pixel.
A grassland represents an intermediate case, as a single grass plant (neglecting the
fact that large swards of genetically identical individuals are known) is about 10
−6
of
a 30 m pixel, a million times smaller, while a large tree may cover a tenth or more
of a 30 m pixel. There is thus not only a huge scale mismatch between plants and
pixels, but the degree of mismatch may vary by 10
9
between ecosystems.
The scale mismatch implies that the plant functional trait measurement from
space characterizes a collection of organisms, a huge number for phytoplankton,
and quite a large one for herbaceous plants, but possibly just a few individuals in a
forest. The interpretation of the functional trait measurements will thus depend
19 Prospects and Pitfalls for Spectroscopic Remote Sensing of Biodiversity…