52
3.1.4 Key Considerations for the Use of Imaging Spectroscopy
Data for Scaling and Mapping Plant Functional Traits
One of the chief challenges to effectively using imaging spectroscopy has been the
acquisition of data of sufficient resolution, quality, and consistency for broad
application in vegetation studies (Table 3.2). This necessitates measurements in the
shortwave infrared (SWIR, 1100–2500 nm) in addition to the visible and near
infrared (VNIR, 400–1100 nm). While VNIR wavelengths are most sensitive to
pigments and overall canopy health, longer wavelengths are required to retrieve
many biochemicals and LMA (Serbin et al. 2014; Kokaly and Skidmore 2015;
Serbin et al. 2015; Singh et al. 2015). Spectral resolution is critical as well, with
10 nm band spacing and 10 nm full-width half maximum (FWHM) generally considered essential to identify traits detected based on narrow absorption features.
Even finer resolution is required to detect spectral features that rely on narrow (<0.5
nm) atmospheric windows, such as solar-induced fluorescence (SIF, Yang et al.
2018). Other key considerations include sufficient signal-to-noise ratio (SNR) to
identify important spectral features, accounting for both coherent and random noise
related to detector sensitivity, dark current, and stray light. Additional sensor characteristics important to using imaging spectroscopy include spectral distortion.
Most sensors are push-broom sensors, in which an image is constructed via the
forward movement of the platform. Spatial samples are measured in the X-dimension
(pixels) of the detector array and spectral wavelengths in the Y-dimension.
Nonuniformity may arise due to differences in detectors in both dimensions, meaning that different detectors in the X-dimension see different central wavelengths
(smile) and offsets in the Y-dimension lead to band-to-band misregistration (keystone). All of these effects can influence the ability to detect traits reliably within
one scene or across multiple scenes using common algorithms. Full understanding
of detector (and thus image) uniformity as well as the measurement point-spread
function in 3-D (spatial X [detector X], spatial Y [platform movement], and spectral
[detector Y]) is critical to accurate retrievals.
All RS data require some level of post-processing. Imaging spectroscopy is no
different; prior to implementing algorithms for trait retrieval (Sect. 3.2.2), additional efforts must be undertaken to ensure consistent measurements in consistent
units such that retrievals from imagery from multiple sources, dates, locations, etc.
can be compared. Minimally, pixel measurements should be converted to radiances
(w m
-2
sr
-1
nm
-1
) based on laboratory calibrations and regular vicarious measurements of stable targets. With proper instrument characterization, keystone, smile,
and other radiometric artifacts can be reduced. Subsequently, atmospheric corrections to convert radiance to reflectance (percent) are essential for cross-site studies.
The approaches to atmospheric correction are numerous and tailored to particular
environments, e.g., terrestrial vs. aquatic systems. Even within terrestrial applications, approaches differ among airborne data products (e.g., NASA’s AVIRISClassic and AVIRIS-NG sensors vs. NEON AOP) and do not necessarily yield
consistent reflectance imagery. Finally, new approaches that take advantage of
S. P. Serbin and P. A. Townsend
3.1.4 Key Considerations for the Use of Imaging Spectroscopy
Data for Scaling and Mapping Plant Functional Traits
One of the chief challenges to effectively using imaging spectroscopy has been the
acquisition of data of sufficient resolution, quality, and consistency for broad
application in vegetation studies (Table 3.2). This necessitates measurements in the
shortwave infrared (SWIR, 1100–2500 nm) in addition to the visible and near
infrared (VNIR, 400–1100 nm). While VNIR wavelengths are most sensitive to
pigments and overall canopy health, longer wavelengths are required to retrieve
many biochemicals and LMA (Serbin et al. 2014; Kokaly and Skidmore 2015;
Serbin et al. 2015; Singh et al. 2015). Spectral resolution is critical as well, with
10 nm band spacing and 10 nm full-width half maximum (FWHM) generally considered essential to identify traits detected based on narrow absorption features.
Even finer resolution is required to detect spectral features that rely on narrow (<0.5
nm) atmospheric windows, such as solar-induced fluorescence (SIF, Yang et al.
2018). Other key considerations include sufficient signal-to-noise ratio (SNR) to
identify important spectral features, accounting for both coherent and random noise
related to detector sensitivity, dark current, and stray light. Additional sensor characteristics important to using imaging spectroscopy include spectral distortion.
Most sensors are push-broom sensors, in which an image is constructed via the
forward movement of the platform. Spatial samples are measured in the X-dimension
(pixels) of the detector array and spectral wavelengths in the Y-dimension.
Nonuniformity may arise due to differences in detectors in both dimensions, meaning that different detectors in the X-dimension see different central wavelengths
(smile) and offsets in the Y-dimension lead to band-to-band misregistration (keystone). All of these effects can influence the ability to detect traits reliably within
one scene or across multiple scenes using common algorithms. Full understanding
of detector (and thus image) uniformity as well as the measurement point-spread
function in 3-D (spatial X [detector X], spatial Y [platform movement], and spectral
[detector Y]) is critical to accurate retrievals.
All RS data require some level of post-processing. Imaging spectroscopy is no
different; prior to implementing algorithms for trait retrieval (Sect. 3.2.2), additional efforts must be undertaken to ensure consistent measurements in consistent
units such that retrievals from imagery from multiple sources, dates, locations, etc.
can be compared. Minimally, pixel measurements should be converted to radiances
(w m
-2
sr
-1
nm
-1
) based on laboratory calibrations and regular vicarious measurements of stable targets. With proper instrument characterization, keystone, smile,
and other radiometric artifacts can be reduced. Subsequently, atmospheric corrections to convert radiance to reflectance (percent) are essential for cross-site studies.
The approaches to atmospheric correction are numerous and tailored to particular
environments, e.g., terrestrial vs. aquatic systems. Even within terrestrial applications, approaches differ among airborne data products (e.g., NASA’s AVIRISClassic and AVIRIS-NG sensors vs. NEON AOP) and do not necessarily yield
consistent reflectance imagery. Finally, new approaches that take advantage of
S. P. Serbin and P. A. Townsend