273
time than its surroundings, this information can be used to time image acquisition
for when the target species appears most spectrally different. The spatial domain
consists of pixel size and overall geographic coverage, and the spectral domain
consists of the number of wavelengths, the position and bandwidth of wavelengths measured, and the spectral range of the sensor at which radiance can be
measured reliably. Sensors typically have trade-offs among these domains based
upon sensor design, size limitations, and data volume limitations. For example,
in the spatial domain, there is a trade-off between overall coverage area or swath
width and pixel size; both can be forced to increase, but at the expense of sensor
size, which limits the platform it can be mounted on. There are also trade-offs
between domains, mostly related to platforms. Most satellite platforms have
larger pixel sizes than other platforms (20–100 s of meters) but have quick revisit
time (days to weeks) and greater geographic coverage. Airborne platforms have
a longer revisit time due to costs and logistics and smaller spatial coverage but
offer smaller pixel size (centimeters to meters) and often support hyperspectral
sensors. Unmanned aircraft systems (UAS) offer quick revisit time, on- demand
deployment, and small pixel size but have very limited spatial coverage and limited spectral resolution due to size restrictions.
When direct detection is not possible due to canopy cover or other factors,
indirect methods can be used to predict species locations. Species knowledge
regarding habitat constraints or coexisting species can be used to govern a model
using other data products. These data include things like digital elevation models
(DEMs), climate layers, soil moisture, and any factor restricting species location.
In some situations, these data can also be combined with direct detection methods to improve results.
3. Enhance data and model/classify. A model or classifier can be thought of as a set
of rules or a mathematical function that uses pixel data to assign or predict class
membership. This can either be supervised, where training data (pixels or spectra
that have been identified previously) are used to define classes, or unsupervised,
where classes are formed based upon pixel spectral/statistical similarity. Usually,
atmospherically corrected surface reflectance data are provided to the classifier.
Often, image enhancement is conducted to increase the information content of
the input data. In addition to reflectance data, enhanced products can also be supplied to the classifier. Methods to enhance spectral data include spectral indices,
principal component analysis (PCA), and minimum noise fraction (MNF).
Spectral indices are combinations of spectral reflectance from two or more wavelengths that highlight a given reflectance or absorption feature and often indicate
relative abundance of features of interest; for example, the Normalized Difference
Vegetation Index (NDVI) is a normalized difference ratio of red and near- infrared
(NIR) bands commonly used as an indicator of vegetation vigor.
With hyperspectral data, many narrowband indices are available that provide
additional data about plant traits, including light use efficiency from the photochemical reflectance index (PRI; Gamon et al. 1997), canopy nitrogen from the
normalized difference nitrogen index (NDNI; Serrano et al. 2002), canopy water
content from the normalized difference water index (NDWI; Gao 1995), and a
large number of leaf pigment indices [see Sims and Gamon (2002) for an
12 Remote Detection of Invasive Alien Species
time than its surroundings, this information can be used to time image acquisition
for when the target species appears most spectrally different. The spatial domain
consists of pixel size and overall geographic coverage, and the spectral domain
consists of the number of wavelengths, the position and bandwidth of wavelengths measured, and the spectral range of the sensor at which radiance can be
measured reliably. Sensors typically have trade-offs among these domains based
upon sensor design, size limitations, and data volume limitations. For example,
in the spatial domain, there is a trade-off between overall coverage area or swath
width and pixel size; both can be forced to increase, but at the expense of sensor
size, which limits the platform it can be mounted on. There are also trade-offs
between domains, mostly related to platforms. Most satellite platforms have
larger pixel sizes than other platforms (20–100 s of meters) but have quick revisit
time (days to weeks) and greater geographic coverage. Airborne platforms have
a longer revisit time due to costs and logistics and smaller spatial coverage but
offer smaller pixel size (centimeters to meters) and often support hyperspectral
sensors. Unmanned aircraft systems (UAS) offer quick revisit time, on- demand
deployment, and small pixel size but have very limited spatial coverage and limited spectral resolution due to size restrictions.
When direct detection is not possible due to canopy cover or other factors,
indirect methods can be used to predict species locations. Species knowledge
regarding habitat constraints or coexisting species can be used to govern a model
using other data products. These data include things like digital elevation models
(DEMs), climate layers, soil moisture, and any factor restricting species location.
In some situations, these data can also be combined with direct detection methods to improve results.
3. Enhance data and model/classify. A model or classifier can be thought of as a set
of rules or a mathematical function that uses pixel data to assign or predict class
membership. This can either be supervised, where training data (pixels or spectra
that have been identified previously) are used to define classes, or unsupervised,
where classes are formed based upon pixel spectral/statistical similarity. Usually,
atmospherically corrected surface reflectance data are provided to the classifier.
Often, image enhancement is conducted to increase the information content of
the input data. In addition to reflectance data, enhanced products can also be supplied to the classifier. Methods to enhance spectral data include spectral indices,
principal component analysis (PCA), and minimum noise fraction (MNF).
Spectral indices are combinations of spectral reflectance from two or more wavelengths that highlight a given reflectance or absorption feature and often indicate
relative abundance of features of interest; for example, the Normalized Difference
Vegetation Index (NDVI) is a normalized difference ratio of red and near- infrared
(NIR) bands commonly used as an indicator of vegetation vigor.
With hyperspectral data, many narrowband indices are available that provide
additional data about plant traits, including light use efficiency from the photochemical reflectance index (PRI; Gamon et al. 1997), canopy nitrogen from the
normalized difference nitrogen index (NDNI; Serrano et al. 2002), canopy water
content from the normalized difference water index (NDWI; Gao 1995), and a
large number of leaf pigment indices [see Sims and Gamon (2002) for an
12 Remote Detection of Invasive Alien Species