Hyperspectral Sensors and Applications
37
1.6.2.3
Vegetation Mapping
Vegetation discrimination has been highlighted as one of the benefits of using
hyperspectral data and the establishment of spectral libraries for different
species has been suggested (Roberts et al. 1998). However, discrimination of
species from single date imagery is only achievable where a combination of
leaf chemistry, structure and moisture content culminate to form a unique
spectral signature. Discrimination from imagery then relies on the extraction
of the pure spectral signature for each species which is dictated by the spatial
resolution of the observing sensor and also the timing of observation (Asner
and Heidebrecht 2002) and requires consideration of the spectral differences
and variations that occur within and between species. Identification of species
and communities can be further enhanced through observation over the same
area at different times within the year. As an example, studies using AVIRIS
data acquired on several dates over Jasper Ridge, California (Merton 1999),
confirmed the existence of distinct hysteresis loops for different vegetation
classes that could be used to refine discrimination. Moreover, these multitemporal hysteresis loops can identify the timing of critical phenological or
environmental-induced events and assist researchers in targeting subsequent
image acquisitions to specific "indicator times" of the year (Merton 1999;
Merton and Silver 2000).
Fundamental to the discrimination of tree/shrub species from fine « 1 m)
spatial resolution imagery is the extraction of signatures from the crown. For
this purpose, a number of procedures have been developed for the delineation
of individual tree crowns/clusters within fine spatial resolution hyperspectral
data (Ticehurst 2001; Culvenor 2002; Held et al. 2003). From these delineated
crowns, representative reflectance data or indices can be derived subsequently
to facilitate discrimination and mapping.
For regional to global mapping, the enhanced spectral characteristics and
radiometric quality of the MODIS sensor has allowed routine mapping ofland
cover types. Using 1 km data acquired between November 2000 and October
2001,17 different land cover types have been classified, following the International Geosphere-Biosphere Program (IGBP) scheme and using decision tree
and artificial neural network techniques (see Chapter 2). These land cover
types include eleven classes of natural vegetation (e. g., savannas, wetlands
and evergreen/deciduous woodlands), several classes of developed and mosaic land, and areas with snow and ice (Friedl et al. 2002). Validation of the
algorithm has been based on a network of sites observed using finer spatial
resolution remote sensing data. These maps provide a better understanding of
global ecosystems and land cover, particularly as they are updated routinely,
and contribute to, for example, global carbon science initiatives.
37
1.6.2.3
Vegetation Mapping
Vegetation discrimination has been highlighted as one of the benefits of using
hyperspectral data and the establishment of spectral libraries for different
species has been suggested (Roberts et al. 1998). However, discrimination of
species from single date imagery is only achievable where a combination of
leaf chemistry, structure and moisture content culminate to form a unique
spectral signature. Discrimination from imagery then relies on the extraction
of the pure spectral signature for each species which is dictated by the spatial
resolution of the observing sensor and also the timing of observation (Asner
and Heidebrecht 2002) and requires consideration of the spectral differences
and variations that occur within and between species. Identification of species
and communities can be further enhanced through observation over the same
area at different times within the year. As an example, studies using AVIRIS
data acquired on several dates over Jasper Ridge, California (Merton 1999),
confirmed the existence of distinct hysteresis loops for different vegetation
classes that could be used to refine discrimination. Moreover, these multitemporal hysteresis loops can identify the timing of critical phenological or
environmental-induced events and assist researchers in targeting subsequent
image acquisitions to specific "indicator times" of the year (Merton 1999;
Merton and Silver 2000).
Fundamental to the discrimination of tree/shrub species from fine « 1 m)
spatial resolution imagery is the extraction of signatures from the crown. For
this purpose, a number of procedures have been developed for the delineation
of individual tree crowns/clusters within fine spatial resolution hyperspectral
data (Ticehurst 2001; Culvenor 2002; Held et al. 2003). From these delineated
crowns, representative reflectance data or indices can be derived subsequently
to facilitate discrimination and mapping.
For regional to global mapping, the enhanced spectral characteristics and
radiometric quality of the MODIS sensor has allowed routine mapping ofland
cover types. Using 1 km data acquired between November 2000 and October
2001,17 different land cover types have been classified, following the International Geosphere-Biosphere Program (IGBP) scheme and using decision tree
and artificial neural network techniques (see Chapter 2). These land cover
types include eleven classes of natural vegetation (e. g., savannas, wetlands
and evergreen/deciduous woodlands), several classes of developed and mosaic land, and areas with snow and ice (Friedl et al. 2002). Validation of the
algorithm has been based on a network of sites observed using finer spatial
resolution remote sensing data. These maps provide a better understanding of
global ecosystems and land cover, particularly as they are updated routinely,
and contribute to, for example, global carbon science initiatives.