between usable images can be longer due to cloud cover (Song et al., 2001). Very high
spatial resolution imagery such as GeoEye’s IKONOS and Digital Globe’s QuickBird
or WorldView 2 images (∼60 cm) have 1–7-day return times depending on latitude
and tasking mode (Anderson and Marchisio, 2012) but in practice their temporal
resolution can be limited by the high cost of images and scheduling constraints.
Also, as a generality, high spatial resolution at the pixel level often involves a
trade-off in extent of coverage. As examples, the swath width of QuickBird is only
16.5 km, compared to 185 km for Landsat and 2330 km for Moderate Resolution
Imaging Spectroradiometer (MODIS) images from the Terra satellite. In contrast to
satellite imagery, tower-mounted, networked digital cameras (phenocams) provide
near-surface images with nearly continuous temporal resolution and spatial resolution
of 1 cm or less, but extent of coverage is generally limited to 1–100 m
2 (Richardson
et al., 2007; Rundel et al., 2009; Sonnentag et al., 2012).
In an early review of issues of scale in remote sensing, Woodcock and Strahler
(1987) pointed out that in most scientific endeavors the investigator selects the scale at
which observations are collected, whereas investigators using spaceborne sensors are
limited to the specific scales of observations inherent in the satellite imagery. At that
time Landsat and AVHRR satellites were used for most Earth observation studies.
However, with the increase in available sensor systems, as noted in the examples
above, it is now possible to combine imagery to cover a wide range of temporal and
spatial scales across image sets to conduct landscape change studies. More recent
reviews reflect the much broader range of choices now available for obtaining data
at multiple scales (e.g., Mulder et al., 2011; Skidmore et al., 2011; Turner 2011;
Abraha and Savage, 2012; Pfeifer et al., 2012). These reviews emphasize the
problems and opportunities in combining ecological observations and models
across spatial and temporal scales, from the cellular level at response times of
minutes to days (e.g., response of photosynthesis to temperature) to the global level
at response times of decades (e.g., response of global net primary productivity to
global warming).
Another scale issue is spectral resolution. Satellite images tend to have just a few
band windows available and currently operational hyperspectral imagery is mainly
acquired from aircraft and a few experimental satellites. Therefore, the detailed
reflectance spectra that can be used to distinguish different types of soils and
vegetation units are not available from application-oriented satellite imagery.
Among other tools such as spectral mixture analysis, researchers often rely on
vegetation indices (VIs), combinations of two or three bands that separate the
landscape into soil, water, and different densities of green vegetation, based
primarily on differences in their reflectance of light in the adjacent red and
near-infrared (NIR) bands (Glenn et al., 2007). The task of monitoring vegetation
by use of VIs is made feasible due to the convergent properties of plant spectral
responses across functional types (Ustin and Gamon, 2010; Ollinger, 2011). VIs
can be used to estimate biophysical parameters such as leaf area index (LAI) and
fractional vegetation cover (f c ), but these relationships are not universal across
different plant communities and need to be calibrated by ground measurements for
each application.
82
CHANGE DETECTION USING VEGETATION INDICES AND MULTIPLATFORM
spatial resolution imagery such as GeoEye’s IKONOS and Digital Globe’s QuickBird
or WorldView 2 images (∼60 cm) have 1–7-day return times depending on latitude
and tasking mode (Anderson and Marchisio, 2012) but in practice their temporal
resolution can be limited by the high cost of images and scheduling constraints.
Also, as a generality, high spatial resolution at the pixel level often involves a
trade-off in extent of coverage. As examples, the swath width of QuickBird is only
16.5 km, compared to 185 km for Landsat and 2330 km for Moderate Resolution
Imaging Spectroradiometer (MODIS) images from the Terra satellite. In contrast to
satellite imagery, tower-mounted, networked digital cameras (phenocams) provide
near-surface images with nearly continuous temporal resolution and spatial resolution
of 1 cm or less, but extent of coverage is generally limited to 1–100 m
2 (Richardson
et al., 2007; Rundel et al., 2009; Sonnentag et al., 2012).
In an early review of issues of scale in remote sensing, Woodcock and Strahler
(1987) pointed out that in most scientific endeavors the investigator selects the scale at
which observations are collected, whereas investigators using spaceborne sensors are
limited to the specific scales of observations inherent in the satellite imagery. At that
time Landsat and AVHRR satellites were used for most Earth observation studies.
However, with the increase in available sensor systems, as noted in the examples
above, it is now possible to combine imagery to cover a wide range of temporal and
spatial scales across image sets to conduct landscape change studies. More recent
reviews reflect the much broader range of choices now available for obtaining data
at multiple scales (e.g., Mulder et al., 2011; Skidmore et al., 2011; Turner 2011;
Abraha and Savage, 2012; Pfeifer et al., 2012). These reviews emphasize the
problems and opportunities in combining ecological observations and models
across spatial and temporal scales, from the cellular level at response times of
minutes to days (e.g., response of photosynthesis to temperature) to the global level
at response times of decades (e.g., response of global net primary productivity to
global warming).
Another scale issue is spectral resolution. Satellite images tend to have just a few
band windows available and currently operational hyperspectral imagery is mainly
acquired from aircraft and a few experimental satellites. Therefore, the detailed
reflectance spectra that can be used to distinguish different types of soils and
vegetation units are not available from application-oriented satellite imagery.
Among other tools such as spectral mixture analysis, researchers often rely on
vegetation indices (VIs), combinations of two or three bands that separate the
landscape into soil, water, and different densities of green vegetation, based
primarily on differences in their reflectance of light in the adjacent red and
near-infrared (NIR) bands (Glenn et al., 2007). The task of monitoring vegetation
by use of VIs is made feasible due to the convergent properties of plant spectral
responses across functional types (Ustin and Gamon, 2010; Ollinger, 2011). VIs
can be used to estimate biophysical parameters such as leaf area index (LAI) and
fractional vegetation cover (f c ), but these relationships are not universal across
different plant communities and need to be calibrated by ground measurements for
each application.
82
CHANGE DETECTION USING VEGETATION INDICES AND MULTIPLATFORM