(Goward et al. 1985; Tarpley 1991; Eidenshink 1992). AVHRR acquires global
coverage daily with its broad scanning swath (2700 km) and as a tradeoff it has a
relatively coarse (1.1 km) spatial resolution. At the turn of the century, a type of
sensor with improved spectral, radiometric, and geometric quality was implemented. The MODIS onboard NASA’s Terra (1999–present) and Aqua (2002–
present) satellites began operating to provide better remote sensing data for
multidisciplinary research on global change (Justice et al. 1998). The MODIS
instruments provide similar global and near-daily repeat coverage and improved
spatial resolutions: 250 m (bands 1–2), 500 m (bands 3–7), and 1 km (bands
8–36). In addition, the EVI was developed along with the implementation of this
sensor, increasing the ability to extract vegetation information. Both AVHRR and
MODIS allow frequent monitoring of vegetation conditions for large geographic
regions and, therefore, are ideal for LSP monitoring at continental and global
scales. However, the presence of clouds has been a major hindrance to acquiring
high temporal resolution image time series of vegetation at a given location. A
maximum value composite method is hence employed to remove cloud contamination for both AVHRR and MODIS vegetation indices time series, with nominal
temporal resolutions of resultant products reduced to about 2 weeks (Thayn and
Price 2008).
Specifically, the available vegetation indices (VI) values from satellite observation over a year comprise time series corresponding to annual growing seasons
of vegetation. Given the cloud contamination problem mentioned previously, the
cloud-free images that allow vegetation of a specific location to be seen are often
few for a growing season. Besides atmospheric interference, the quality of data is
also affected by noises from sensor systems, surface properties, and solar and
sensor viewing angles. Therefore data smoothing approaches from simple running
median to curve-fitting models are employed as a necessary procedure in image
preprocessing (Reed et al. 1994; Bradley et al. 2007). Obvious spurious data points
such as extremely low VI values caused by snow, clouds, or water bodies can also
be removed manually. Such an approach is meant to reduce the signal contaminations and retain the essence of annual and interannual variations of phenology.
As stated earlier, VI products available are typically processed with the maximum
value composite method to remove cloud effects (Holben 1986). For a 16-day
composite window, the maximum VI value is retained to indicate the vegetation
condition during that time period. This temporal limitation of applicable data
entails the use of interpolation approaches for generating continuous VI curves for
LSP estimation.
Curve geometry of VI time series is used for further extracting satellite pixelbased phenological patterns, because VI values fluctuate in annual cycles in
accordance with vegetation phenology. For instance, the rise and drop of VI levels
reflect respectively the increase and decrease of vegetation activities. Conventional
phenology studies pay most attention to the critical annual events, such as leaf bud
burst, flower bloom, and sometimes leaf fall and growing season length. Remotely
sensed phenology has been mostly focused on leaf phenology, given the spatial
resolution available that makes detecting flower phenology in mixed landscapes
4 Land Surface Phenology
105
coverage daily with its broad scanning swath (2700 km) and as a tradeoff it has a
relatively coarse (1.1 km) spatial resolution. At the turn of the century, a type of
sensor with improved spectral, radiometric, and geometric quality was implemented. The MODIS onboard NASA’s Terra (1999–present) and Aqua (2002–
present) satellites began operating to provide better remote sensing data for
multidisciplinary research on global change (Justice et al. 1998). The MODIS
instruments provide similar global and near-daily repeat coverage and improved
spatial resolutions: 250 m (bands 1–2), 500 m (bands 3–7), and 1 km (bands
8–36). In addition, the EVI was developed along with the implementation of this
sensor, increasing the ability to extract vegetation information. Both AVHRR and
MODIS allow frequent monitoring of vegetation conditions for large geographic
regions and, therefore, are ideal for LSP monitoring at continental and global
scales. However, the presence of clouds has been a major hindrance to acquiring
high temporal resolution image time series of vegetation at a given location. A
maximum value composite method is hence employed to remove cloud contamination for both AVHRR and MODIS vegetation indices time series, with nominal
temporal resolutions of resultant products reduced to about 2 weeks (Thayn and
Price 2008).
Specifically, the available vegetation indices (VI) values from satellite observation over a year comprise time series corresponding to annual growing seasons
of vegetation. Given the cloud contamination problem mentioned previously, the
cloud-free images that allow vegetation of a specific location to be seen are often
few for a growing season. Besides atmospheric interference, the quality of data is
also affected by noises from sensor systems, surface properties, and solar and
sensor viewing angles. Therefore data smoothing approaches from simple running
median to curve-fitting models are employed as a necessary procedure in image
preprocessing (Reed et al. 1994; Bradley et al. 2007). Obvious spurious data points
such as extremely low VI values caused by snow, clouds, or water bodies can also
be removed manually. Such an approach is meant to reduce the signal contaminations and retain the essence of annual and interannual variations of phenology.
As stated earlier, VI products available are typically processed with the maximum
value composite method to remove cloud effects (Holben 1986). For a 16-day
composite window, the maximum VI value is retained to indicate the vegetation
condition during that time period. This temporal limitation of applicable data
entails the use of interpolation approaches for generating continuous VI curves for
LSP estimation.
Curve geometry of VI time series is used for further extracting satellite pixelbased phenological patterns, because VI values fluctuate in annual cycles in
accordance with vegetation phenology. For instance, the rise and drop of VI levels
reflect respectively the increase and decrease of vegetation activities. Conventional
phenology studies pay most attention to the critical annual events, such as leaf bud
burst, flower bloom, and sometimes leaf fall and growing season length. Remotely
sensed phenology has been mostly focused on leaf phenology, given the spatial
resolution available that makes detecting flower phenology in mixed landscapes
4 Land Surface Phenology
105