these SOS measures showed the ability to detect general interannual and large-scale
geographic variations of LSP.
An alternative approach to bridging in situ phenology with satellite phenology
was attempted using Landsat data as an intermediary (Fisher et al. 2006; Fisher
and Mustard 2007). Relatively high spatial resolution (30 m) data from Landsat
TM and ETM ? have a repeat coverage of 16 days (can be reduced to 8 days if
data from multiple platforms are used). Due to the data loss from cloud effects,
estimating vegetation phenology using data from a single year of Landsat data is
not practical. Fisher et al. (2006) utilized multi-year Landsat data to approximate
phenological development over an annual cycle. A modified double logistic model
akin to that of Zhang et al. (2003) was used to fit a growth season curve. Fractional
vegetation cover, which was estimated using a sub-pixel spectral mixture analysis
(SMA) method (estimating vegetation percentage cover at the subpixel level), was
used instead of vegetation indices. Fisher et al. (2006) suggested that the SOS is
marked when vegetation reaches half maximum greenness, similar to the SMN
approach of White et al. (1997).
The Fisher et al. (2006) approach accepted a compromise in using 19-year
composited time series to take advantage of the higher spatial resolution of
Landsat data. In spite of the unavoidable loss of interannual change information
when multi-year data are merged, results seemed to show spatial variations of
phenology with interesting details. In particular, the micro-topographic and microclimatological gradients seem to play an evident role in influencing spring leaf
phenology. Cold air drainage in the New England area delays the phenology of
trees grown in low laying areas. The sequence of phenological development as
observed onsite with both visual estimates and photography agreed with findings
from satellite data. Further, maritime effects seem to be responsible for the delayed
coastal vegetation phenology due to a lag of spring temperature increase. The
urban heat island effect that advances phenology is also documented in the
Landsat-based LSP estimates.
The effort to relate ground and satellite phenology is furthered from considering
only Landsat data to using Landsat data as a bridge to connect to MODIS data
(Fisher and Mustard 2007). This is a necessary step because most of the large-scale
vegetation studies are carried out at coarser spatial resolutions as provided by
MODIS (250 m–1 km) or AVHRR (1 km), which are better suited for global
monitoring. In this follow-up study, Fisher and Mustard (2007) attempted to
address the need of interannual phenology comparison in relation to detecting
climatic influences on phenology. This was not possible in the previous study with
time-integrated Landsat data. The investigation of interannual change mainly
relied on the MODIS time series and field-observed data. The spatial variation was
checked mainly between Landsat-based and MODIS-based phenological measurements. These two sets of independent satellite measurements appeared to show
relatively coherent spatial variations when averaged phenology is of concern. This
study further compared MODIS phenology with field-observed phenology at two
sites, incorporating interannual variations. Year-to-year change of MODIS phenology seemed to match the general variations observed in situ at the field sites,
110
J. M. Hanes et al.
geographic variations of LSP.
An alternative approach to bridging in situ phenology with satellite phenology
was attempted using Landsat data as an intermediary (Fisher et al. 2006; Fisher
and Mustard 2007). Relatively high spatial resolution (30 m) data from Landsat
TM and ETM ? have a repeat coverage of 16 days (can be reduced to 8 days if
data from multiple platforms are used). Due to the data loss from cloud effects,
estimating vegetation phenology using data from a single year of Landsat data is
not practical. Fisher et al. (2006) utilized multi-year Landsat data to approximate
phenological development over an annual cycle. A modified double logistic model
akin to that of Zhang et al. (2003) was used to fit a growth season curve. Fractional
vegetation cover, which was estimated using a sub-pixel spectral mixture analysis
(SMA) method (estimating vegetation percentage cover at the subpixel level), was
used instead of vegetation indices. Fisher et al. (2006) suggested that the SOS is
marked when vegetation reaches half maximum greenness, similar to the SMN
approach of White et al. (1997).
The Fisher et al. (2006) approach accepted a compromise in using 19-year
composited time series to take advantage of the higher spatial resolution of
Landsat data. In spite of the unavoidable loss of interannual change information
when multi-year data are merged, results seemed to show spatial variations of
phenology with interesting details. In particular, the micro-topographic and microclimatological gradients seem to play an evident role in influencing spring leaf
phenology. Cold air drainage in the New England area delays the phenology of
trees grown in low laying areas. The sequence of phenological development as
observed onsite with both visual estimates and photography agreed with findings
from satellite data. Further, maritime effects seem to be responsible for the delayed
coastal vegetation phenology due to a lag of spring temperature increase. The
urban heat island effect that advances phenology is also documented in the
Landsat-based LSP estimates.
The effort to relate ground and satellite phenology is furthered from considering
only Landsat data to using Landsat data as a bridge to connect to MODIS data
(Fisher and Mustard 2007). This is a necessary step because most of the large-scale
vegetation studies are carried out at coarser spatial resolutions as provided by
MODIS (250 m–1 km) or AVHRR (1 km), which are better suited for global
monitoring. In this follow-up study, Fisher and Mustard (2007) attempted to
address the need of interannual phenology comparison in relation to detecting
climatic influences on phenology. This was not possible in the previous study with
time-integrated Landsat data. The investigation of interannual change mainly
relied on the MODIS time series and field-observed data. The spatial variation was
checked mainly between Landsat-based and MODIS-based phenological measurements. These two sets of independent satellite measurements appeared to show
relatively coherent spatial variations when averaged phenology is of concern. This
study further compared MODIS phenology with field-observed phenology at two
sites, incorporating interannual variations. Year-to-year change of MODIS phenology seemed to match the general variations observed in situ at the field sites,
110
J. M. Hanes et al.