with satellite-derived onset dates situated between the earlier bud break-based
onset dates and later leaf expansion-based onset dates.
Two notable efforts to validate different LSP methodologies developed for
AVHRR NDVI and MODIS NDVI/EVI highlighted large uncertainties in the
connection between satellite and ground phenology measurements (White et al.
2009; Schwartz and Hanes 2010). Both studies were conducted at the continental
scale and were specifically aimed towards validating SOS estimates. White et al.
(2009) performed an inter-comparison among 10 selected SOS derivation techniques using 8 km 15-day composite AVHRR NDVI data for 1982–2006. Groundbased measurements used included phenology network records, phenology model
predictions, as well as cryospheric/hydrologic seasonality data. Results indicated
that SOS dates predicted by different methods deviate from each other up to
2 months, in addition to large variations across geographic regions. Schwartz and
Hanes (2010) utilized 1 km 16-day composite MODIS NDVI and EVI data from
2000 to 2006, and compared 10 sets of SOS estimates derived from selected
methods. Some important SOS methods, such as those using double logistic
functions (Zhang et al. 2003; Fisher et al. 2006) and modified TIMESAT (Tan
et al. 2008), which were missing in the analyses of White et al. (2009), were
covered. The inter-comparison was made for eastern North America against surface data, which included both modeled phenology (SI) and local species phenology data from several study sites across the region. The comparison results
alike suggested a lack of significant correlations between the satellite and ground
measurements and the performance of different SOS methods varied across sites.
Though targeting on LSP estimates from different satellite data sources, both
studies noted here implied that the differences in spatial scales and temporal
resolutions of satellite and ground data are a major hurdle to validation tasks.
These studies also demonstrated that ground phenology data as collected in an
extensive manner at discrete locations (e.g., from the Nature’s Notebook of the
USA National Phenology Network) need to be complemented with more detailed
field observations in order to be used for validating LSP.
More specifically, remote sensing and in situ observation of plant phenology
both investigate essentially the same phenomena, but deviate in coverage and
resolution, as well as in the forms of information recorded. Observer-based
recording of phenology relies on the synthetic capability of human eyes in
detecting integrated phenological development phases, such as bud burst or leaf
expansion. Often some statistical criteria are employed to guide the characterization of differential growth within the same canopy. Human eyes do not provide
quantitatively precise recordings when multiple observations are translated into a
series of values, but do capture ordinal progression sequences. For key phenological events, visual observations are most reliable. Remote sensing, on the other
hand, would allow quantification of vegetation growth at more differentiable levels
as allowed by the sensor’s radiometric resolution, but lacks the direct correspondence with traditionally defined phenological events. In addition, the majority of
local species ground phenology data from observation networks are available for
limited plant individuals registered to approximate geographic locations, while
4 Land Surface Phenology
111
onset dates and later leaf expansion-based onset dates.
Two notable efforts to validate different LSP methodologies developed for
AVHRR NDVI and MODIS NDVI/EVI highlighted large uncertainties in the
connection between satellite and ground phenology measurements (White et al.
2009; Schwartz and Hanes 2010). Both studies were conducted at the continental
scale and were specifically aimed towards validating SOS estimates. White et al.
(2009) performed an inter-comparison among 10 selected SOS derivation techniques using 8 km 15-day composite AVHRR NDVI data for 1982–2006. Groundbased measurements used included phenology network records, phenology model
predictions, as well as cryospheric/hydrologic seasonality data. Results indicated
that SOS dates predicted by different methods deviate from each other up to
2 months, in addition to large variations across geographic regions. Schwartz and
Hanes (2010) utilized 1 km 16-day composite MODIS NDVI and EVI data from
2000 to 2006, and compared 10 sets of SOS estimates derived from selected
methods. Some important SOS methods, such as those using double logistic
functions (Zhang et al. 2003; Fisher et al. 2006) and modified TIMESAT (Tan
et al. 2008), which were missing in the analyses of White et al. (2009), were
covered. The inter-comparison was made for eastern North America against surface data, which included both modeled phenology (SI) and local species phenology data from several study sites across the region. The comparison results
alike suggested a lack of significant correlations between the satellite and ground
measurements and the performance of different SOS methods varied across sites.
Though targeting on LSP estimates from different satellite data sources, both
studies noted here implied that the differences in spatial scales and temporal
resolutions of satellite and ground data are a major hurdle to validation tasks.
These studies also demonstrated that ground phenology data as collected in an
extensive manner at discrete locations (e.g., from the Nature’s Notebook of the
USA National Phenology Network) need to be complemented with more detailed
field observations in order to be used for validating LSP.
More specifically, remote sensing and in situ observation of plant phenology
both investigate essentially the same phenomena, but deviate in coverage and
resolution, as well as in the forms of information recorded. Observer-based
recording of phenology relies on the synthetic capability of human eyes in
detecting integrated phenological development phases, such as bud burst or leaf
expansion. Often some statistical criteria are employed to guide the characterization of differential growth within the same canopy. Human eyes do not provide
quantitatively precise recordings when multiple observations are translated into a
series of values, but do capture ordinal progression sequences. For key phenological events, visual observations are most reliable. Remote sensing, on the other
hand, would allow quantification of vegetation growth at more differentiable levels
as allowed by the sensor’s radiometric resolution, but lacks the direct correspondence with traditionally defined phenological events. In addition, the majority of
local species ground phenology data from observation networks are available for
limited plant individuals registered to approximate geographic locations, while
4 Land Surface Phenology
111