different percentage characterization for canopies undergoing a particular
phenological stage transition.
To fully utilize the data from high resolution phenology for the purpose of
validating satellite phenology, adequate scaling methods are required. Liang and
Schwartz (2009) proposed the concept of landscape phenology, which integrates
discrete phenology information to the synthetic levels of ecosystem organization.
Observations for individual plants may be aggregated to form population, community, and landscape phenology representations with additional data related to
species dominance, abundance, and distribution. The concept was fully implemented in a practical scaling ladder that effectively produced landscape phenology
(LP) indices that are readily comparable with LSP metrics (Liang et al. 2011).
Liang et al. (2011) developed a step-wise nested hierarchical scaling approach
linking landscape phenology and hierarchical patch dynamics (Wu and Loucks
1995). This scaling design allowed practicality of data aggregation and simultaneously took into consideration the predefined ecosystem structure and coherence,
as well as the patchiness of forest landscapes.
To achieve each step of scale transition following the individual-populationcommunity-landscape (ecosystem patch) sequence, a suite of digital image processing techniques (with ancillary high resolution imagery) was employed to
characterize the forest landscapes and provide required additional information for
spatially aggregating phenology data (for details see Liang et al. 2011). Two sets
of LP indices were developed with one being compatible with landscape spectral
reflectance and the other retaining field protocol meanings. MODIS VI time series
were first compared with the time series of reflectance calibrated LP index. Then
full bud burst dates as estimated from the LP indices (which retained biological
meanings) for deciduous and coniferous species respectively were compared with
LSP SOS dates. The maximum curvature logistic approach (Zhang et al. 2003) was
used to develop SOS estimates given its use in the MODIS global land cover
dynamics products (Ganguly et al. 2010). With the high resolution field data, the
spatial matching with MODIS satellite pixels was able to be conducted in a more
precise manner. Pixel values were weighted with corresponding spectral contributions of the areal fractions overlapping with the study areas. Detailed comparisons revealed relatively close connections of satellite phenology with the ground
observations. Instead of a linear relationship, the phenological development on the
ground appeared to influence LSP exponentially. The phenology protocol used in
the field describes the progression of canopy growth from buds to leaves. The
primary impact of phenology on surface greening occurs after bud burst and
accelerates with rapid leaf unfolding and expansion. This general trend also agrees
with the initiating pattern of a logistic curve. Coniferous LP index time series
lagged behind those of LSP and deciduous LP index, implying the major role of
deciduous land covers in influencing LSP. Maximum absolute errors between LP
deciduous full bud burst dates and SOS dates were 15 days and 2 days for NDVI
and EVI, respectively.
In addition to the effort to validate LSP using traditional observer-based
phenology, there has been much work done to utilize in situ observations from
4 Land Surface Phenology
113
phenological stage transition.
To fully utilize the data from high resolution phenology for the purpose of
validating satellite phenology, adequate scaling methods are required. Liang and
Schwartz (2009) proposed the concept of landscape phenology, which integrates
discrete phenology information to the synthetic levels of ecosystem organization.
Observations for individual plants may be aggregated to form population, community, and landscape phenology representations with additional data related to
species dominance, abundance, and distribution. The concept was fully implemented in a practical scaling ladder that effectively produced landscape phenology
(LP) indices that are readily comparable with LSP metrics (Liang et al. 2011).
Liang et al. (2011) developed a step-wise nested hierarchical scaling approach
linking landscape phenology and hierarchical patch dynamics (Wu and Loucks
1995). This scaling design allowed practicality of data aggregation and simultaneously took into consideration the predefined ecosystem structure and coherence,
as well as the patchiness of forest landscapes.
To achieve each step of scale transition following the individual-populationcommunity-landscape (ecosystem patch) sequence, a suite of digital image processing techniques (with ancillary high resolution imagery) was employed to
characterize the forest landscapes and provide required additional information for
spatially aggregating phenology data (for details see Liang et al. 2011). Two sets
of LP indices were developed with one being compatible with landscape spectral
reflectance and the other retaining field protocol meanings. MODIS VI time series
were first compared with the time series of reflectance calibrated LP index. Then
full bud burst dates as estimated from the LP indices (which retained biological
meanings) for deciduous and coniferous species respectively were compared with
LSP SOS dates. The maximum curvature logistic approach (Zhang et al. 2003) was
used to develop SOS estimates given its use in the MODIS global land cover
dynamics products (Ganguly et al. 2010). With the high resolution field data, the
spatial matching with MODIS satellite pixels was able to be conducted in a more
precise manner. Pixel values were weighted with corresponding spectral contributions of the areal fractions overlapping with the study areas. Detailed comparisons revealed relatively close connections of satellite phenology with the ground
observations. Instead of a linear relationship, the phenological development on the
ground appeared to influence LSP exponentially. The phenology protocol used in
the field describes the progression of canopy growth from buds to leaves. The
primary impact of phenology on surface greening occurs after bud burst and
accelerates with rapid leaf unfolding and expansion. This general trend also agrees
with the initiating pattern of a logistic curve. Coniferous LP index time series
lagged behind those of LSP and deciduous LP index, implying the major role of
deciduous land covers in influencing LSP. Maximum absolute errors between LP
deciduous full bud burst dates and SOS dates were 15 days and 2 days for NDVI
and EVI, respectively.
In addition to the effort to validate LSP using traditional observer-based
phenology, there has been much work done to utilize in situ observations from
4 Land Surface Phenology
113