4.4 Validation Efforts Using In Situ Measurements
Satellite-derived phenology has advantages of covering broad geographic regions,
integrating information from multiple species, and providing regular multi-temporal capability that is useful for change detection. However, common limitations
and signal contaminations associated with remote sensing data reduce the accuracy
of LSP estimates. The primary limitations include the relatively coarse spatial
resolution (250 m–1 km) and the fairly large temporal resolution of useable
observations at a given location caused primarily by clouds. The standard VI
products are usually available as 8–16 day maximum value composites, but there
have been attempts to use daily MODIS surface reflectance to compute customized
NDVI for optimized temporal precision in LSP derivation (Graham et al. 2010; Ju
et al. 2010). Other uncertainties may be from additional atmospheric path radiance
effects, sensor systematic errors, and ground reflectance noises. The different
approaches used for the extraction of phenology metrics may also contain inherent
biases. Therefore, validation of LSP with appropriate ground measurements is
essential for improving phenological monitoring abilities (Schwartz and Reed
1999).
Given the lack of in situ data that are comparable with LSP in spatial coverage
and landscape representativeness, the initial attempt to bridge ground phenology
with satellite phenology was through using bioclimatic models (Schwartz and
Reed 1999; Schwartz et al. 2002). A suite of climate data-driven phenological
models (spring indices, SI) was developed using phenological records of indicator
species: Syringa chinensis (a lilac) and two varieties of honeysuckle (Lonicera
tararica and Lonicera korolkowii) distributed across eastern North America
(Schwartz 1990, 1994, 1997). Schwartz (1994, 1997) outlined the utility of SI as a
‘‘backbone’’ to produce spatially continuous phenological measures for large
regions and thus enable comparison of surface phenology with satellite phenology.
Simulated phenology as derived according to ground-based phenology and independently from LSP serves as a proxy of in situ phenology to compare with LSP
over broad regions where actual phenology data are not available. An initial
comparison was made between satellite-derived SOS and SI first leaf and first
bloom estimates using a land cover stratified correlation approach (Schwartz and
Reed 1999). The first bloom phenology was used as a reference of late spring
events rather than to correspond with the landscape greening as detected by satellites. The SOS dates were calculated from 1 km AVHRR NDVI data using an
updated version of the DMA approach described previously and in Reed et al.
(1994). This study found that satellite-derived SOS was systematically earlier than
the SI first leaf, suggesting that satellite signals record the greening of understory
vegetation before the onset of tree foliage growth. A follow-up study compared SI
with both DMA SOS and SMN SOS over the conterminous United States
(Schwartz et al. 2002). Results pointed out that SMN SOS corresponds more
closely in timing with the late spring phenology event (e.g., SI first bloom date).
Regardless of the lack of explicit linkages with ground biophysical processes,
4 Land Surface Phenology
109
Satellite-derived phenology has advantages of covering broad geographic regions,
integrating information from multiple species, and providing regular multi-temporal capability that is useful for change detection. However, common limitations
and signal contaminations associated with remote sensing data reduce the accuracy
of LSP estimates. The primary limitations include the relatively coarse spatial
resolution (250 m–1 km) and the fairly large temporal resolution of useable
observations at a given location caused primarily by clouds. The standard VI
products are usually available as 8–16 day maximum value composites, but there
have been attempts to use daily MODIS surface reflectance to compute customized
NDVI for optimized temporal precision in LSP derivation (Graham et al. 2010; Ju
et al. 2010). Other uncertainties may be from additional atmospheric path radiance
effects, sensor systematic errors, and ground reflectance noises. The different
approaches used for the extraction of phenology metrics may also contain inherent
biases. Therefore, validation of LSP with appropriate ground measurements is
essential for improving phenological monitoring abilities (Schwartz and Reed
1999).
Given the lack of in situ data that are comparable with LSP in spatial coverage
and landscape representativeness, the initial attempt to bridge ground phenology
with satellite phenology was through using bioclimatic models (Schwartz and
Reed 1999; Schwartz et al. 2002). A suite of climate data-driven phenological
models (spring indices, SI) was developed using phenological records of indicator
species: Syringa chinensis (a lilac) and two varieties of honeysuckle (Lonicera
tararica and Lonicera korolkowii) distributed across eastern North America
(Schwartz 1990, 1994, 1997). Schwartz (1994, 1997) outlined the utility of SI as a
‘‘backbone’’ to produce spatially continuous phenological measures for large
regions and thus enable comparison of surface phenology with satellite phenology.
Simulated phenology as derived according to ground-based phenology and independently from LSP serves as a proxy of in situ phenology to compare with LSP
over broad regions where actual phenology data are not available. An initial
comparison was made between satellite-derived SOS and SI first leaf and first
bloom estimates using a land cover stratified correlation approach (Schwartz and
Reed 1999). The first bloom phenology was used as a reference of late spring
events rather than to correspond with the landscape greening as detected by satellites. The SOS dates were calculated from 1 km AVHRR NDVI data using an
updated version of the DMA approach described previously and in Reed et al.
(1994). This study found that satellite-derived SOS was systematically earlier than
the SI first leaf, suggesting that satellite signals record the greening of understory
vegetation before the onset of tree foliage growth. A follow-up study compared SI
with both DMA SOS and SMN SOS over the conterminous United States
(Schwartz et al. 2002). Results pointed out that SMN SOS corresponds more
closely in timing with the late spring phenology event (e.g., SI first bloom date).
Regardless of the lack of explicit linkages with ground biophysical processes,
4 Land Surface Phenology
109