Results were quantified as the percent of annual green leaf cover or NDVI
*
PC lost
due to defoliation for each year at each camera station. All cameras captured at least a
portion of the initial greening of canopies which started on about April 1 each year and
reached a maximum value by mid-June. This was followed by a rapid loss of nearly all
green leaves during the defoliation period and in some cases a subsequent production
of new green leaves in August and September. Plants then dropped their leaves and
entered winter dormancy in early October. The extent of defoliation was estimated by
projecting each curve over a complete growing season (April 1 to October 15), then
inferring what the shape of the seasonal curve would have been in the absence of
defoliation, by assuming that peak green cover had been reached before beetles were
active, as documented by Hultine et al. (2009) at this site. The “missing” portion of the
curves due to defoliation were then estimated on paper printouts of plots by weighing
cut-out sections of the curves on an analytical balance, following a method used to
determine the area of irregular shapes (e.g., Patil and Bodhe, 2001). For each curve,
the weight of the lost-production cut-out were divided by the weight of the cut-out
representing the total area under the curve times 100 to get the percentage of
reduction.
The defoliation processes as viewed on the Virgin River from wide-angle cameras
is shown in a time series of panchromatic images in Figure 5.2 taken in 2011, the first
full season of defoliation at this site. Plants greened up in May 2011 but were
completely defoliated by July 2011 and did not refoliate by September 1.
On the other hand, plants at the Dolores River (Rio Mesa) site responded to
defoliation by producing new leaves in August of each year, and no net decrease in
spring green-up was noted over three years of defoliation, 2008–2010 (Figure 5.3).
Estimates of lost leaf production from visible-band and NDVI cameras were similar
across years and sites: 32.0 and 30.8% per season, respectively (F = 0.05, p = 0.838,
df = 1,10) (Nagler et al., 2012).
On balance, more information was obtained from the visual green leaf counts than
from the NDVI data. The onset of defoliation was clearly visible on RGB images even
in its early stages, when some of the green leaves abruptly turned brown, indicating
the arrival of beetle. Visual analysis of images also allowed us to distinguish between
leaves subjected to defoliation from those that were lost due to detachment of twigs by
wind and to visually account for differences in greenness caused by shadows within
the canopy or cloud cover. Furthermore, the results were expressed as fractional green
cover, a meaningful biophysical variable, rather than as a derived vegetation index.
Finally, the uncalibrated band reflectance values differed among cameras, and by
scaling NDVI between 0 and 1.0 individually for each camera, it was not possible to
compare values among cameras and sites, whereas the visual green leaf counts could
be compared directly.
5.2.4 Landsat Imagery to Compare NDVI and ET Before and After Beetle
Arrival along Six River Systems
The Landsat series of satellites has provided near-continuous coverage of Earth’s
surface since 1972, with a spatial resolution of 15–60 m in the visible and NIR bands
88
CHANGE DETECTION USING VEGETATION INDICES AND MULTIPLATFORM
*
PC lost
due to defoliation for each year at each camera station. All cameras captured at least a
portion of the initial greening of canopies which started on about April 1 each year and
reached a maximum value by mid-June. This was followed by a rapid loss of nearly all
green leaves during the defoliation period and in some cases a subsequent production
of new green leaves in August and September. Plants then dropped their leaves and
entered winter dormancy in early October. The extent of defoliation was estimated by
projecting each curve over a complete growing season (April 1 to October 15), then
inferring what the shape of the seasonal curve would have been in the absence of
defoliation, by assuming that peak green cover had been reached before beetles were
active, as documented by Hultine et al. (2009) at this site. The “missing” portion of the
curves due to defoliation were then estimated on paper printouts of plots by weighing
cut-out sections of the curves on an analytical balance, following a method used to
determine the area of irregular shapes (e.g., Patil and Bodhe, 2001). For each curve,
the weight of the lost-production cut-out were divided by the weight of the cut-out
representing the total area under the curve times 100 to get the percentage of
reduction.
The defoliation processes as viewed on the Virgin River from wide-angle cameras
is shown in a time series of panchromatic images in Figure 5.2 taken in 2011, the first
full season of defoliation at this site. Plants greened up in May 2011 but were
completely defoliated by July 2011 and did not refoliate by September 1.
On the other hand, plants at the Dolores River (Rio Mesa) site responded to
defoliation by producing new leaves in August of each year, and no net decrease in
spring green-up was noted over three years of defoliation, 2008–2010 (Figure 5.3).
Estimates of lost leaf production from visible-band and NDVI cameras were similar
across years and sites: 32.0 and 30.8% per season, respectively (F = 0.05, p = 0.838,
df = 1,10) (Nagler et al., 2012).
On balance, more information was obtained from the visual green leaf counts than
from the NDVI data. The onset of defoliation was clearly visible on RGB images even
in its early stages, when some of the green leaves abruptly turned brown, indicating
the arrival of beetle. Visual analysis of images also allowed us to distinguish between
leaves subjected to defoliation from those that were lost due to detachment of twigs by
wind and to visually account for differences in greenness caused by shadows within
the canopy or cloud cover. Furthermore, the results were expressed as fractional green
cover, a meaningful biophysical variable, rather than as a derived vegetation index.
Finally, the uncalibrated band reflectance values differed among cameras, and by
scaling NDVI between 0 and 1.0 individually for each camera, it was not possible to
compare values among cameras and sites, whereas the visual green leaf counts could
be compared directly.
5.2.4 Landsat Imagery to Compare NDVI and ET Before and After Beetle
Arrival along Six River Systems
The Landsat series of satellites has provided near-continuous coverage of Earth’s
surface since 1972, with a spatial resolution of 15–60 m in the visible and NIR bands
88
CHANGE DETECTION USING VEGETATION INDICES AND MULTIPLATFORM