234
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
or simulated snowmelt runoff from the base period. Therefore, SCA estimates are
required for the base period in order to run the historical simulations. However, this
base period predates the launch of the Terra satellite, so MODIS snow cover measurements were not made at that time, and S curves or CDCs must be synthesized by
some alternatives.
One possible alternative may be derived from a comparison of time series of
SWE data from SNOTEL stations with CDCs from MODIS data in the years they
are available. Unlike the MODIS CDCs that represent basin or zone spatial aggregations, SNOTEL station time series represent point measurements of SD or, similarly, SWE. Nevertheless, the behavior of an individual SNOTEL station’s SWE time
series relative to the entire basin or zone is often consistent.
The strongest direct connection between individual station SWE and MODIS
SCA corresponds to the point in time when snow disappears from a SNOTEL station, that is, the case when SWE becomes zero. Ideally, this would correspond to
the day when the MODIS snow cover pixel corresponding to the location of the
SNOTEL station would change from “snow” to “snow-free land.”
There are spatial and temporal scale differences between MODIS and SNOTEL
data, namely, we have used the MODIS 8-day temporal filter product whose spatial
resolution is 500-m at nadir, whereas SNOTEL measurements cover 7 m 2 and are
taken in daily time steps here. These scale differences along with random variability prevent perfect pixel-to-SNOTEL-station correlation. Taking a more statistical
approach, we assumed that each station generally corresponds to the same fraction
of the basin or zone that melts out concurrently. In other words, a station that melts
out early 1 year, say around the date, t, when S(t) reduces to 0.9, will generally melt
out with the first 10% of area every year; similarly, stations that melt out approximately with the 30th, 50th, or 90th percentiles of area do so relatively consistently
from year to year. Others have pointed out that factors such as invariant topography
and physically forced wind directions might produce consistent spatial patterns of
melt out from year to year (Kirnbauer and Bloschl 1994; Sturm et al. 1995) and that
persistence of spatial patterns may enable forms of depletion curves that are standard from year to year (Luce et al. 1999; Luce and Tarboton 2004). In the present
application, this spatial consistency is demonstrated in Figure 10.7, which displays
the time series of 2003 MODIS Snow Cover Fraction and the fitted Gaussian S curve
in Figure 10.6, SRM zone 2.
In addition, the melt-out dates of all 11 SNOTEL stations within the basin are
plotted directly on the Gaussian curve in Figure 10.7. This is done by assigning the
same snow cover fraction, S(t), to each melt-out date as that corresponding to the
Gaussian curve on the same day. Zone snow–covered fraction on the day of melt
out for a given station is assumed to be consistent from year to year. Consequently,
each year, it is assumed that, regardless of when it actually occurs, the melt-out date
for a given station corresponds to the same zone snow cover fraction as it did in
2003. Figures 10.8 through 10.10 plot the melt-out dates from the subsequent years
2004–2006 with the same snow cover fraction as assigned to the station from 2003.
These are plotted together with the zone 2 MODIS snow-covered fractions from
the corresponding year. Despite modest scatter, there is good agreement between
the SNOTEL and MODIS data in each panel. It is important to note that, apart from
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
or simulated snowmelt runoff from the base period. Therefore, SCA estimates are
required for the base period in order to run the historical simulations. However, this
base period predates the launch of the Terra satellite, so MODIS snow cover measurements were not made at that time, and S curves or CDCs must be synthesized by
some alternatives.
One possible alternative may be derived from a comparison of time series of
SWE data from SNOTEL stations with CDCs from MODIS data in the years they
are available. Unlike the MODIS CDCs that represent basin or zone spatial aggregations, SNOTEL station time series represent point measurements of SD or, similarly, SWE. Nevertheless, the behavior of an individual SNOTEL station’s SWE time
series relative to the entire basin or zone is often consistent.
The strongest direct connection between individual station SWE and MODIS
SCA corresponds to the point in time when snow disappears from a SNOTEL station, that is, the case when SWE becomes zero. Ideally, this would correspond to
the day when the MODIS snow cover pixel corresponding to the location of the
SNOTEL station would change from “snow” to “snow-free land.”
There are spatial and temporal scale differences between MODIS and SNOTEL
data, namely, we have used the MODIS 8-day temporal filter product whose spatial
resolution is 500-m at nadir, whereas SNOTEL measurements cover 7 m 2 and are
taken in daily time steps here. These scale differences along with random variability prevent perfect pixel-to-SNOTEL-station correlation. Taking a more statistical
approach, we assumed that each station generally corresponds to the same fraction
of the basin or zone that melts out concurrently. In other words, a station that melts
out early 1 year, say around the date, t, when S(t) reduces to 0.9, will generally melt
out with the first 10% of area every year; similarly, stations that melt out approximately with the 30th, 50th, or 90th percentiles of area do so relatively consistently
from year to year. Others have pointed out that factors such as invariant topography
and physically forced wind directions might produce consistent spatial patterns of
melt out from year to year (Kirnbauer and Bloschl 1994; Sturm et al. 1995) and that
persistence of spatial patterns may enable forms of depletion curves that are standard from year to year (Luce et al. 1999; Luce and Tarboton 2004). In the present
application, this spatial consistency is demonstrated in Figure 10.7, which displays
the time series of 2003 MODIS Snow Cover Fraction and the fitted Gaussian S curve
in Figure 10.6, SRM zone 2.
In addition, the melt-out dates of all 11 SNOTEL stations within the basin are
plotted directly on the Gaussian curve in Figure 10.7. This is done by assigning the
same snow cover fraction, S(t), to each melt-out date as that corresponding to the
Gaussian curve on the same day. Zone snow–covered fraction on the day of melt
out for a given station is assumed to be consistent from year to year. Consequently,
each year, it is assumed that, regardless of when it actually occurs, the melt-out date
for a given station corresponds to the same zone snow cover fraction as it did in
2003. Figures 10.8 through 10.10 plot the melt-out dates from the subsequent years
2004–2006 with the same snow cover fraction as assigned to the station from 2003.
These are plotted together with the zone 2 MODIS snow-covered fractions from
the corresponding year. Despite modest scatter, there is good agreement between
the SNOTEL and MODIS data in each panel. It is important to note that, apart from