205
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
limitations of the model structure and different types of model inputs. Su et al. (2008)
assimilated MODIS snow cover fraction into continental SWE fields simulated by a
highly complex land surface model. The evaluation over North America showed that
the assimilation method more accurately simulated the seasonal variability of SWE
and reduced the uncertainties in the ensemble spread. Kuchment et al. (2010) applied
MOD10L2 data for validating and refining the parameterization of a physically based
snowpack model. They found that the model allowed a satisfactory reproduction of
SCA temporal changes for open areas, but a decreasing accuracy was found when
forested pixels were included in the evaluation. The effect of forest was also examined by Roy et al. (2010), who reported an underestimation of SCA in the forested
study region. They assimilated SCA from MODIS into a one-layer energy budget
model and found that the direct assimilation improved the stream flow simulation for
the spring periods. The runoff model efficiency and stream flow peak identification
improved by 0.11%–0.13% and 19%–36%, respectively.
The methodology used for assimilating MODIS data into hydrologic models needs to account for the differences between these two snow representations.
Typically, hydrologic models simulate the amount (volume) of water stored in the
form of snow (in millimeters SWE), whereas the MODIS snow cover data show
only whether the spatial unit of the snow mapping (pixel) is covered by snow or land
or is classified as missing information (mostly clouds). The main implication of the
different representations is that some relationship between the (modeled) SWE and
the presence of snow at the pixel scale (from MODIS) needs to be established. This
relationship usually takes on the form of thresholds (i.e., no snow coverage assumed
below an SWE threshold; catchment assumed snow-free if percentage of SCA of a
catchment is below an SCA threshold). Sensitivity analyses presented by Parajka and
Blöschl (2008b) and Nester et al. (in press) indicated that the magnitude of the snow
model efficiency was sensitive to the choice of the SWE threshold but not sensitive
to the choice of the SCA threshold. The analysis of the seasonal distribution of snow
underestimation errors indicated that the MODIS misclassification errors, especially
in the summer months, may significantly affect the magnitude of the snow model
efficiency. Parajka and Blöschl (2008b) hence suggested a 25% threshold value of
SCA for robust snow underestimation error assessment. Roy et al. (2010) tested a
simple direct insertion assimilation approach based on an empirical SWE threshold
compensating for the small amount of snow that satellite sensors cannot identify during the melting period. They found the best runoff model performance when a 3–6
cm threshold was added to the model in the case that MODIS indicated snow and the
model indicated less snow than this threshold.
The selection of the cloud threshold affects how much information is used in the
evaluation of the snow model performance and how representative the MODIS SCA
is. Parajka and Blöschl (2008b) and Su et al. (2008) found that a 60% and 50% cloud
cover threshold, respectively, is a reasonable compromise between snow data availability and SCA robustness. Udnaes et al. (2007) and Roy et al. (2010) estimated and
integrated snow cover data into hydrologic modeling only when clouds obscured less
than 30% of the catchment. Andreadis and Lettenmaier (2006) used a 20% cloud
cover threshold to decide whether or not it is necessary to assimilate the MODIS
observations into the macroscale hydrologic model. On the other hand, Rodell and
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
limitations of the model structure and different types of model inputs. Su et al. (2008)
assimilated MODIS snow cover fraction into continental SWE fields simulated by a
highly complex land surface model. The evaluation over North America showed that
the assimilation method more accurately simulated the seasonal variability of SWE
and reduced the uncertainties in the ensemble spread. Kuchment et al. (2010) applied
MOD10L2 data for validating and refining the parameterization of a physically based
snowpack model. They found that the model allowed a satisfactory reproduction of
SCA temporal changes for open areas, but a decreasing accuracy was found when
forested pixels were included in the evaluation. The effect of forest was also examined by Roy et al. (2010), who reported an underestimation of SCA in the forested
study region. They assimilated SCA from MODIS into a one-layer energy budget
model and found that the direct assimilation improved the stream flow simulation for
the spring periods. The runoff model efficiency and stream flow peak identification
improved by 0.11%–0.13% and 19%–36%, respectively.
The methodology used for assimilating MODIS data into hydrologic models needs to account for the differences between these two snow representations.
Typically, hydrologic models simulate the amount (volume) of water stored in the
form of snow (in millimeters SWE), whereas the MODIS snow cover data show
only whether the spatial unit of the snow mapping (pixel) is covered by snow or land
or is classified as missing information (mostly clouds). The main implication of the
different representations is that some relationship between the (modeled) SWE and
the presence of snow at the pixel scale (from MODIS) needs to be established. This
relationship usually takes on the form of thresholds (i.e., no snow coverage assumed
below an SWE threshold; catchment assumed snow-free if percentage of SCA of a
catchment is below an SCA threshold). Sensitivity analyses presented by Parajka and
Blöschl (2008b) and Nester et al. (in press) indicated that the magnitude of the snow
model efficiency was sensitive to the choice of the SWE threshold but not sensitive
to the choice of the SCA threshold. The analysis of the seasonal distribution of snow
underestimation errors indicated that the MODIS misclassification errors, especially
in the summer months, may significantly affect the magnitude of the snow model
efficiency. Parajka and Blöschl (2008b) hence suggested a 25% threshold value of
SCA for robust snow underestimation error assessment. Roy et al. (2010) tested a
simple direct insertion assimilation approach based on an empirical SWE threshold
compensating for the small amount of snow that satellite sensors cannot identify during the melting period. They found the best runoff model performance when a 3–6
cm threshold was added to the model in the case that MODIS indicated snow and the
model indicated less snow than this threshold.
The selection of the cloud threshold affects how much information is used in the
evaluation of the snow model performance and how representative the MODIS SCA
is. Parajka and Blöschl (2008b) and Su et al. (2008) found that a 60% and 50% cloud
cover threshold, respectively, is a reasonable compromise between snow data availability and SCA robustness. Udnaes et al. (2007) and Roy et al. (2010) estimated and
integrated snow cover data into hydrologic modeling only when clouds obscured less
than 30% of the catchment. Andreadis and Lettenmaier (2006) used a 20% cloud
cover threshold to decide whether or not it is necessary to assimilate the MODIS
observations into the macroscale hydrologic model. On the other hand, Rodell and