195
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
product (MOD10C1), and its effectiveness was tested by data assimilation in the
Noah land surface model. The results showed that the filtered snow-covered product
improved the SD bias efficiency of data assimilation by 8%.
The spatial filter approach replaces pixels classified as clouds by the class (land or
snow) of the majority of noncloud pixels in an eight-pixel neighborhood. The spatial
filter applied to the combined Terra/Aqua product was examined by Parajka and
Blöschl (2008a). They found that the spatial merging resulted in a further 6% reduction in cloud cover and only a slight 0.7% decrease in the OA. Tong et al. (2009b)
applied the spatial filter to the 8-day MOD10A2 product and reported a reduction
of the percentage of cloudy days from 15% to 9% in the Quesnel River Basin. The
percentage of cloudy days was even more reduced with respect to MOD10A1 (see
evaluation in Table 9.3). At the same time, the OA of the spatially filtered product increased by about 2% compared to MOD10A2 and by about 10% compared to
MOD10A1 at higher elevations.
An alternative to spatial filters for cloud impact reduction is the method based on
snow line elevation. This approach assumes that the vertical snow cover distribution
is similar within a region. Parajka et al. (2010) tested the snow line elevation method
over Austria and found that this approach was remarkably robust, including for cases
where only a few percentage of the pixels were cloud free. The cornerstone of this
method is a reclassification of pixels assigned as clouds based on a comparison of
their elevation with the mean elevation of all snow and land pixels. The assessment of
the OA for cloud-free pixels was similar to the MOD10A1 product and only slightly
decreased for cases when clouds covered more than 90% of Austria. When considering clouds as false classification, the decrease in cloud extent can be translated
into a significantly higher mapping performance of the snow line elevation method.
The overall annual accuracy ranged from 48.7% to 81.5%, depending on the cloud
threshold used compared, with 38.5% for the original MOD10A1 product. A more
favorable mapping performance of the snow line approach was found, especially for
cases when the snow cover started to build or melt, which is documented by higher
mapping accuracies in November, December, and April.
The combination of different spatial and temporal filters was examined by
Gafurov and Bárdossy (2009). They tested a sequence of six methods (combination
of Aqua and Terra, temporal and spatial filters, snow line, and climatologic method),
which resulted in total removal of clouds in the Kokcha River Basin. The accuracy
against the artificially masked MOD10A1 product was above 90%.
Multisensor approaches take advantage of the high spatial resolution of MODIS
images and the cloud penetration of passive microwave sensors (Gao et al. 2010a,b).
The resulting maps thus provide daily cloud-free snow cover maps at coarse spatial resolution. The combination of MODIS images with the passive microwave
AMSR-E product is presented in the work of Liang et al. (2008a) and Gao et al.
(2010b). Liang et al. (2008a) reported 75% accuracy of the combined product against
20 in situ observations, instead of 34% accuracy of MOD10A1 in all weather conditions (in all weather condition assessment, the pixels with clouds are considered as
mapping error). An 86% accuracy in all weather conditions was obtained by Gao et
al. (2010b), which was much higher than the 31%, 45%, and 49% accuracies of the
Terra, Aqua, and Terra/Aqua combined snow cover products, respectively.
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
product (MOD10C1), and its effectiveness was tested by data assimilation in the
Noah land surface model. The results showed that the filtered snow-covered product
improved the SD bias efficiency of data assimilation by 8%.
The spatial filter approach replaces pixels classified as clouds by the class (land or
snow) of the majority of noncloud pixels in an eight-pixel neighborhood. The spatial
filter applied to the combined Terra/Aqua product was examined by Parajka and
Blöschl (2008a). They found that the spatial merging resulted in a further 6% reduction in cloud cover and only a slight 0.7% decrease in the OA. Tong et al. (2009b)
applied the spatial filter to the 8-day MOD10A2 product and reported a reduction
of the percentage of cloudy days from 15% to 9% in the Quesnel River Basin. The
percentage of cloudy days was even more reduced with respect to MOD10A1 (see
evaluation in Table 9.3). At the same time, the OA of the spatially filtered product increased by about 2% compared to MOD10A2 and by about 10% compared to
MOD10A1 at higher elevations.
An alternative to spatial filters for cloud impact reduction is the method based on
snow line elevation. This approach assumes that the vertical snow cover distribution
is similar within a region. Parajka et al. (2010) tested the snow line elevation method
over Austria and found that this approach was remarkably robust, including for cases
where only a few percentage of the pixels were cloud free. The cornerstone of this
method is a reclassification of pixels assigned as clouds based on a comparison of
their elevation with the mean elevation of all snow and land pixels. The assessment of
the OA for cloud-free pixels was similar to the MOD10A1 product and only slightly
decreased for cases when clouds covered more than 90% of Austria. When considering clouds as false classification, the decrease in cloud extent can be translated
into a significantly higher mapping performance of the snow line elevation method.
The overall annual accuracy ranged from 48.7% to 81.5%, depending on the cloud
threshold used compared, with 38.5% for the original MOD10A1 product. A more
favorable mapping performance of the snow line approach was found, especially for
cases when the snow cover started to build or melt, which is documented by higher
mapping accuracies in November, December, and April.
The combination of different spatial and temporal filters was examined by
Gafurov and Bárdossy (2009). They tested a sequence of six methods (combination
of Aqua and Terra, temporal and spatial filters, snow line, and climatologic method),
which resulted in total removal of clouds in the Kokcha River Basin. The accuracy
against the artificially masked MOD10A1 product was above 90%.
Multisensor approaches take advantage of the high spatial resolution of MODIS
images and the cloud penetration of passive microwave sensors (Gao et al. 2010a,b).
The resulting maps thus provide daily cloud-free snow cover maps at coarse spatial resolution. The combination of MODIS images with the passive microwave
AMSR-E product is presented in the work of Liang et al. (2008a) and Gao et al.
(2010b). Liang et al. (2008a) reported 75% accuracy of the combined product against
20 in situ observations, instead of 34% accuracy of MOD10A1 in all weather conditions (in all weather condition assessment, the pixels with clouds are considered as
mapping error). An 86% accuracy in all weather conditions was obtained by Gao et
al. (2010b), which was much higher than the 31%, 45%, and 49% accuracies of the
Terra, Aqua, and Terra/Aqua combined snow cover products, respectively.