194
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
products). As clouds vary more quickly in time than the snow cover does, one would
expect that combining the data decreases the cloud coverage significantly. However,
one would also expect that the accuracy of the snow cover maps so obtained would be
lower than that of the original MODIS product because of the time and space shifts
introduced. Table 9.3 summarizes the studies investigating different approaches
and the tradeoff between cloud impact reduction and overall mapping accuracy. It
includes the study region and time period of the evaluation, method used for cloud
impact reduction and cloud coverage, and OA obtained by the reduction approach.
Table 9.3 indicates that numerous studies examined the performance of combining
MODIS data from the Terra and Aqua satellites, whose observations are shifted only
by a few hours. Parajka and Blöschl (2008a) reported a reduction of clouds from 63%
to 51.7% and practically the same 95% OA. Wang et al. (2009) examined the mapping accuracy of the combined (MOD10A1/MYD10A1) MODIS product separately
for land, snow, and fractional snow classification. They found a 7%–17% decrease in
clouds and 7%–10%, 2%–15%, and 7%–17% increases in land, fractional snow, and
snow cover mapping accuracy, respectively. Gao et al. (2010b) investigated the Terra
and Aqua combination over Alaska and found a 7%–12% reduction in clouds. The
corresponding overall clear sky accuracy of the combined product was 92.3%, which
means a slight decrease with respect to the 94.1% accuracy of Terra, but an increase
with respect to the 91.6% accuracy of Aqua. Gao et al. (2010a) assessed the accuracy of the combined product over the Pacific Northwest. They showed that a Terra/
Aqua combination reduced cloud coverage by 5%–14% on monthly and 8%–12% on
annual time scales. The OA of the combined product was 89.7%, which is 0.7% less
and 1.4% more than the reported MOD10A1 and MYD10A1 accuracies, respectively.
Alternative options for temporal merging include the replacement of cloud pixels
by noncloud observations that have occurred at the same pixels within a predefined
temporal window. Different types of fixed or flexibly defined temporal windows have
been tested in recent years. Gao et al. (2010a) tested the accuracy of fixed 2-, 4-, 6-,
and 8-day combined products and found 25% (2-day) to 48% (8-day) cloud impact
reduction in comparison to MOD10A1 and a corresponding 0.9%–2.6% decrease
in accuracy. Parajka and Blöschl (2008a) examined the performance of fixed 1-, 3-,
5- and 7-day windows and reported 18% (1-day) to 47% (7-day) cloud impact reduction, and the corresponding 1.1%–3.4% decrease in OA, respectively. An example of
the seasonal tradeoff between accuracy and cloud impact reduction for January and
October is shown in Figure 9.2.
Flexible temporal filters replace cloud-covered pixels by using multiple MODIS
images until a predefined maximum cloud coverage threshold is reached. Gao et al.
(2010a) tested the performance of the flexible filter with 10% cloud threshold and
reported 34% reduction in clouds and 0.5% decrease in OA with respect to the combined Terra/Aqua product. The same method and cloud threshold were examined
by Wang et al. (2009) and Xie et al. (2009). They reported similar accuracies as for
the standard 8-day product and 25% to 30% cloud impact reduction in Colorado and
Xinjiang, respectively. The average number of images used in the composition was,
however, between 2 and 3. A similar method was presented by Hall et al. (2010), who
replaced the cloud pixels with the most recent cloud-free observations. The cloudgap-filling approach was applied to the 0.05° resolution CMG daily snow cover
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
products). As clouds vary more quickly in time than the snow cover does, one would
expect that combining the data decreases the cloud coverage significantly. However,
one would also expect that the accuracy of the snow cover maps so obtained would be
lower than that of the original MODIS product because of the time and space shifts
introduced. Table 9.3 summarizes the studies investigating different approaches
and the tradeoff between cloud impact reduction and overall mapping accuracy. It
includes the study region and time period of the evaluation, method used for cloud
impact reduction and cloud coverage, and OA obtained by the reduction approach.
Table 9.3 indicates that numerous studies examined the performance of combining
MODIS data from the Terra and Aqua satellites, whose observations are shifted only
by a few hours. Parajka and Blöschl (2008a) reported a reduction of clouds from 63%
to 51.7% and practically the same 95% OA. Wang et al. (2009) examined the mapping accuracy of the combined (MOD10A1/MYD10A1) MODIS product separately
for land, snow, and fractional snow classification. They found a 7%–17% decrease in
clouds and 7%–10%, 2%–15%, and 7%–17% increases in land, fractional snow, and
snow cover mapping accuracy, respectively. Gao et al. (2010b) investigated the Terra
and Aqua combination over Alaska and found a 7%–12% reduction in clouds. The
corresponding overall clear sky accuracy of the combined product was 92.3%, which
means a slight decrease with respect to the 94.1% accuracy of Terra, but an increase
with respect to the 91.6% accuracy of Aqua. Gao et al. (2010a) assessed the accuracy of the combined product over the Pacific Northwest. They showed that a Terra/
Aqua combination reduced cloud coverage by 5%–14% on monthly and 8%–12% on
annual time scales. The OA of the combined product was 89.7%, which is 0.7% less
and 1.4% more than the reported MOD10A1 and MYD10A1 accuracies, respectively.
Alternative options for temporal merging include the replacement of cloud pixels
by noncloud observations that have occurred at the same pixels within a predefined
temporal window. Different types of fixed or flexibly defined temporal windows have
been tested in recent years. Gao et al. (2010a) tested the accuracy of fixed 2-, 4-, 6-,
and 8-day combined products and found 25% (2-day) to 48% (8-day) cloud impact
reduction in comparison to MOD10A1 and a corresponding 0.9%–2.6% decrease
in accuracy. Parajka and Blöschl (2008a) examined the performance of fixed 1-, 3-,
5- and 7-day windows and reported 18% (1-day) to 47% (7-day) cloud impact reduction, and the corresponding 1.1%–3.4% decrease in OA, respectively. An example of
the seasonal tradeoff between accuracy and cloud impact reduction for January and
October is shown in Figure 9.2.
Flexible temporal filters replace cloud-covered pixels by using multiple MODIS
images until a predefined maximum cloud coverage threshold is reached. Gao et al.
(2010a) tested the performance of the flexible filter with 10% cloud threshold and
reported 34% reduction in clouds and 0.5% decrease in OA with respect to the combined Terra/Aqua product. The same method and cloud threshold were examined
by Wang et al. (2009) and Xie et al. (2009). They reported similar accuracies as for
the standard 8-day product and 25% to 30% cloud impact reduction in Colorado and
Xinjiang, respectively. The average number of images used in the composition was,
however, between 2 and 3. A similar method was presented by Hall et al. (2010), who
replaced the cloud pixels with the most recent cloud-free observations. The cloudgap-filling approach was applied to the 0.05° resolution CMG daily snow cover