226
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
Second, each of the studies agreed that cloud obscuration is the main limitation
in utilizing snow products from optical sensors including MODIS (Xie et al. 2009;
Wang et al. 2008; Parajka and Blöschl 2008b). As with other optical sensors, the
accuracy of MODIS snow products under cloudy conditions has been the subject
of many studies, because cloud cover obscures snow cover monitoring capability.
According to Xie et al. (2009), the snow classification accuracy of Terra MODIS
daily snow cover product (MOD10A1) was reported to be 44% in all weather conditions, whereas that of Aqua MODIS (MYD10A1) was 34%. Gao et al. (2010b)
reported slightly higher snow accuracies of 46.8% and 39.5% for MOD10A1 and
MYD10A1, respectively, in all weather conditions.
In order to benefit from the high spatial and temporal resolution of MODIS
snow products and also improve the accuracy of MODIS snow data products under
unfavorable conditions caused by clouds, rainfall, or darkness, several researchers
attempted various strategies to enhance the overall accuracy of the MODIS snow
cover product under all weather conditions (Gafurov and Bardossy 2009; Gao et al.
2010a,b; Parajka et al. 2010; Parajka and Blöschl 2008a; Molotch and Margulis 2008;
Hall and Riggs 2007; Dozier et al. 2008). Approaches used to address cloud interference include spatial and/or temporal filtering and combinations of various sensors
(multisensor; e.g., see the work of Gao et al. 2010a and Parajka and Blöschl 2008b).
Combination of MODIS products from the Aqua and Terra satellite platforms
involves merging the two products, observed on the same day shifted by several
hours, on a pixel-by-pixel basis as described by Parajka and Blöschl (2008a). Values
of cloud-obscured pixels from one platform can be replaced by the corresponding
values of cloud-free pixels from the other platform (Gafurov and Bardossy 2009).
In spatial filtering, cloud pixels are replaced by the value held by the majority of
cloud-free pixels (either land or snow) in an eight-pixel neighborhood surrounding
the obscured pixel. In the case of a tie among the values of the neighboring pixels,
Parajka and Blöschl (2008a) assigned the pixel value to be snow covered.
In the third approach, called temporal filtering or temporal deduction, the value
of cloud pixels is replaced with the value from the same pixel under cloud-free conditions from the preceding and following days (Gao et al. 2010a). The idea behind
using this approach is that it allows the use of information from the cloud-free days
to represent the observation during the cloudy days. For example, out of 5 days of
observation, if the first and last 2 days indicate snow coverage while the middle day
is cloudy, the cloud-covered day would be assigned as being snow covered. Three
situations may occur if a pixel is obstructed by clouds in the image of a particular
day and the preceding and following days are clear. The situations and the temporal
filtering in the work of Gao et al. (2010a) are as follows:
1. If the corresponding pixel values in both images of the preceding and following
days have snow, the cloud pixel of the current day is presumed to have snow.
2. If the corresponding pixel values in both images of the preceding and following days are land, the cloud pixel of the current day is presumed to be land.
3. If the corresponding pixel values in both images of the preceding and following days are different, for example, one is snow and the other is land, the
cloud pixel is left unchanged.
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
Second, each of the studies agreed that cloud obscuration is the main limitation
in utilizing snow products from optical sensors including MODIS (Xie et al. 2009;
Wang et al. 2008; Parajka and Blöschl 2008b). As with other optical sensors, the
accuracy of MODIS snow products under cloudy conditions has been the subject
of many studies, because cloud cover obscures snow cover monitoring capability.
According to Xie et al. (2009), the snow classification accuracy of Terra MODIS
daily snow cover product (MOD10A1) was reported to be 44% in all weather conditions, whereas that of Aqua MODIS (MYD10A1) was 34%. Gao et al. (2010b)
reported slightly higher snow accuracies of 46.8% and 39.5% for MOD10A1 and
MYD10A1, respectively, in all weather conditions.
In order to benefit from the high spatial and temporal resolution of MODIS
snow products and also improve the accuracy of MODIS snow data products under
unfavorable conditions caused by clouds, rainfall, or darkness, several researchers
attempted various strategies to enhance the overall accuracy of the MODIS snow
cover product under all weather conditions (Gafurov and Bardossy 2009; Gao et al.
2010a,b; Parajka et al. 2010; Parajka and Blöschl 2008a; Molotch and Margulis 2008;
Hall and Riggs 2007; Dozier et al. 2008). Approaches used to address cloud interference include spatial and/or temporal filtering and combinations of various sensors
(multisensor; e.g., see the work of Gao et al. 2010a and Parajka and Blöschl 2008b).
Combination of MODIS products from the Aqua and Terra satellite platforms
involves merging the two products, observed on the same day shifted by several
hours, on a pixel-by-pixel basis as described by Parajka and Blöschl (2008a). Values
of cloud-obscured pixels from one platform can be replaced by the corresponding
values of cloud-free pixels from the other platform (Gafurov and Bardossy 2009).
In spatial filtering, cloud pixels are replaced by the value held by the majority of
cloud-free pixels (either land or snow) in an eight-pixel neighborhood surrounding
the obscured pixel. In the case of a tie among the values of the neighboring pixels,
Parajka and Blöschl (2008a) assigned the pixel value to be snow covered.
In the third approach, called temporal filtering or temporal deduction, the value
of cloud pixels is replaced with the value from the same pixel under cloud-free conditions from the preceding and following days (Gao et al. 2010a). The idea behind
using this approach is that it allows the use of information from the cloud-free days
to represent the observation during the cloudy days. For example, out of 5 days of
observation, if the first and last 2 days indicate snow coverage while the middle day
is cloudy, the cloud-covered day would be assigned as being snow covered. Three
situations may occur if a pixel is obstructed by clouds in the image of a particular
day and the preceding and following days are clear. The situations and the temporal
filtering in the work of Gao et al. (2010a) are as follows:
1. If the corresponding pixel values in both images of the preceding and following
days have snow, the cloud pixel of the current day is presumed to have snow.
2. If the corresponding pixel values in both images of the preceding and following days are land, the cloud pixel of the current day is presumed to be land.
3. If the corresponding pixel values in both images of the preceding and following days are different, for example, one is snow and the other is land, the
cloud pixel is left unchanged.