193
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
daily and 8-day products (from both Terra and Aqua satellites) in the hydrologic year
2004 and reported accuracies (for SDs ≥4 cm) in a range of 96.3% (MYD10A2) to
98.8% (MOD10A1 and MYD10A1).
The validation studies against in situ observations were biased to the MOD10A1
product. The median of OA of the MOD10A1 validation studies was above 94%.
Larger mapping errors were reported only at a small number of stations, which were
likely affected by specific local meteorological and/or physiographic conditions
(e.g., low solar illumination conditions or false land/water mask along coastline). A
detailed discussion of the source and propagation of MODIS snow mapping errors is
presented by Riggs and Hall (2011). They note that “aside from potential mapping or
geolocation errors, most snow detection errors are associated with non-ideal conditions for snow detection or with snow/cloud discrimination.” Although the problem
with cloud obscuration could be partly alleviated by compositing of MODIS images
(e.g., as in the MODIS 8-day products), the cloud cover and snow/cloud discrimination is still considered the main limitation of MODIS snow cover products.
9.4 METHODS FOR CLOUD IMPACT REDUCTION
The validation studies summarized in the previous section drew two main conclusions.
First, the MODIS snow cover products are, overall, in good agreement with available satellite and ground-based snow data sets. The mapping accuracy depends on the
region and season, but very often, it is within a range that makes the data very useful
and attractive for hydrologic applications. The second conclusion is that clouds may
severely limit the application of MODIS snow cover products. Again, cloud coverage
depends on region and season, but very often, it is a real problem instead of an artifact
of the MODIS snow mapping algorithm. As shown by Parajka and Blöschl (2006),
for example, clouds cover 63% of Austria on the average, and cloud coverage is even
larger in the winter. A similar average cloud cover of about 70% is indicated by Tong
et al. (2009b) for the Quesnel River Basin, 50%–60% for Alaska (Gao et al. 2010b),
or 45% in North America (Zhou et al. 2005). Wang et al. (2009) reported 44%–47%
cloud coverage on the average in the Xinjiang region, which was, interestingly, larger
than 75% during fractional snow conditions. This indicates that the MODIS cloud
mask has the tendency to map edges of areas of patchy or thin snow as cloud.
There is a continuous effort to reduce cloud obscuration in the MODIS snow
data product by improving the cloud mask (e.g., Ackerman et al. 1998; Riggs and
Hall 2003; Lyapustin et al. 2008), which permits more snow to be mapped if it is
present (Hall and Riggs 2007). Extensive testing of liberal cloud masks showed that,
although it provided excellent results in some areas of the globe, it may cause problems in other areas (Hall et al. 2010). Thus, it is not available as part of the most
recent MODIS Collection-5 snow cover product suite. Future revision of the MODIS
mapping algorithm foresees further improvements in the clouds/snow discrimination
technique (Riggs and Hall 2011). However, in many regions, the expected improvements will not be large, as the cloud coverage is real.
An alternative idea of cloud impact reduction in MODIS snow cover products
is based on combining MODIS data in time (temporal filter), space (spatial filter),
or with products from different (multisensor) platforms (e.g., passive microwave
MODIS-Based Snow Cover Products, Validation, and Hydrologic Applications
daily and 8-day products (from both Terra and Aqua satellites) in the hydrologic year
2004 and reported accuracies (for SDs ≥4 cm) in a range of 96.3% (MYD10A2) to
98.8% (MOD10A1 and MYD10A1).
The validation studies against in situ observations were biased to the MOD10A1
product. The median of OA of the MOD10A1 validation studies was above 94%.
Larger mapping errors were reported only at a small number of stations, which were
likely affected by specific local meteorological and/or physiographic conditions
(e.g., low solar illumination conditions or false land/water mask along coastline). A
detailed discussion of the source and propagation of MODIS snow mapping errors is
presented by Riggs and Hall (2011). They note that “aside from potential mapping or
geolocation errors, most snow detection errors are associated with non-ideal conditions for snow detection or with snow/cloud discrimination.” Although the problem
with cloud obscuration could be partly alleviated by compositing of MODIS images
(e.g., as in the MODIS 8-day products), the cloud cover and snow/cloud discrimination is still considered the main limitation of MODIS snow cover products.
9.4 METHODS FOR CLOUD IMPACT REDUCTION
The validation studies summarized in the previous section drew two main conclusions.
First, the MODIS snow cover products are, overall, in good agreement with available satellite and ground-based snow data sets. The mapping accuracy depends on the
region and season, but very often, it is within a range that makes the data very useful
and attractive for hydrologic applications. The second conclusion is that clouds may
severely limit the application of MODIS snow cover products. Again, cloud coverage
depends on region and season, but very often, it is a real problem instead of an artifact
of the MODIS snow mapping algorithm. As shown by Parajka and Blöschl (2006),
for example, clouds cover 63% of Austria on the average, and cloud coverage is even
larger in the winter. A similar average cloud cover of about 70% is indicated by Tong
et al. (2009b) for the Quesnel River Basin, 50%–60% for Alaska (Gao et al. 2010b),
or 45% in North America (Zhou et al. 2005). Wang et al. (2009) reported 44%–47%
cloud coverage on the average in the Xinjiang region, which was, interestingly, larger
than 75% during fractional snow conditions. This indicates that the MODIS cloud
mask has the tendency to map edges of areas of patchy or thin snow as cloud.
There is a continuous effort to reduce cloud obscuration in the MODIS snow
data product by improving the cloud mask (e.g., Ackerman et al. 1998; Riggs and
Hall 2003; Lyapustin et al. 2008), which permits more snow to be mapped if it is
present (Hall and Riggs 2007). Extensive testing of liberal cloud masks showed that,
although it provided excellent results in some areas of the globe, it may cause problems in other areas (Hall et al. 2010). Thus, it is not available as part of the most
recent MODIS Collection-5 snow cover product suite. Future revision of the MODIS
mapping algorithm foresees further improvements in the clouds/snow discrimination
technique (Riggs and Hall 2011). However, in many regions, the expected improvements will not be large, as the cloud coverage is real.
An alternative idea of cloud impact reduction in MODIS snow cover products
is based on combining MODIS data in time (temporal filter), space (spatial filter),
or with products from different (multisensor) platforms (e.g., passive microwave