Processing Steps
The algorithm maps 500 m MOD10A1 observations into 0.05° (approx. 5 km) CMG cells and bins and counts each observation by type, for example snow cover, cloud cover, and snow-free land. For the purpose of counting, the binning algorithm converts NDSI snow cover to a snow/no snow flag assuming that an NDSI > 0 indicates the presence of at least some snow. Snow and cloud cover extent are computed as the ratio of the number of snow or cloud observations to the total number of land observations that were mapped into the CMG cell. These ratios are expressed as percentages in the SDSs.
The snow map also includes lake ice coverage. The number of inland water body observations are counted using the water flag in the MOD10A1 NDSI_Snow_Cover_Algorithm_Flags_QA
SDS. If the water body has more lake ice observations than open water, it is interpreted as lake ice and a value of 107 is set in the output. Lakes that are cloud obscured are output as cloud obscured with a value of 250.
A CMG-specific, 0.05° land mask is used with the binning algorithm. This land mask was derived from the University of Maryland 1km Land Cover data set. CMG cells which contain 12% or greater land are considered land and analyzed; cells with less than 12% land are considered ocean. This threshold was selected as a balance between remaining sensitive enough to map snow along coasts and minimizing snow detection errors in these regions.
Viewing conditions in the CMG cell, relative to cloud cover, are represented by the Clear Index (CI). This value reports the percentage of all land observations in the cell that were clear, thus providing an estimate of the amount of land surface that was observable. Though calculated independently from observation counts, the clear index is essentially 100 minus the percentage of cloud cover and can be used to assess the quality of that cell's snow cover value. A high CI indicates predominantly clear-sky conditions; low values correspond to cells with extensive cloud cover and indicate that the snow cover estimate was derived from a partial view of the land surface.
Polar darkness extent is based on the latitude of the CMG cell nearest the equator that is full of night observations. All CMG cells poleward of that latitude are filled as night. This approach was adopted so that a neat demarcation of night and day is visible in the CMG.
Antarctica has been masked as 100% snow covered to improve the visual quality of data. As such, this data set cannot be used to map snow in Antarctica. For users who wish to evaluate Antarctica, the MOD10_L2 data set offers a higher resolution and contains more data and information about accuracy and error.
Finally, a global mask is applied at the end of the algorithm to eliminate erroneous snow cover detections in regions where snow is extremely unlikely, such as the Amazon, the Sahara, and the Great Sandy Desert. The presence of snow cover in these regions stems from erroneous snow detections in the MOD10_L2 data set which are carried forward through the processing chain to the CMG. The mask is specifically designed to eliminate extremely unlikely snow cover in the CMG while allowing it in regions where snow may be a rare event.
Version History
See the MODIS | Data Versions page for the history of MODIS snow and sea ice data versions.
Error Sources
The NDSI technique has proven to be a robust indicator of snow cover. Numerous investigators have utilized MODIS snow cover data sets and reported accuracy in the range of 88% to 93%. The daily CMG offers a synoptic view of snow cover extent, with cloud cover and optionally the clear index which can be added to construct a synoptic view of snow cover with the cloud mask overlaid. Snow cover and cloud cover are written to separate data arrays so that users may interpret or combine the data which is most relevant to their research or applications. Snow commission errors are typically associated with cloud cover and thus snow errors may appear on any day in conjunction with cloud cover. Users should consider how best to interpret and use the snow cover data, or whether to combine it with the cloud cover data.
Because of the great difficulty in discriminating between clouds and snow over Antarctica in the swath-level snow detection and cloud mask algorithms, data quality is low over Antarctica and thus masked as 100% snow cover. Although masking improves the visual quality of the image, this approach excludes using the data for scientific study in Antarctica. In addition, to reduce erroneous snow cover detections in regions of the world that climatologically should never have snow, a putative snow “impossible” mask is applied in the algorithm. This mask improves the synoptic quality of the product, but at the expense of detecting unprecedented snowfall with this data set. The MOD10_L2/MYD10_L2 and MOD10A1/MOD10A1 data sets should be used to investigate such an event.
Snow errors are ultimately propagated from MOD10_L2 to MOD10A1 and then into this data set. For more detail about potential error sources in the input data, see the Derivation Techniques and Algorithms section in the MOD10_L2 documentation and the MODIS Snow Products Collection 6 User Guide.