Fifteen of the 20 band 6 detectors on the Aqua MODIS failed shortly after launch, a 75 percent signal loss that has precluded using this band for snow detection. However, a Quantitative Image Restoration (QIR) technique was recently developed (Gladkova et al., 2012) that restores Aqua MODIS band 6 data to scientific quality. Version 6 incorporates this technique to produce an intermediate, calibrated radiances product with band 6 restored: MYD02HKM_QIR (this product is not retained). Aside from this step, the snow detection algorithm is the same for Aqua and Terra.
For MYD10_L2, the algorithm detects snow by computing the Normalized Difference Snow Index (NDSI) (Hall and Riggs, 2011) from MYD02HKM_QIR. Data screens are then applied to alleviate errors of commission and to flag uncertain snow detections. The final output consists of NDSI snow cover plus the location of clouds, water bodies, and other algorithm results of interest to data users.
To generate the daily 500 m data set, a gridding algorithm maps all MYD10_L2 swaths into an intermediate snow cover product (MYD10GA) which is then used as input. Starting with this version (Version 6), the algorithms which select the day's best snow cover observation and compute snow albedo have been incorporated into the MYD10GA generation process. Note that MYD10GA is an intermediate product and is not archived at NSIDC.
Once the data have been gridded, the selection algorithm uses several criteria to identify the best observation from the one to several MYD10_L2 swaths which were mapped into each grid cell. The criteria were chosen to obtain the best sensor view of the surface for detecting snow cover: specifically, observations which were acquired nearest local solar noon, nearest the orbit nadir track, and which offer the greatest coverage in the cell. The MYD10GA generation process stores the selected observation's NDSI_Snow_Cover, NDSI_Snow_Cover_Basic_QA, NDSI_Snow_Cover_Basic_QA, and NDSI (raw) as separate SDSs, calculates descriptive QA statistics, and then writes the data and metadata into MYD10A1.
For more information about the MYD10_L2 data set, see the MODIS/Aqua Snow Cover 5-Min L2 Swath 500m, Version 6 documentation.
Although snow albedo in Version 6 is computed during the MYD10GA generation process, the algorithm is the same as Version 5. Once the best MYD10_L2 observations have been selected, snow albedo is calculated for the corresponding pixels in the MYD09GA land-surface reflectance product using the MYD09GA visible and near infrared (VNIR) bands. Land cover type is read from the MODIS combined land cover product (MCDLCHKM) and an anisotropic response function corrects for anisotropic scattering effects in non-forested areas. Snow-covered forests are assumed to be Lambertian reflectors. The snow albedo algorithm is described in Klein and Stroeve, 2002. Additional details about all the MODIS snow cover data sets are available in the Algorithm Theoretical Basis Document (ATBD).
See the MODIS | Data Versions page for the history of MODIS snow and sea ice data versions.
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%. For this data set, choosing a single, best observation of the day results in a weave or stitch pattern along the edges of adjacent swaths. This pattern is most apparent where cloud cover changed between the acquisition times of overlapping swaths. In addition, users may encounter interwoven cloud and clear observations in images with snow cover. Differences in viewing geometry can also produces discontinuities in regions where adjacent swaths overlap.
Geolocation error may be visible due to: a) uncertainty in swath geolocation; and b) the process of gridding and projecting the swaths into the MODIS Sinusoidal Tile Grid from day to day. This latter effect, a so-called geolocation wobble, is most commonly observed as daily shifts in the position of a lake by one or more cells in the horizontal or vertical directions. Thus compositing tiles over the course of several consecutive days may result in blurred lake outlines.
Snow albedo is estimated to be within 10% of surface measured values, based on both published studies (see Klein and Stroeve, 2002 and Tekeli et al., 2006) and unpublished evaluations. However, this estimate assumes optimal conditions for the algorithm, such as a level surface and complete snow cover in the cell. Errors could be much higher where the conditions are less favorable for determining snow albedo, for example over steep mountain terrain. Note that this data set does not report snow albedo-specific QA. The MODIS Science Team is still investigating the best way to express this metric.
Finally, anomalies in the input data can propagate to the output. Table 3 in the MYD10_L2 documentation lists the products that are used as input to the snow cover algorithm. For a more detailed discussion of potential sources of error, including examples, consult the MODIS Snow Products Collection 6 User Guide.