Persistence-based temporal filtering for MODIS snow products

- Automated methodology for producing cloud-sparse daily snow covered area maps
- Mitigate errors in snow/cloud classification using temporal persistence of SCA
- Methodology significantly improves MODSCAG consistency/accuracy.
- Results validated by comparing to manually produced/quality-assured maps

Single-day snow covered area (SCA) products are incomplete and often inadequate representations of ground conditions due to short term variation in cloud cover, snow cover, and sensor geometry. To mitigate these effects, we developed a by-pixel filtering algorithm to produce relatively cloud-free SCA products from 16 days of MODIS imagery. The algorithm uses previous days' data to estimate the current SCA value of each pixel and uses a simple persistence test to reduce the effects of spurious SCA/cloud classifications in the input products. To be positively identified as SCA, a pixel must be snow-covered in the two most recent, cloud-free scenes of the 16-day period. We applied this time-domain-filtering (TDF) methodology to two single-day MODIS fractional snow cover products (MOD10A1 and MODSCAG) over the MODIS period of record (2000-present) and compared the outputs to the unfiltered products, to filtered maps generated using the cloud-gap-filled algorithm (CGF, Hall et al., 2010), and to historical snow assessment reports from the U.S. Army Corps of Engineers, Cold Regions Research and Engineering Laboratory (CRREL). The CRREL reports were manually generated and quality-controlled by an analyst and are treated as ground truth. We find that, when applied to MODSCAG, the TDF algorithm successfully fills in gap pixels and limits the effects of snow/cloud confusion and produces a filtered product that is more consistent and accurate than the MODSCAG CGF product and comparable to the MOD10A1 CGF product.