3.1 MODIS Snow Cover
8-day composites of the MODIS MOD10A2 version 6 snow cover product (Hall & Riggs, 2016) were downloaded from the National Snow and Ice Data Centre. This product is derived from the temporal aggregation of MOD10A1 daily snow cover in 8-day composites and it provides the maximum snow cover extent (i.e. a pixel is classified as snow if snow is present on any day of the period). Snow classification is based on the ‘snowmap’ algorithm (Crane & Anderson, 1984), using the normalized difference snow index. MOD10A2 also includes information about lakes, lake ice and cloud cover. The first two are obtained using a lake mask, while cloud cover is reported if a pixel is cloudy in each daily image. The accuracy of snow classification in the MOD10A1 product has been reported as 93% (Hall & Riggs, 2007), and no further validation was performed in this study. We acquired tiles h22v03, h23v03, h23v04, h24v04 for each year between 2000 and 2017 and days of year (DOY) 57-201, corresponding to 26th February – 20th July of a non-leap year, to enable analysis of the full snowmelt period (Table 1). Raw data tiles were reprojected from a MODIS sinusoidal grid to UTM45N WGS84, with uniform spacing of 500 m per pixel, mosaicked into a single raster layer and extracted for the area 40-54°N, 80-94°E. We processed yearly data using a cloud cover interpolation algorithm similar to previously reported in the literature (Gascoin et al., 2015; Tong et al., 2009) and calculated snow cover extents for each DOY and year (see Supplementary material).
Few data gaps were present in the original MOD10A2 and MOD10A1 datasets during 2001 (DOY 169 and 177) and 2002 (DOY 81), see also Dietz et al. (2013). Snow cover extent on these dates was calculated using linear interpolation based on the adjacent dates.