In 1997, the National Ice Center (NIC) introduced the Interactive Multisensor Snow and Ice Mapping System (IMS). Inputs for IMS evolved to a more diverse set of products including satellite imagery, snow and ice analysis maps, National Centers for Environmental Prediction (NCEP) model data, and surface observations. Among these inputs, time sequenced satellite imagery improved the discrimination between snow and clouds following the assumption that any short-term changes in reflectance can only be introduced by clouds (Lyapustin et al., 2008; Gafurov and Bárdossy, 2009). Since its inception, the IMS has served as a more effective and modernized approach to snow mapping compared to the historical approach. Using the IMS tool, snow extent output has been produced at spatial and temporal resolutions corresponding to a 24 km daily product (Ramsay, 1998; Helfrich et al., 2007).

Both the historical weekly SCE maps and the higher-resolution IMS products were independently produced during a 2-year overlap period from June 1997 to May 1999. To generate a pseudo-weekly product using daily IMS SCE (to ensure continuity in evaluating the full satellite-era snow extent record), it was necessary to compare the two independently produced versions generated during the overlap period. SCE area calculations performed with these data revealed that Monday IMS output best matched the corresponding weekly product output. A reduction in the 24 km IMS maps to 190.6 km resolution was also performed using a threshold to determine the location of SCE. To designate a weekly grid cell as snow covered, ≥ 42 % of the IMS land pixels falling within the cell must indicate snow. This threshold provided the best match between weekly SCE areas and Monday IMS SCE areas during the overlap period. While a longer overlap period may have improved the analysis, the 2-year sample provided by NOAA was invaluable to preserving continuity in the CDR time series.

Starting in June 1999 the CDR is completely derived in this manner, reducing the spatial resolution of IMS output to produce weekly granules conforming to the historical weekly SCE product. Although the IMS data used in this process are dated Monday, satellite imagery from the 36 h leading up to the analysis time can be used to more accurately delineate SCE boundaries (Robinson et al., 1999).

The transition from weekly to daily maps has not resulted in an artificial step change in spring continental-scale SCE (Brown and Robinson, 2011; Robinson et al., 1999). More research is needed to determine whether SCE analysis in mountainous regions (e.g., the Tibetan Plateau) shows systematic change during this time period. While there are differences in the production of the historical data compared with IMS, the conversion of IMS snow output to weekly SCE maps is automated and reproducible.

To date, the weekly SCE data product has been widely used by the cryospheric community. The CDR is an improved version of the data series, with efforts to achieve CDR status described in the following sections.

The weekly data are available in Network Common Data Form (netCDF), which is self-describing, platform independent, and archivable. All weekly CDR granules are stored in a single netCDF-4 file. Metadata elements comply with climate and forecast (CF-1.6) conventions, and collection-level metadata adheres to ISO 19115-2 standards for geographic information to facilitate data set discovery. These netCDF file metadata are structured following guidelines recommended by the NCDC satellite Climate Data Record Program (CDRP, 2011b).

The latest CDR product includes refinements to the NH Cartesian grid previously distributed by NOAA's Satellite and Information Service. Cell center geographic coordinates were evaluated against a regular grid in polar stereographic coordinates (PSC) with a cell size of 190.6 × 190.6 km. The total calculated area of the two grids agreed within -0.74 % indicating that the grid provided by NOAA (B. Ramsay, personal communication, 1998) and released in an earlier version of the CDR slightly underestimates total area. Except for three grid cells located over the pole, the longitude and latitude coordinates were found to be accurate to within ±26 km, and cell areas were accurate to within ±459 km2 (J. Biard, personal communication, 2012). All geographic coordinates and cell areas in the latest version of the CDR product have been corrected to correspond to the regular grid in PSC. These refinements do not impact the presence or absence of snow at these grid locations since the determination of snow is made in row and column space, not by latitude and longitude coordinates.

Cell areas range from ∼ 10 700 km2 near the Equator to ∼ 41 800 km2 near the pole. To generate the CDR, values from an 88 × 88 subset of the historical weekly 89 × 89 SCE matrices are used to populate the SCE variable in the netCDF-4 binary file. This allows the CDR product to provide data only for grid cells that lie entirely north of the Equator.

Mean NH snow cover extent reaches its maximum in January and minimum in August, ramping up quickly in the fall and melting at a slower pace in the spring. The mean annual CDR SCE is 25.1 million km2, with a standard deviation (SD) of 0.9 million km2. Mean annual maximum SCE totals 47.4 million km2 (SD = 1.5), which is 44.4 million km2 more extensive than the mean annual minimum of 3.0 million km2 (SD = 0.7). The highest annual maximum SCE occurred in February 1978, totaling 51.3 million km2. At the opposite end of the rankings, the lowest annual minimum SCE of 2.1 million km2 was observed in August 1968 (Table 2).

The CDR shows 1978 as the snowiest year (i.e., September 1977 to August 1978), with an annual mean SCE of 27.2 million km2. In contrast, annual mean SCE in 1989 was the lowest, totaling 23.5 million km2. During the most recent 27 years of the CDR, mean annual SCE has continued to exhibit lower snow extents relative to the data period ending in mid-1987. This step change in NH SCE was first identified by Robinson and Dewey (1990).

Seasonal trends observed in the CDR include a pronounced decline in SCE during the spring melt season (Fig. 3). Fall and winter show an opposite trend towards increased extent, with winter SCE growing 0.19 million km2 per decade. Fall seasonal SCE means have increased by 0.26 km2 per decade. By contrast, spring SCE has decreased by -0.58 million km2 per decade, declining at 3 times the pace of the winter trend. This has also left the summer season with less snow on the ground, with summer SCE decreasing by -0.81 million km2 per decade.

Brown and Derksen (2013) found that over Eurasia the positive trend in fall SCE is internal to the CDR and is likely a result of improved snow detection during the October snow onset period. SCE observations from other sources compared in the study exhibit a reduction in October SCE from 1982 to 2011, which is consistent with warming fall surface temperatures observed during this period.

Conversely, the negative trend in spring SCE shown by the CDR has been corroborated using observations from multiple independent data sources (Brown et al., 2010; Brown and Robinson, 2011). While the data sets show different areas of Arctic SCE, all data sources including the CDR exhibit similar SCE anomalies, which indicate significant reductions in Arctic SCE over the period 1967–2008.

Brown and Robinson (2011) derived estimates of uncertainty through the statistical analysis and comparison of multiple SCE data sources. The results indicate that the 95 % confidence interval in March and April continental SCE is ±3–5 % during the satellite era beginning in 1967. The analysis also shows larger uncertainties in spring SCE monitoring over Eurasia compared to North America, in part due to larger variability between data sources over northern Europe and north-central Russia.

The CDR excels at portraying regional SCE in areas with high illumination, stable snow cover, and frequently clear skies. Despite the CDR's general homogeneity, difficulties in historical SCE charting over the Tibetan Plateau due to frequent cloud cover resembling patchy snow on the surface have resulted in higher uncertainty over this region. Positive biases have been shown in SCE mapping during summer months in the Canadian Arctic; this also relates to frequent cloud cover over the area (Wang et al., 2005; Brown et al., 2007).

The NOAA climate data record of Northern Hemisphere snow cover extent, Version 1, is archived and distributed by NCDC's satellite Climate Data Record Program. The CDR can be downloaded via FTP or from the NCDC THREDDS data server (10.7289/V5N014G9). The CDR is forward processed operationally every month, along with figures and tables made available at the website http://climate.rutgers.edu/snowcover/.