2. DESCRIPTIONS, DATA, AND EXAMPLES OF THE NEW SNOW METRICS
The new snow metrics described below are meant to improve upon and augment established snow metrics. The snow metrics described in the following section are intended to add value through extended spatial coverage, improved spatial resolution, more nuanced interpretation, improved data access, and user-driven flexibility.
2.1 Snow Cover Frequency (SCF; Global)
Snow Cover Frequency (SCF) is a global, satellite-derived, gridded product representing the observed frequency of snow cover in a grid cell. It is computed as the number of days that snow is observed on the land surface divided by the number of valid observations for a specified period (Eq. 1). We use the global, 500-m, daily, gridded snow cover product, version 6 from NASA’s Moderate Resolution SpectroRadiometer (MODIS/Terra MOD10A1). Because the MOD10A1 product is daily and global, it allows the user to compute SCF anywhere globally for the MODIS period of record (February 2000–present). It is statistically valid for any period of 30 days or longer. In related work, Crumley et al. (2020) have created a version of this SCF metric for the Northern Hemisphere Water Year (WY, October 1–September 30) for WY 2001–2019. Their SCF snow metric is available as a web-based product (SnowCloudMetrics.app) developed using the Google Earth Engine (GEE) framework. In its current form, the app approach has somewhat limited flexibility for user-defined temporal and spatial subsetting. Even greater flexibility is gained if a user is willing to run the SCF code within GEE (i.e., as a GEE developer). In this case, users can select and compute SCF for any sub-annual set of sequential days within the MOD10A1 record. Users can also spatially subset SCF by user-defined polygon, US state boundary or Canadian province, elevation range, or USGS watershed. For instance, a user can draw a polygon around the California Sierra Nevada range and compute a regional SCF. The code used to produce the SCF metric, as described here, is available to users who wish to compute this product on their own using GEE.
Here, we present both a global and a regional example of SCF using MODIS data. Figure 1a shows global SCF for WY 2015. Over the western US, one can see significant differences in SCF between the two years; 2015 was a record low snow year for the western United States, and 2017 snowpack was anomalously high over the region. Figures 1b and 1c take a closer look, focusing on SCF over the California Sierra Nevada for WY 2015 and WY 2017.
In another GEE application of the SCF metric, Sproles et al. (2018) found that SCF was valuable for monthly streamflow forecasting. They used SCF in their web-based, cloud-computing tool SnowCloudHydro, which combines basin-scale SCF with a simple hydrologic model for use in snow-dominated and data-sparse watersheds. Crumley et al. (2020) noted a strong association between measured SNOTEL-derived SCF and MODIS-derived SCF (using SnowCloudMetrics). We are not claiming that SCF is analogous to SWE though this may be the case in much of the western US, especially for elevations where snowpacks are very temperature sensitive. Indeed, in cold regions such as the Arctic boreal region, any relationship between SCF and SWE might be weak because the cold, thin snow cover can persist for many months but have a low SWE. However, we note that the SCF metric is valuable in these regions because it can detect interannual variability in the snow line elevation in the spring months (Verbyla et al. , 2017). This is important for wildlife such as Dall sheep who produce lambs at that time and for whom late-season snow cover reduces forage access, and for caribou who migrate long distances (Verbyla et al. , 2017; Mahoney et al. , 2018; Van De Kerket al. , 2018; Boelman et al. , 2019).

2.2 Snow Disappearance Date (SDD, Northern Hemisphere extent)

The Snow Disappearance Date is a global, satellite-derived, gridded product that maps the last day of the WY when snow is last detected in a pixel. As with SCF, SDD also uses the global, 500-m, daily, gridded snow cover product from MODIS. Starting on the last day of each WY, the algorithm searches back in the WY for the longest period without snow after a minimum of 5 days of snow cover (accounting for cloudy days). Crumley et al. (2020) computed SDD for WY 2001–2019. Like SCF, SDD has been used in wildlife studies for locations where spring snow disappearance affects the survival of young (Van De Kerk et al. , 2018). To date, SDD has not been as widely used as SCF. Still, we anticipate its value for areas where snow disappearance date has been related to the onset of the wildfire season (Westerling et al. , 2006b) and where spring vegetation phenology varies with snow cover (Huang et al. , 2018; Xie et al. , 2020).
With our revised metric, a user can compute SDD globally in about 20 seconds or use the polygon drawing tool to select a region such as the California Sierra Nevada. Figure 2a shows global SDD for WY2015 and SDD subset by polygon for the Sierra Nevada (Figures 2b and 2c), providing the ability to compare a low snow year (WY 2015) with a high snow year (WY2017).
At this time, SDD spatial coverage is for the Northern Hemisphere only. In addition to SDD through the SnowCloudMetrics.app, users can also download the SDD code and run it independently using Google Earth Engine. SDD is still computed only for the Northern Hemisphere WY, but users have greater spatial subsetting options, similar to those for SCF, as described above.
[Insert Figure 2]

2.3 | Snowstorm Temperature (ST, western US extent)

Snowstorm temperature is the mean daily temperature on days when snow accumulation occurred (Hu and Nolin, 2019, 2020). This is our only snow metric that is derived from SNOTEL station data. A storm day is determined based on measured changes in SWE. Spatial coverage is for the western United States, excluding Alaska. The snowstorm temperature data set is composed of 33 years of daily meteorological data (1984–2016) from 567 SNOTEL sites and a homogenized daily temperature dataset (TopoWx). The SNOTEL data were used to determine dates on which a snowstorm occurred and whether the snow water equivalent increased, decreased, or stayed the same. Storm day is defined as one with a measurable positive change in precipitation. SWE-gaining and SWE-losing days were recorded. TopoWx temperature data (Oyler et al. , 2015) were used to produce the dry-day and storm-day data for each of the 573 SNOTEL sites. Disaggregating winter storm days from non-storm (dry) days turns out to have a major impact on the interpretation of temperature trends. For instance, Hu and Nolin (2020) found that storm days are warming at roughly twice the rate of dry days. They also found that there were significantly fewer storm days in November. Such a trend indicates not only a delay in the start of the wet season but with declining spring snowpacks (Mote et al. , 2005, 2018), means a lengthening of the preceding dry season.

2.4 “At-Risk” Snow (ARS, conterminous US extent)

At-Risk Snow is computed in a manner similar to that in Nolin and Daly (2006) except that we use climate model output in this updated version of the ARS metric. ARS is computed using the gridded NASA Earth Exchange Downscaled Climate Projections (NEX-DCP30) 30 arcsec (approximately 800-m) dataset (Thrasher et al. , 2013), which covers the conterminous United States. We compute mean temperature from their average monthly maximum and minimum temperature data from 33 downscaled climate models and four Representative Concentration Pathways (RCPs) (Meinshausen et al. , 2011). The data set includes retrospective model runs covering the historical period from 1950 to 2005, and prospective model runs for 2006 to 2099. The spatial domain is the conterminous United States (Figure 3). The ARS gridded map product depicts model grid cells where December–February monthly mean temperatures will reach the 0°C threshold.
We have produced a fully functional code running on GEE that allows the user to compute At-Risk Snow for any range of years in the period covered by the NASA NEX-DCP30 downscaled climate data. Users can spatially and temporally subset, visualize, explore, and download the FWW data of interest. A user can filter by emissions scenario, global climate model, spatial extent, and period. A user can select from any of four CMIP5 emissions scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) and from numerous downscaled (800-m) model output (23 models for RCP2.6; 33 models for RCP4.5; 17 models for RCP6.0; and 31 models for RCP8.5). If desired, a user can subset by watershed using the USGS hydrologic unit codes (HUC levels 2-12).
[Insert Figure 3]

2.5 Frequency of a Warm Winter (FWW, conterminous US extent)

As with the At-Risk Snow metric, FWW uses the monthly maximum and minimum temperature data from NEX-DCP30, computing mean temperature from their average. The definition of FWW is the same as in Nolin and Daly (2006), but the geographic scope extends to cover the conterminous US at 800-m spatial resolution. The temporal range is 1950–2005 for historic FWW and 2006–2099 for future FWW.
For users to compute FWW, we have produced code that runs on GEE. This allows the user to calculate FWW for any range of years in the period covered by the NASA NEX-DCP30 downscaled climate data. As with ARS, users can spatially and temporally subset, visualize, explore, and download the FWW data of interest. A user can filter by emissions scenario, global climate model, spatial extent, and range of years. If desired, users can subset by watershed using the USGS hydrologic unit codes (HUC levels 2-12). When computing FWW, we recommend that users specify a range of at least 30 years so that the calculated values are the empirical equivalent of a statistical probability. For instance, a user can compare FWW for 1979-2009 (historical) with FWW for 2035-2069 (mid-century) or 2079-2099 (late-century). The 800-m gridded data are sufficiently fine spatial resolution that users can explore projected FWW changes for spatial extent as large as the conterminous United States and as small as a headwater catchment, an urban area, and even along an elevation gradient in a ski area.
Figure 4 shows an example of FWW for the conterminous United States and selected ski areas, for three time periods: historic (1970-1999) and RCP8.5 mid-century (2035-2064), and RCP8.5 late-century (2070-2099).
In a recent application of FWW, Nolin assisted the city of Whitefish, Montana with their Climate Action Plan (City of Whitefish, 2018). The winter economy of Whitefish depends on the success of the nearby ski area, Whitefish Mountain Resort. For this report, Non subset and computed FWW for the lower, mid- and summit elevations of the ski area for different climate scenarios and for mid-century and late-century. This is a straightforward example of the use of the FWW metric for non-scientists seeking climate change information at the local-to-regional scale. Table 1 gives FWW values for the selected ski areas shown in Figure 4. These represent major ski resorts in a maritime snow climate (Squaw Valley, CA; Mt. Bachelor, OR), northern Rocky Mountains (Big Sky, MT), central Rocky Mountains (Vail, CO), and northern Appalachian Mountains (Killington, VT).
[Insert Figure 4]
[Insert Table 1]