Every value should say where it came from, so a reader can tell a measurement from a calculation. Snotel's rule is that stored observations are observed values only, and anything else is labeled as what it is.
Four kinds of value
| Kind | Meaning | Stored as an observation? |
|---|---|---|
| Observed | A station measured it and the source reported it. | Yes. |
| Derived | Calculated from observations by a stated rule, such as a difference between two days. | No. |
| Inferred | Estimated from observations plus assumptions, such as how much new snow fell. | No. |
| Forecast | A prediction about the future. | No. |
A derived, inferred or forecast value must carry its label wherever it appears, and say what it was derived from.
Weather from the National Weather Service is kept apart from SNOTEL observations, in its own tables and with its own labels in the weather endpoint: an NWS forecast is a prediction for the station's grid cell, and an NWS observation was measured at a different station, which is always named with its distance and elevation. Neither is ever stored or shown as the SNOTEL station's own measurement.
Derived values: Snowpack Context
Snowpack Context is Open SNOTEL's derived layer, computed from the observations and stored apart from them: 24 h and 72 h changes, how today compares with a station's own history, accumulation events and rankings. Every one of its values carries category: "derived" and the derivation (an id and a version) that produced it.
- The baseline. Historical comparisons use each station's own water years 1991 to 2020 (WY1991–2020), as Open SNOTEL computed them from the station's observations. A station installed later has fewer years, and a day with fewer than ten has no comparison.
- Percent of median is not percentile. Percent of median says how today's value compares with the middle of the station's history for the day (118% is 18% above the median). The historical percentile says where today falls among those years (the 72nd percentile is above 72% of them). Neither is an official NRCS percent of normal.
- Unavailable and stale. A derived value that can't be computed is
unavailablewith a reason, such as too few years of record or a zero median, and never a zero. One older than its limit isstale, shown as old. One computed before the newest readings arrived is withheld until it is recomputed. - Versions. A new derivation version can change a derived number. The observations it was computed from never change because of it.
What travels with a value
An observed value keeps these, so it can be traced and cited:
- The station triplet, element and unit it came from.
- The flags the source attached, unchanged.
- The time it was observed, the time the source last revised it, and the time Snotel ingested it.
- A revision count, so a corrected value can be told from an uncorrected one.
Quality and freshness explain the flags and times.
Attribution
When you show data from Snotel, credit it as Data: USDA NRCS, NOAA/NWS via Open Snotel, with a link to opensnotel.com where practical, and Resort data: OpenSkiData contributors, ODbL for the resort registry. Snotel is an independent project, not affiliated with or endorsed by USDA NRCS or NOAA. The Terms set this out.
A depth change is not snowfall
It is tempting to read the change in snow depth as the snow that fell. It is not. At Banner Summit, Idaho, the station reported these daily values for three days in February 2024:
| Date | Snow depth (in) | Snow water equivalent (in) |
|---|---|---|
| 2024-02-01 | 40 | 10.7 |
| 2024-02-02 | 43 | 11.0 |
| 2024-02-03 | 45 | 11.1 |
Depth rose by 5 inches over the two days, but the water in the snowpack rose by only 0.4 inches. New snow that falls light and fluffy adds a lot of depth and little water, and settling takes depth away again, so depth alone cannot say how much snow fell. Snotel therefore stores the depths and the water equivalents as observed, and does not store a snowfall number.
If a Snotel surface shows a change in depth, it is a derived value and is labeled so. If it ever estimates snowfall, that is inferred and carries its assumptions.