How snow day predictions work
A snow day prediction is really two forecasts stacked together: a weather forecast (how much snow, ice and cold, and when) and a guess about how a school district will respond to it. Here is exactly how SnowCall does both — no "secret formula".
1. The data
- Hourly forecast from Open-Meteo: snowfall, rain, temperature, wind chill ("feels like"), wind gusts, precipitation probability and snow depth, 8 days ahead.
- Four global models — GFS (US), ECMWF (Europe), ICON (Germany) and GEM (Canada) — fetched side by side to measure agreement.
- National Weather Service gridded snowfall and ice-accumulation forecasts for US locations, blended 50/50 with the model forecast, plus active NWS alerts.
2. The closure window
When snow falls matters as much as how much. We weight each hour by how likely it is to affect the morning commute:
| Hours | Weight | Why |
|---|---|---|
| Day before, 6 AM – 6 PM | 0.5 | Crews have time to plow and treat roads overnight |
| Day before, before 6 AM | 0.2 | Mostly cleared by the next morning |
| 6 PM – 8 AM | 1.0 | Overnight and bus-run hours drive closure decisions |
| School day, 8 AM – 2 PM | 0.45 | Can trigger precautionary closures or early dismissal |
3. From weather to "effective snow"
Ice is weighted heavily because a tenth of an inch of freezing rain can make bus routes impassable. In a "regular snow" climate, 0.1″ of ice counts the same as 5″ of snow.
4. Local snow tolerance
Each location gets a tolerance tier from its state (with adjustments for snowbelts and mountains) or, for district pages, a hand-assigned tier. The threshold is the amount of effective snow at which we put closure odds at about 50%:
| Tier | Snow at ~50% | Ice at ~50% | Cold line (wind chill) |
|---|---|---|---|
| Rarely sees snow | 1″ | 0.03″ | 10°F |
| Occasional snow | 2.5″ | 0.06″ | 0°F |
| Regular snow | 5″ | 0.1″ | -15°F |
| Snow country | 7″ | 0.15″ | -25°F |
| Deep-winter climate | 9″ | 0.2″ | -35°F |
Your settings shift the threshold: rural districts ×0.85 (more cautious), urban ×1.12; districts with 4–5 snow days used ×1.10 and 6+ ×1.22 (they get stingier), none yet ×0.95.
5. Probability
The logistic curve means odds rise steeply around the threshold: half the threshold gives about 12%, one-and-a-half times gives about 88%. We cap closures at 97% because no forecast is certain. An active NWS Winter Storm, Blizzard or Ice Storm Warning nudges the odds up slightly, since official warnings make precautionary closures more likely.
6. Confidence
Confidence starts from lead time (High within about 18 hours of the morning, dropping steadily beyond two days) and is reduced when the four models and the NWS disagree about the snow-equivalent in the window. If every source agrees there is almost nothing coming, confidence is high even several days out.
How accurate are snow day calculators?
Be skeptical of any site that claims near-perfect accuracy without publishing how it was measured. Closure decisions depend on things no forecast can see: road crew staffing, a broken boiler, how the last closure was received, state rules on make-up days and whether the district will switch to remote learning instead. SnowCall has not yet published a verified accuracy record, and we will not invent one. What we do instead is show the inputs, the confidence, and the formula so you can judge the estimate yourself.
Practical tips: check the evening before; trust high-confidence estimates more; and remember that a 40% chance means a closure would not be surprising, not that one is expected.
What the model does not know
- Your district's actual closure history and policies
- Road and bridge closures, power outages or building problems
- Holidays, teacher workdays, and whether school is already out
- Whether your district will use a remote-learning day instead of a traditional snow day
Try it: run the snow day calculator, or see how much snow typically closes schools by region.
Frequently asked questions
Why does the percentage change during the day?
Forecast models update several times a day. A shift of a few miles in the rain–snow line or a band of lake-effect snow can change the expected snowfall, and the estimate follows.
Why is my probability low even though a lot of snow is forecast?
Usually because of timing (the snow ends well before the night), temperature (wet snow near freezing on warm pavement), or because your area is used to snow. Open "What's driving this estimate" to see each factor.
Does SnowCall use historical closing data for my school?
Not yet. Closure decisions are not published in a consistent national dataset, so the model uses regional tolerance levels instead. We say this openly rather than claiming school-by-school accuracy we cannot verify.
What does the confidence level mean?
It reflects how far ahead the forecast is and how much the weather models disagree about the snow and ice in the closure window. High confidence means the forecast is stable, not that the school will make a particular decision.