How Our Snow Day Predictions Are Built
Most snow day tools ask you to trust a number without ever explaining where it comes from. This page exists so you don't have to. It covers exactly what data feeds every estimate, how different weather factors are weighted, why your specific region matters, and where the model's real limitations are
Where the data comes from
Every estimate draws on publicly available forecast data from national weather authorities, including NOAA's National Weather Service in the United States, and Environment and Climate Change Canada north of the border. These are official public weather sources rather than forecasts created independently by SnowDayScore, not a private or proprietary weather feed built in isolation.
The six factors we evaluate
Every location is scored against six weather inputs: expected snowfall, ice and freezing rain accumulation, minimum temperature, wind chill, wind speed, and the timing of the storm relative to the school day. None of these are weighted equally, and none of them are evaluated alone
Why temperature, wind speed, and wind chill aren't double-counted
Wind chill is calculated from temperature and wind speed, so those two raw inputs are not separately re-scored on top of the wind chill figure. Temperature and wind speed are used to calculate wind chill accurately and to catch edge cases wind chill alone can miss, such as a very cold, still morning that doesn't cross a wind chill danger threshold but still affects road conditions. The model treats these three inputs as one connected calculation, not three independent scores added together.
Why ice generally carries more weight than snowfall
Ice and freezing rain generally increase closure risk more sharply than an equivalent depth of dry snow, since a thin layer of ice can be more difficult for road crews to treatquickly and can remain hazardous after it stops falling. This is consistent with general winter weather safety guidance from NOAA's National Weather Service, and it's also why our ice-versus-snow page documents specific, sourced examples (including a two-day school closure in Portland, Oregon triggered by roughly a quarter to half inch of ice) rather than asserting a blanket national rule
Why storm timing changes the estimate
Identical snowfall arriving at 3 a.m. and arriving at 3 p.m. produce different estimates, because road crews have vastly different amounts of time to treat routes before the first bell. Timing is treated as a multiplier on the other factors, not a separate, disconnected input.
Regional and school-type adjustment
The same raw weather inputs produce different estimates in different places, because closure decisions are shaped by local road infrastructure, plow fleet size, and district risk tolerance as much as by the forecast itself. To be specific about what this currently means: for the handful of districts where we have confirmed, named closure policies (for example Anoka-Hennepin and Duluth Public Schools in Minnesota), the model uses that real district-level data directly. Everywhere else, the model relies on broader regional patterns, for example treating a low-snowfall region as generally more sensitive to a given snowfall total than a heavy-snowbelt region, rather than a specific district's own confirmed threshold. We are direct about this distinction rather than implying every region has the same level of granular, named data behind it. School type is adjusted for separately: where supported by confirmed institutional policies or historical outcomes, colleges and universities may be treated differently from K-12 districts.
What the estimate is, and what it isn't
Every result is labeled as a Closure Risk Score, not a statistical percentage or probability, and not a guarantee. That distinction matters: a true statistical probability requires validation against a large history of real outcomes. The score will not be described as a calibrated probability unless it has been tested against a sufficiently large and geographically diverse set of verified outcomes, not simply after one season has passed. Results are shown as a tier plus a range, or a score out of 100, deliberately avoiding a single bare percentage that could look more precisely calibrated than the underlying model currently supports.
How we're tracking accuracy
Once live, this model will log every prediction against the outcome the relevant school district actually reports, and that record will be compiled and published on an ongoing basis, including the weeks the model gets it wrong, rather than only a general claim of being accurate. We believe a dated, public track record isa stronger basis for trust than a claim of accuracy on its own.
How the model changes over time
As more real closure outcomes are logged, factor weightings are reviewed and adjusted where the data supports a change. Any substantive update to the model is dated and noted here, rather than silently changed without a record.
Limitations, stated plainly
This model cannot see everything a superintendent weighs, including last-minute road conditions, bus driver availability, or localized events like a power outage or water main break. Estimates in the moderate range reflect genuine uncertainty, not a flaw in the tool, and often mean your district is weighing the same close call the model is. Estimates made more than a day or two in advance should be treated as anearly signal, not a firm answer, since storm tracks and totals can still shift significantly. Nothing on this site is an official closure notice. Always confirm with your school district directly.
Is this a real statistical probability?
No. It is labeled as a Closure Risk Score because it has not yet been validated against a sufficiently large and geographically diverse set of verified outcomes. See the accuracy tracking section above for how that will be measured going forward.
What weather factor mattersmost?
No single factor dominates. Ice and extreme wind chill are weighted more heavily than an equivalent snowfall depth, and storm timing acts as a multiplier across all factors, since the same snowfall arriving overnight is treated differently than the same snowfall arriving midday.
Does the model account for my specific school district?
It adjusts by region, and school type is considered where confirmed institutional policies or sufficient historical outcomes are available, rather than applying one national curve. As more named district data is confirmed, results become more specific to individual districts over time.
How often is the model updated?
The model is reviewed as new closure outcomes are logged. Any substantive change is dated and noted on this page, rather than changed without a visible record.
Related reading: Learn more about the team behind SnowDay Score, or try the manual snow day calculator to test your own scenario against this methodology. Back to the SnowDay Score predictor.
Accuracy Tracker
We are not going to put a made-up accuracy percentage on this page just to have a number here. Real accuracy tracking requires logging predictions and checking them against what school districts actually decide, over enough time and enough regions for the result to mean something. Here is where that stands right now, updated honestly rather than left to imply more than we can back up:
Once tracking begins, this section will update with the number of predictions made, how many have been checked against real school closure decisions, and a reported accuracy rate broken down by region and season. Results will be published as they are, including the weeks the model gets it wrong, rather than only a flattering average. Check back as the upcoming winter season progresses.