How It Works
The Science Behind the Snow Day Predictor
We believe in transparency. This page explains exactly how our snow day probability score is calculated — what data we use, how each factor is weighted, why we cap predictions at 98%, and where our algorithm performs less reliably. If you want to understand the number you see on screen, this is the place to start.
Our Data Sources
Environment Canada — Meteorological Service of Canada
Environment Canada’s Meteorological Service is the official weather authority for Canada and our primary data source. Their network of weather stations, radiosonde balloon launches, and forecast model outputs covers every populated area of Canada. We access Environment Canada’s data through the Open-Meteo API, which integrates their official feeds for Canadian locations and updates hourly during active weather events.
GDPS and GFS Weather Models
Our predictions draw on two complementary weather models. The Global Deterministic Prediction System (GDPS) is Environment Canada’s own national numerical weather prediction model — our primary model for Canadian locations. The Global Forecast System (GFS) from the US National Weather Service serves as a secondary reference. When both models agree on a forecast, our confidence is higher and our probability score reflects this. When they diverge significantly, we apply more conservative probability estimates and the uncertainty is reflected in the output.
Historical Canadian School Board Closure Records
Weather data alone cannot predict school closures — the same storm produces different outcomes in different cities and different boards. We have compiled school board closure records from publicly available sources going back five years. This historical data trains our algorithm to understand how each specific board responds to specific weather conditions: what snowfall thresholds actually triggered closures in the past, how much wind chill matters for Prairie boards vs. Ontario boards, and how each board weights storm timing in its decision.
The Probability Algorithm — Step by Step
Step 1 — Identify the Decision Window
School boards make their closure decisions during a specific window: roughly 9:00 PM tonight to 6:00 AM tomorrow morning. Our algorithm focuses on forecast conditions within this exact window. We ask: how much snow will fall between 10 PM and 7 AM? What will the wind chill be at 5 AM when transportation supervisors are driving bus routes? Is the precipitation type changing from snow to freezing rain during the critical overnight period?
Step 2 — Score Each Weather Factor
Each weather variable is scored based on its proven historical impact on Canadian school board closure decisions:
- Snowfall accumulation (0–50 points): The most important factor. 20+ cm earns the maximum 50 points. Scale: 20+ cm = 50 pts | 15 cm = 44 pts | 10 cm = 37 pts | 7 cm = 29 pts | 5 cm = 21 pts | 3 cm = 13 pts | under 1 cm = 0 pts.
- Wind chill temperature (0–20 points): Critical in Prairie provinces and Northern Ontario. -35°C or colder = 20 pts | -30°C = 17 pts | -25°C = 13 pts | -20°C = 8 pts | above -20°C = 0–4 pts based on context.
- Wind speed and blowing snow (0–15 points): High winds reduce visibility and create dangerous drifting on rural roads. 70+ km/h = 15 pts | 55 km/h = 11 pts | 40 km/h = 7 pts | under 30 km/h = 0–3 pts.
- Existing snow depth (0–10 points): A new storm on top of already-deep snow cover lowers the closure threshold. 40+ cm on the ground = 10 pts | 25 cm = 7 pts | 15 cm = 4 pts | under 5 cm = 0 pts.
- Precipitation type (0–5 bonus points): Freezing rain earns 5 bonus points regardless of other factors. Blizzard conditions earn 3–5 points. Ice pellet accumulation earns 2–3 points.
Step 3 — Apply Location Calibration
The raw point score is calibrated against the location’s historical baseline — this is the most important step that distinguishes snowpredictor.ca from generic weather tools. The same raw score of 35 points produces a different final probability in Barrie (where the SCDSB has a long history of proactive closures during lake-effect events) versus downtown Toronto (where the TDSB has significantly higher thresholds and better urban infrastructure). Our calibration data reflects actual board behaviour over five years of observed closures.
Step 4 — Calculate and Express the Probability
The calibrated score is converted to a 0–98% probability using a sigmoid function. We cap all outputs at 98% because no weather forecast tool can claim 100% certainty — there is always a non-zero probability that a board keeps schools open even under severe conditions, whether due to an unexpected improvement in conditions, political decisions, or factors outside weather entirely. The 98% cap is a commitment to honest, calibrated outputs rather than false certainty.
Accuracy Rates
Based on back-testing our algorithm against five years of Canadian school board closure records:
- Major snow events (15 cm or more): 85–90% accuracy
- Moderate events (7–15 cm): 75–82% accuracy
- Marginal events (3–7 cm): 68–75% accuracy — genuine uncertainty exists in this range
- Extreme cold events (wind chill -35°C or colder): 80–88% accuracy
- Freezing rain events: 78–85% accuracy
What the Algorithm Cannot Account For
We believe in being honest about limitations:
- Individual board decisions: A board administrator can choose to keep schools open under seemingly closure-worthy conditions, or close under seemingly manageable conditions, based on factors beyond weather data (community pressure, recent incidents, fiscal considerations).
- Highly localised microclimate events: Georgian Bay lake-effect snow bands, Niagara Escarpment enhancement, and similar hyper-local phenomena can produce dramatically more snowfall than regional models predict. Our calibration partially accounts for this in affected cities but cannot fully model it. Read more about microclimates and snow days.
- Transportation-only cancellations: In Ontario, boards sometimes cancel bus service without closing schools. Our algorithm is trained on full school closures, not transportation-only decisions.
- Rapid overnight forecast changes: A storm that intensifies or weakens significantly between 9 PM and 5 AM may produce an outdated prediction at 9 PM. Always recheck at 5 AM for the most current assessment.
Best Practices for Using the Predictor
- Check at 9 PM the night before for an initial assessment — useful for planning but less accurate than the 5 AM check
- Check at 5 AM on the morning of the storm — this is when the overnight weather observations are in the model and the board’s transportation team is already on the road
- Always verify with your school board’s official channels — our tool is a probability predictor, not a substitute for the official announcement
- For lake-effect-prone areas (Barrie, Kingston, Hamilton) — treat our probability conservatively; these events are harder to model and actual totals sometimes exceed the forecast
Frequently Asked Questions
Why is the maximum score 98% and not 100%?
Because no prediction tool is 100% certain, and showing 100% would be misleading. Even in a blizzard with 50 cm of snow forecast, there is always a non-zero probability that the board keeps schools open. The 98% cap reflects our commitment to honest, calibrated probability outputs.
How often does the weather data update?
Our weather feed refreshes every hour for all Canadian locations. During active weather events, Environment Canada pushes more frequent model updates which our system automatically captures. This means a prediction checked at 9 PM may differ from one checked at midnight as the overnight storm evolves.
Why was the predictor wrong for my storm?
The most common reasons are: the weather forecast changed significantly between when you checked and when the board decided; the storm was a highly localised lake-effect event that regional models underestimated; or the board made an unusual decision that deviated from its historical pattern. We log these cases and use them to improve calibration each season.
Want to learn more? Check out our complete methodology here.