Wind and Rain Are the Only Weather That Matters for NFL Totals
Every winter, the same ritual: a storm system rolls toward a stadium, the broadcast shows a shaky flag on the uprights, and everyone assumes the game will stay under the total. Snow game? Under. Freezing cold? Under. A little drizzle? Under.
Most of that intuition is wrong. We spent months testing every weather input against 23 seasons of NFL totals results — as part of building our prediction models — and the conclusion is blunt: two weather conditions genuinely move totals outcomes, and everything else fans react to is noise.
The two that matter
Wind at 15+ mph: ~58% of games finish under
Across every outdoor NFL game since 2003 with sustained wind of 15 mph or more — 553 games, roughly 24 per season — 57.7% finished under the total. Against a ~50% norm, that's nearly an 8-point deviation, sustained across two decades.
What makes this trend worth trusting where most fail:
- It's era-stable. Split the sample into 2003–2021 and 2022–present and the under rate barely moves: 57.7% vs 58.2%. Real effects replicate; data-mined ones collapse out of sample.
- It's monotonic. Calm games run ~49% under (dead on the norm). Moderate wind (10–14 mph) nudges to ~54%. At 15+ it jumps to ~58%. The effect scales with the physical cause — exactly what you'd expect if wind, not luck, is doing the work.
- It's season-robust. The under rate finished above 50% in 17 of 23 seasons, and the strongest years are scattered across the timeline, not clustered in some ancient low-scoring era.
- The mechanism is obvious. Wind degrades the passing and kicking game — the two most efficient ways to score — and no amount of expectation-setting changes physics.
Forecast rain: ~59–60% of games finish under
Rain is the second real signal, and in some ways the stronger one. Using the weather-service conditions recorded for every outdoor game (about 14 rain games per season):
- Games with rain in the forecast finished under 59–60% of the time, and it's era-stable too: 59.3% in 2003–2021, 61.0% since 2022.
- When rain actually falls, it's ~70% under. The gap between 60% and 70% is forecast false alarms — games where rain was called and never materialized played like normal games. Seventy percent is the physical ceiling; sixty is the number you can actually know before kickoff.
- The dry baseline is 50.4% across 4,231 games. That control matters: absent weather, totals outcomes match expectations to the decimal. Weather is the specific exception, not one of many.
And they stack
In rain games with calm wind, ~55% finished under. In rain games with 10+ mph wind, ~65% finished under. Each condition contributes roughly independently — wet ball plus hard wind is the strongest scoring suppressor the data can find.
The weather that doesn't matter
This is the half of the article that corrects the folklore.
- Snow: statistical noise. The most iconic "low-scoring weather" in football shows no reliable totals effect in our data — the sample is small and the direction flips between eras. The blizzard-game under is an aesthetic, not a pattern.
- Cold: nothing measurable. Temperature graded out as one of the weakest inputs we tested across 360 candidate model features. Scoring expectations for cold-weather games are already well calibrated.
- Short rest / Thursday games: at the norm (~50.5% under). Tired-legs-equals-low-scoring is folklore.
- Pace, dome vs. outdoor by itself, totals mean-reversion: all tested, all at expectation. Totals absorb everything scheme- and situation-shaped. What they systematically miss is physics on game day.
The finding that surprised us most
Here's the result that made us trust this analysis: we couldn't use wind to improve our prediction model — even though it's the single most correlated weather feature we have.
When we added wind speed as an input to our machine-learning totals model, accuracy went down. The reason is instructive. In roughly 88% of games, wind is irrelevant — feeding it to a model just adds noise across the whole schedule. The value only exists in the ~12% of games past the threshold. Wind is a filter, not a factor: "expect a lower-scoring game when wind is 15+" works precisely because it ignores every game where wind doesn't matter.
That distinction — between a real conditional pattern and a model-worthy signal — is the kind of thing you only learn by testing honestly. It's also why we're comfortable publishing this: totals expectations are remarkably accurate, these two narrow exceptions are the entire weather story, and any analysis built on snow, cold, or "weather games" broadly is built on the 88% that doesn't matter.
The honest caveats
- Low volume. ~24 qualifying wind games and ~14 forecast-rain games per season. This is a narrow situational pattern, not an every-week signal.
- Off years happen. The wind-game under rate finished at 37% in 2010, 36% in 2020, and 45% in 2024. A 58% tendency still runs cold on 20-game samples — that's what a real but modest effect looks like.
- Forecasts move. The meaningful read is the forecast in the final day before kickoff, not the Tuesday guess.
Check the conditions yourself
Game-time weather — including wind speed — is tracked for every game on the SpreadTrends Schedule and Matchups pages, with forecasts updated continuously until kickoff and frozen at game time. When a 15-mph flag shows up on a Sunday slate, you'll see it there first.
Data: outdoor NFL games, 2003–2025 seasons. Wind figures use cleaned, cross-verified wind readings (we found and fixed a unit-conversion bug in one data source by reconciling against a second, independent feed — the corrected data made the signal stronger and cleaner, not weaker). Games landing exactly on the total are excluded from percentages.