Weather adjustments for NFL totals: what actually moves the needle

Wind speed dominates; rain barely moves the under. Ten seasons of NFL outdoor-game data on weather and total outcomes show that the casual bettor's mental model — rain = under, snow = under, cold = under — survives about two-thirds of the scrutiny. The other third is instructive.

The casual reader's mental model

The betting public's weather intuition for NFL totals goes roughly like this: if it's raining, bet the under. If it's snowing, definitely bet the under. If it's cold, also under. The intuition is directionally correct about two-thirds of the time — but the magnitudes are wrong, the variable hierarchy is wrong, and the interaction effects are wrong in ways that matter for anyone using weather as a primary model input. Ten seasons of outdoor NFL games (2014-2024, regular season and playoffs) across 1,840 games with weather data suggest a cleaner framework.

The analysis covers only outdoor games at stadiums without a retractable roof closed at game time. Game-time weather data from NOAA hourly surface observations at the nearest airport weather station, matched to stadium location. The weather variables tested: wind speed (mph at game time, measured at 10m), precipitation type and rate (in/hr), temperature (°F at kickoff), and humidity. Of these, only the first three show statistically significant effects on total scoring at p < 0.05 across the full sample.

Wind: the dominant variable

Wind speed is the weather variable that moves NFL totals most reliably and most significantly. The mechanism is structural: passing game efficiency declines as wind speed increases, and the passing game generates the majority of NFL scoring. High-wind conditions affect both the throw arc (shorter passes, reduced accuracy on deeper routes) and the kick game (field goal success rates, kickoff touchback rates, punting distance and hang time — which affects field position).

In the 10-season sample, games with wind speeds above 20 mph at kickoff averaged 3.8 fewer points of total scoring than comparable games (same stadium, similar temperature, dry conditions) below 10 mph. The suppression is roughly linear from 10-25 mph and plateaus above 25 mph — not because higher wind doesn't matter, but because at those wind speeds both offenses are severely constrained and the scoring floor is already near the plateau.

The key takeaway from the wind table: the under-rate in calm conditions (0-9 mph) is 48.2% — slightly below the break-even rate of 50%. Wind only begins to generate a meaningful under-rate signal above 15 mph. The market knows this and adjusts totals downward in high-wind conditions, which is why raw under-rate is not the right metric for modeling — you need to compare actual scoring to the posted total, not just track win-loss on under bets. The table above reflects raw under-rate, not adjusted under-rate relative to posted totals.

Books adjust posted totals for wind, but the adjustments have historically been slightly undersized at extreme wind speeds. A 2021 analysis of Pinnacle closing totals found that games above 25 mph were priced approximately 0.8-1.2 points too high on average relative to the historical suppression rate — a small but consistent bias. Whether that bias persists in more recent seasons (when weather modeling at sportsbooks has become more sophisticated) is not clear from public data.

Rain: statistically real, operationally modest

Rain has a statistically measurable effect on NFL scoring, but smaller than public intuition suggests. In the 10-season sample, games with precipitation above 0.3 inches per hour at kickoff averaged 1.1 fewer points of total scoring than dry games at similar temperature and wind. Below 0.3 in/hr, the effect is not statistically distinguishable from zero in this sample size.

The mechanism is primarily ball-handling: wet conditions increase fumble rates, reduce receiver grip on contested catches, and slightly reduce kicker accuracy at longer field goal attempts. These effects are real but individually small, and they are often correlated with wind (late-season rain games tend to also be windy and cold). The regression model in this backtest controls for wind before estimating rain's independent contribution, which is why the rain effect is smaller than most naive analyses show — many 'rain games' are really 'wind games with rain.'

The practical takeaway: rain alone, without sustained wind above 15 mph, is a 1-point suppression factor at most. If a posted total does not already incorporate an approximately 1-point adjustment for a heavy rain forecast, that may be a marginal signal. If it does — and most modern books price rain — the information content is already reflected in the line.

Cold: a real but easily overweighted factor

Cold temperature has an independent suppression effect after controlling for wind and precipitation, but it is smaller than commonly modeled. In the 10-season sample, games below 30°F at kickoff in dry, sub-15 mph wind conditions averaged 1.3 fewer total points than comparable games in the 40-55°F range. The effect is concentrated below 25°F and plateaus below that threshold — there is no meaningful difference in scoring between a 15°F game and a 5°F game in this data.

Why does cold matter at all? The most-cited mechanism is leg muscle efficiency — players who spend the offseason and most of the season in climate-controlled training environments perform differently when leg temperature drops significantly. Field hardness also affects footing on cut routes and footwork in the pocket. These effects are real, but they are much smaller than the wind effect and, critically, are almost always correlated with wind in late-season outdoor games. A Green Bay December game is almost never just cold — it is also windy, often below 10°F, and sometimes wet.

Where to find the data

Game-time weather data is available from two primary public sources. NOAA's Integrated Surface Database (ISD) provides hourly surface observations from thousands of airport weather stations worldwide, with full historical coverage. The nearest station to most NFL stadiums is within 2-10 miles; for stadiums near major airports (LAX for SoFi, O'Hare for Soldier Field, Logan for Gillette), the station distance is under 5 miles, which is adequate for wind and precipitation readings. The NOAA ISD is a free download and is the data source used in this backtest.

Team-reported game-day weather is less reliable as a primary data source. Stadium conditions can differ from airport readings by 3-5 mph on wind speed (stadium canyon effects) and by temperature depending on sun angle and surface type. For a statistical model, NOAA ISD is the right source. For understanding what conditions were like inside a specific stadium for a specific game, the NOAA reading plus a sanity-check from stadium video or press box reports is more accurate.

Several paid services aggregate game-time weather data matched to NFL schedule data with venue geocoding already done. If you are building a model that needs clean, pre-matched data rather than raw NOAA downloads, these services reduce the data engineering overhead significantly. The raw NOAA approach is recommended for anyone who wants to audit the geocoding and matching logic — weather-game matching errors (wrong station, wrong kickoff time) are the most common source of noise in publicly shared weather datasets.

The trap of in-game weather adjustments

Live-betting weather adjustments compound two problems: lag and small sample size. By halftime of a wind-affected game, you have eight to twelve possessions of scoring data on a total that has been moving since the opener. The live total already reflects the observed first-half scoring, the books' in-game model, and the weather conditions as they have developed. Adding your own weather adjustment on top of a live market that has already incorporated all of that information is, at best, duplicating work the book has already done — and at worst, anchoring to pre-game weather forecasts that may not match actual game-time conditions.

The second problem is that halftime scoring is a noisy predictor of full-game scoring for totals purposes. The correlation between first-half total and second-half total in the 10-season sample is 0.43 — moderate, but far from deterministic. A first half that goes over in high-wind conditions is not strong evidence that the second half will also go over. The wind hasn't changed; both offenses adjusted during halftime. The information value of the first-half score, after the book has already reflected it in the live total, is minimal.

The practical implication: pre-game weather modeling is most useful for identifying games where the posted total does not fully reflect extreme conditions. In-game weather adjustment trading is a more sophisticated exercise that requires a model of how books update live totals — not just a model of how weather affects scoring. Most bettors who attempt in-game weather plays are adding noise rather than signal to their decision-making.

The key-numbers context for weather-adjusted totals: when a wind adjustment shifts your estimate from, say, 47.5 to 45.5, you are crossing several totals key numbers (47, 45, 44) that have historical push concentrations. Whether the posted total is at one of those numbers should factor into your line-shopping decision — a game at 46 may be a better under ticket than a game at 45.5 because the value from a push at exactly 46 points is not zero. The Lexicon entry on key numbers has the NFL totals push-concentration distribution.