Yes. Adding six-hour stadium weather forecasts reduced held-out weekly PPR projection error by 0.568% overall (95% game-cluster bootstrap interval: +0.328% to +0.810%), winning all six future-season windows from 2019 through 2024.
We tested whether cutoff-safe six-hour stadium weather forecasts improve weekly fantasy football projections. On 20,700 held-out NFL player-weeks from the 2019 through 2024 seasons, the weather-augmented model reduced weekly PPR mean absolute error by 0.568% versus the otherwise identical champion — a small but consistent edge that showed up in every season tested.
The headline number
The weather challenger posted a weekly PPR MAE of 4.635 points against the champion's 4.662 on the same 20,700 player-weeks. That is a 0.568% improvement, with a 95% game-cluster bootstrap interval from +0.328% to +0.810% — the interval excludes zero, so the improvement is not noise. The bootstrap resampled NFL games 4,000 times, keeping all player-weeks from a sampled game together, which respects the fact that everyone in one game shares the same weather.
| Model | Player-weeks | MAE | 80% interval coverage |
|---|---|---|---|
| Current champion | 20700 | 4.662 | 79.2% |
| Weather challenger | 20700 | 4.635 | 79.0% |
Consistent across every season
The challenger won all six locked future-season windows. The largest gap came in 2019, when weather cut MAE by 0.067 points (4.777 vs 4.844); the closest was 2024, a near-dead heat at 4.573 vs 4.575. Each season was predicted only from earlier seasons, and every weather forecast was archived before its six-hour kickoff cutoff, so no future information leaked into the predictions.
Where the gains came from
Quarterbacks and wide receivers drove the improvement. QB MAE fell 1.12% (6.526 to 6.454) and WR MAE fell 1.09% (4.639 to 4.589). Tight ends were essentially flat at -0.06%. Running backs were the one position that regressed, and only by 0.24% — the largest position-level regression in the study, small enough that it does not offset the overall gain.
| Position | Player-weeks | Champion MAE | Weather MAE | Change |
|---|---|---|---|---|
| QB | 2206 | 6.526 | 6.454 | -1.12% |
| RB | 5413 | 4.795 | 4.807 | +0.24% |
| WR | 8716 | 4.639 | 4.589 | -1.09% |
| TE | 4365 | 3.600 | 3.597 | -0.06% |
What this means for your lineup
Weather is now part of the active weekly projection champion, because the challenger passed every locked model gate — calibration, leakage, missingness, reproducibility, cost, subgroup stability, and the primary metric. For fantasy managers, the practical takeaway is modest: weather adds a measurable, repeatable layer of accuracy to weekly projections, especially at the pass-catching positions, but a 0.57% average error reduction is a population-level estimate. It does not prove that weather improves every individual start-sit decision, and it says nothing about how any specific player will perform in wind or rain this week.
This study estimates average weekly projection error for eligible historical player-weeks from 2019 through 2024. It does not prove that weather improves every player decision, and the position-level results are associations in projection error, not evidence that weather causes specific player outcomes.
METHODS · SOURCES · REVIEW TRAILOpen the Research File+
QUESTION
Does adding cutoff-safe six-hour stadium weather improve held-out weekly fantasy football projection accuracy versus the existing model?
METHOD
- Held-out evaluation on 20,700 NFL player-weeks across six rolling future-season windows, 2019 through 2024.
- Each season predicted only from earlier seasons; every weather forecast archived before its six-hour kickoff cutoff.
- Weather features: pregame air temperature, wind, precipitation probability, and a wind-by-precipitation interaction.
- Uncertainty from a 4,000-draw deterministic cluster bootstrap resampling 916 NFL games, with all player-weeks from a game moving together.
- Champion and challenger scored on identical player-week denominators; weather-missing games removed from both models.
LIMITS
- The 0.568% improvement is an average across eligible player-weeks and does not apply to every individual decision.
- Running backs regressed slightly (+0.24% MAE), the largest position-level regression observed.
- The 2024 season gap was nearly zero (-0.003 MAE points), so the edge varies by season.
- Results cover the 2019 through 2024 seasons and do not establish that weather helps in any specific future game.
SOURCES
- NOAA High-Resolution Rapid Refresh forecast archiveUS Government work; NOAA data are generally public domain
SHA-256 d36e91b1f7e564a5a836bd42673b3645f37f472a8b00f0b23b441ebfa9be52e0 - nflverse weekly NFL player statisticsCC BY 4.0
SHA-256 c34f15e2a79205fa0257f6b3890eac7b545689d5e21e1a20459a32705e13685b
REVIEW TRAIL
Statistics and figures come from deterministic code. The Research File contains the registration, provenance, specialist reviews, and gate decision.