Measure the condition, not the label
| Variable | Minimum record | Frequent error |
|---|---|---|
| Temperature | Value, unit, source, station and time | Using daily maximum as match value |
| Humidity | Relative humidity and time | Treating it as interchangeable with heat |
| Wind | Speed, gust, direction and height convention | Using regional average without venue check |
| Precipitation | Type, amount and observation interval | Recording forecast rain as observed rain |
| Altitude | Venue elevation and team acclimatisation context | Treating all away teams alike |
| Pitch | Surface, drainage and observed condition source | Inferring pitch state from rain alone |
A systematic review of real-match professional soccer studies found heterogeneous evidence across weather, altitude, physical output and technical performance (JSAMS Plus). That review supports precise measurement and cautious transfer, not a universal weather adjustment.
Align observations to the match clock
Use venue-level observations nearest the decision timestamp, retain source and update time, and distinguish measured conditions from forecasts. A live feature should not use a post-match weather report that was unavailable at the decision point.
Choose one target
Weather could be tested against physical output, pass completion, shot process, total goals, injury events or another defined target. These are not interchangeable. A change in running output does not automatically imply a change in goal probability.
Join football and weather data reproducibly
The Wyscout open-data paper documents match and spatio-temporal event data that can form the football side of a research join. Preserve venue identifiers, kickoff UTC, period, event time, score, player count and data version, then document the weather-station mapping and maximum accepted distance or time gap.
Illustrative sensitivity table
| Model version | Weather input | Question |
|---|---|---|
| Baseline | None | How well do score, time and team strength perform? |
| Measured | Venue observation at cutoff | Does contemporaneous weather add later information? |
| Forecast | Pre-match forecast | Could the feature have existed before kick-off? |
| Regional | Distant station | How sensitive is the result to source quality? |
| Missingness | Missing-weather flag | Is availability itself associated with competition coverage? |
Do not select only the weather specification with the strongest historical result. Predeclare the primary source and use alternatives as sensitivity checks.
Actual playing time
Under the IFAB duration law, allowance for time lost varies. If the target is an event rate, use actual exposure and account for interruptions that may correlate with weather.
Validation checklist
- Units, location, source and timestamp are retained.
- Forecast and observed weather are separate fields.
- Pitch condition is not inferred without evidence.
- The target and causal language are declared precisely.
- Baseline and weather-augmented models use the same later test set.
- Performance is reported by competition, season and missingness.
- Results include uncertainty and sensitivity to source mapping.
Continue with Weather and Pitch Conditions for the pre-match version of this workflow.
Continue learning
- Next guide: In-Play Betting Explained
- Related guide: Hedging Football Bets and Managing Open Positions
Assumptions and limitations
Weather studies differ in competition, measurement, target and model design. Venue microclimates and pitch treatment may not be captured by public stations. An incremental predictive result would not by itself establish causation or a favourable live price.
GambleAware
GamCare
Gambling Therapy
GAMSTOP