Major League Soccer in 2026 will look unlike any earlier season: more teams, longer travel, and a deeper appetite for analytics. This article lays out a practical, evidence-based approach to building a goal model that explicitly accounts for flights and climate — factors that matter in a continent-spanning league. I walk through the why, the data, the modeling choices, and how coaches and analysts could use such a model in real time.
Why flights and climate matter in modern MLS
MLS stretches from Vancouver to Miami, from the desert Southwest to the high plains of Colorado, and that geography has grown more consequential as the schedule densifies. Long flights, multiple time-zone crossings, and sharply different temperatures and humidity levels create physiological and logistical stresses that can change on-field performance.
Teams arriving after overnight flights, facing late kickoffs, or playing in sweltering heat are not competing on a level playing field compared to rested, acclimatized opponents. The differences are subtle for a single shot or run, but they add up across a match and across a season.
Recognizing those effects matters for two reasons: predictive accuracy and decision support. A model that ignores travel or climate risks biasing predictions and missing levers that coaching staffs can actually manage.
What traditional goal models capture — and what they miss
Most modern goal expectancy frameworks start at the shot level. They model the probability a shot becomes a goal using features such as shot location, body part, assist type, defensive pressure, and game context. Providers like StatsBomb and Opta have made high-resolution event data the standard input for expected goals (xG) models.
Those shot-level models are powerful for scouting and in-game decisions, but they usually treat each event in isolation. Match-level and seasonal effects such as travel fatigue, circadian misalignment, heat stress, and altitude exposure are often relegated to team fixed effects or ignored altogether.
Integrating travel and climate explicitly means adding another layer: how environmental and logistical context shifts the baseline probability of scoring or conceding, either by changing player physiology or by influencing tactical choices.
How flights and climate change performance: mechanisms and evidence
Circadian disruption from crossing time zones alters alertness, reaction time, and decision making. Sleep loss from red-eye flights and hotel disturbances compounds that effect. Sports-science reviews show these factors can blunt cognitive and physical performance for 24–72 hours after travel, depending on the direction and magnitude of time-zone change.
Heat and humidity degrade aerobic capacity and sprint repeatability, impairing the all-important transition moments that create high-quality chances. Thermoregulatory strain alters pacing and increases perceptual load, which can reduce the precision of finishing under pressure.
Altitude reduces available oxygen and can reduce high-intensity work rate for unacclimatized players. Mountain venues — Denver’s Empower Field or Salt Lake City’s stadiums — consistently present physiological challenges that are distinct from heat or travel fatigue.
Data sources: building the input layer
A robust model needs three classes of data: event-level match data, travel and schedule metadata, and climate/environmental observations. Event data comes from providers such as StatsBomb or Opta and supplies the standard xG inputs.
Flight and schedule data can be constructed from team itineraries (when available), public flight schedules, and league fixtures. Useful fields include departure and arrival timestamps, total elapsed travel time, time zones crossed, and layover duration. In league-level work, the U.S. Bureau of Transportation Statistics and airline schedules are practical starting points.
Climate inputs should include temperature, relative humidity, wet-bulb globe temperature (WBGT) where possible, precipitation, wind speed, and stadium altitude. NOAA, local meteorological services, and global reanalysis datasets provide consistent historical observations for every match location.
Feature engineering: what to add and why
Turn raw inputs into features that the model can use. For travel: hours since arrival, time zones crossed, red-eye indicator (overnight flight within 24 hours), and cumulative travel minutes in the prior seven days. For climate: kickoff temperature, humidity, WBGT, and whether the match is indoors or outdoors.
Complement these with team- and player-level controls: rolling Elo or expected-goals-based rating for team strength, days of rest, rotation index, and injury/suspension flags. These controls reduce confounding when estimating travel and climate effects.
A practical feature table helps communicators and modelers align expectations. The table below summarizes common features and the direction a model might expect them to push goal probability.
| Feature | Type | Expected effect on scoring probability |
|---|---|---|
| Hours since arrival | Travel | Fewer hours → lower scoring probability |
| Time zones crossed | Travel | More zones → lower scoring probability |
| Red-eye flight (yes/no) | Travel | Yes → lower scoring probability |
| Kickoff temperature (°F/°C) | Climate | High temp → lower scoring probability for high-intensity plays |
| WBGT | Climate | High WBGT → reduced physical output → lower scoring probability |
| Stadium altitude (meters) | Climate/Geography | High altitude → potential advantage for home team |
Model architecture: integrating contextual modifiers with xG
There are two broad strategies: adjust xG at the shot level with contextual modifiers, or build a hierarchical model that nests shots within match and travel contexts. Both approaches are viable; the best practical route for MLS analytics is a mixed-effects shot-level model.
Start with a baseline shot-level xG model (logistic or gradient-boosted tree) trained on standard event features. Then layer a mixed-effects component where travel and climate variables act as match-level predictors with random intercepts for teams and seasons. This structure lets the model capture how the same environmental stress shifts baseline probabilities without discarding the high-resolution shot features.
For interpretability and deployment, a two-stage approach works well: compute baseline xG for each shot, then compute an environmental multiplier that scales xG up or down. The multiplier can be learned with a separate regression or with a Bayesian hierarchical model that produces credible intervals for the effect size.
Algorithmic choices and explainability
Gradient-boosted machines (LightGBM, XGBoost) often produce strong shot-level xG performance, especially with complex interactions like shot location × assist type. However, tree models are less transparent about causal effects. Mixed-effects logistic regression gives clearer parameter interpretation for travel and climate coefficients.
A recommended hybrid is: use gradient boosting for the baseline xG, then fit a generalized linear mixed model (GLMM) on the residuals with travel and climate covariates. This preserves predictive power while yielding interpretable estimates of environmental impacts.
For teams that require online predictions, a simpler calibrated multiplier approach can run in real time: the precomputed baseline xG is adjusted by a small factor based on travel and weather inputs that are known before kickoff.
Validation: how to test the model
Validation needs to be twofold: predictive and quasi-experimental. For predictive validation, use season-over-season backtests and holdout years, paying attention to calibration and Brier score at the shot and match levels. Compare the adjusted model against a baseline xG-only model to quantify incremental gain.
For quasi-experimental validation, leverage natural experiments. Examples include midweek vs. weekend travel, abrupt changes in kickoff time, or multi-match road trips. Difference-in-differences designs — controlling for opponent and venue fixed effects — can estimate causal shifts associated with defined travel events.
Cross-validation must preserve time ordering: train on earlier seasons and test on later ones. That prevents leakage from calendar-dependent features like schedule density and climate patterns.
What effect sizes are realistic?
Be cautious with claims about magnitudes. The literature and operational projects suggest travel and climate create modest but consistent effects: small changes in shot conversion rates and slightly larger shifts in team-level outcomes like expected goal difference. Those changes, multiplied across a season, can explain meaningful parts of the variation in standings.
Pragmatically, expect environmental modifiers to adjust shot-level probabilities by single-digit percentage points in most cases; the largest effects appear in extreme heat, substantial altitude differences, or immediate post-red-eye matches. Those modifiers are enough to change win probability in tightly contested fixtures.
Practical applications for clubs and the league
Clubs can use the model for planning travel windows, resting key players, and deciding substitution timing. If the model flags a meaningful decline in finishing probability after a particular flight pattern, coaches can prioritize conserving energy or altering pressing schemes to reduce high-intensity sprints.
The league can use aggregated model outputs to assess fairness in the schedule and to optimize kickoff times to limit circadian disadvantages. For broadcast partners and bettors, a more accurate model produces fairer lines and better fan engagement through richer pregame narratives.
Real-world tweaks are small but actionable: shifting practice time to match local game time after travel, implementing targeted cold-water immersion in hot venues, or arranging non-stop flights to reduce layover stress. Those interventions are inexpensive relative to player wages but can move the needle when margins are thin.
Roster management and player-level considerations
Not every player responds to travel or heat the same way. Older players, those with limited travel tolerance, or players with preexisting sleep disorders may be disproportionately affected. Incorporating player-level susceptibility — from historical performance after travel and wearables like sleep metrics — refines the model’s recommendations.
Teams with deep benches can substitute proactively in matches flagged as high-risk for fatigue-related lapses. A goal model that returns both team-level and player-level adjustments enables smarter rotation without unnecessary starting-lineup changes.
Case studies and hypothetical scenarios
Consider a midweek match where a Pacific Northwest team flies to Miami for a Sunday evening kickoff after a red-eye. The combined travel and circadian stress could reduce the expected conversion rate of transition shots late in the match. The model would lower the team’s xG for high-intensity sequences and flag an increased need for controlled possession to reduce chance quality reliance.
Contrast that with a home match in Denver following a week where the visiting team spent the previous week at sea level. The altitude differential can manifest as more high-quality chances for the home side, especially early in the second half when oxygen debt accumulates among unacclimatized players. Coaches can exploit that window by pressing more aggressively after halftime.
In a project I worked on for a different league, incorporating travel metadata improved season-long predictive performance and led to measurable operational changes in how one club scheduled rest days. Those changes were modest but correlated with a small increase in points-per-game over congested stretches.
Deployment: pipelines, latency, and operational constraints
Operationalizing this model requires a reliable pipeline: automated ingestion of event data, daily weather pulls for upcoming locations, and an up-to-date travel dataset. Precompute multipliers for scheduled matches and update them if itineraries change.
Latency matters less for match-day analytics and more for pregame planning. Clubs should prioritize model runs 48–72 hours before kickoff to allow time for tactical or roster adjustments. Live in-game adjustments can use the model’s pregame forecast combined with in-match telemetry and player substitutes.
Secure handling of team itinerary data is essential. Clubs may treat travel plans as sensitive; if internal access is limited, models can revert to public flight schedules and still capture most of the variation in travel-related stress.
Limitations, pitfalls, and ethical considerations
Correlation is not causation. Even with strong covariate control, unobserved variables — like psychological readiness or locker-room disputes — can confound estimates of travel and climate effects. Avoid overinterpreting small coefficients as actionable truths without operational testing.
There is also an ethical layer. Using a model to justify canceling or disproportionally benching players could unfairly penalize individuals who are not actually impacted. Transparent, team-level dialogue and respecting players’ welfare must guide deployment.
Finally, data quality is a limiting factor. Incomplete or noisy travel logs, imprecise local microclimate measurements, and inconsistent wearable data can all reduce the reliability of estimated effects. Plan for data audits and versioned models.
Next steps and research opportunities for 2026
As MLS grows and the sport embraces more wearable and physiological data, the most promising direction is integrating individualized biomarkers into the environmental model. Sleep metrics, heart-rate variability, and external load from GPS can transform a population-level adjustment into a player-specific prediction.
Another avenue is causal field experiments. Clubs and the league can test simple interventions — altered travel schedules, in-stadium cooling strategies, or differing recovery protocols — and use the model to measure outcomes. Those randomized or quasi-randomized trials would strengthen causal claims.
Finally, expanding validation to include alternative outcomes — such as high-quality scoring chances, pressing success, or injury risk — will make the model richer and more useful across sports operations.
How teams can pilot this approach in the 2026 season
Start by building a minimum viable product that merges baseline xG with a small set of travel and climate modifiers. Validate it on historical MLS seasons and run a shadow trial during a congested spell, offering the outputs to coaching staff as advisory rather than prescriptive guidance.
Document cases where the model’s signal aligns with observed fatigue or climate-driven shifts in performance. Use those examples to secure buy-in for more ambitious integrations, such as linking the model to training periodization or substitution strategy planning.
A disciplined rollout, combined with careful measurement, will be the difference between an academic exercise and a tool that influences points on the table.
Sources and expert contributors
- Major League Soccer — https://www.mlssoccer.com
- StatsBomb analytics and blog — https://statsbomb.com
- StatsPerform / Opta — https://www.statsperform.com
- National Oceanic and Atmospheric Administration (NOAA) — https://www.noaa.gov
- U.S. Bureau of Transportation Statistics — https://www.bts.gov
- FIFA Football Medicine and Research — https://www.fifa.com/what-we-do/medical
- John Waterhouse, PhD — circadian rhythms and sport (selected reviews on PubMed)
- Julien Périard, PhD — thermoregulation and sports performance (selected publications)
Full analysis of the information was conducted by experts from sports-analytics.pro
