📄 earth_science

Heat, wind, and the calendar: why a warming climate makes its performance effects unidentifiable at the IRONMAN World Championship

Using a comprehensive dataset of 42,605 finishers from 2002 to 2025, this study demonstrates that common analytical flaws and the strong correlation between race-day heat and the calendar year render heat effects on IRONMAN performance unidentifiable, leaving wind load as the only measurable environmental factor.

Original authors: Andreas Will¹, Romuald Lepers², Beat Knechtle³

Published 2026-07-14

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Andreas Will¹, Romuald Lepers², Beat Knechtle³

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the IRONMAN World Championship in Kona, Hawaii, as the ultimate science lab for endurance athletes. For years, scientists and fans have looked at the race results from 2002 to 2025—a massive dataset of over 42,000 finishers—and tried to answer a simple question: Does the heat make people slower?

The answer, according to this new study, is a bit of a plot twist. While the heat is definitely real, and the climate is definitely getting warmer, this specific race archive cannot prove that the heat is slowing the runners down. In fact, the study suggests that the famous "heat penalty" people see in the data is mostly a trick of the math, a confusion of cause and effect, and a case of mistaking talent for weather.

Here is the story of what the researchers found, broken down into three main acts.

Act 1: The "Circular" Stopwatch Trap

First, the researchers looked at how people used to measure "heat load." Imagine you are trying to see if running in the rain makes you slower. You decide to measure "rain-load" by multiplying how hard it's raining by how long you run.

The problem? If you run for 4 hours, you get a huge "rain-load" number. If you run for 3 hours, you get a smaller number. But if you are slower because of the rain, you run for 4 hours! So, your "rain-load" number is high because you were slow, not the other way around. It's a circular trap.

The study found that the old way of calculating heat for Kona was exactly like this. They used a "cumulative heat load" (measured in °C·h, or degree-Celsius hours) that was mathematically tied to how long the race took. The correlation was incredibly high (r = 0.93), making it look like heat was the main villain. But when the researchers switched to a "duration-independent" measure (just the temperature intensity, ignoring the time), the magic spell broke. The apparent link between heat and slowness dropped significantly. The old "heat penalty" was largely an artifact of the math, not a physical law.

Act 2: The Calendar Confusion

Next, the team tackled the biggest mystery: Why does it look like heat is ruining the race?

They discovered that the Kona race days have been getting hotter. The data shows a clear warming trend of +1.28 °C per decade in the Wet-Bulb Globe Temperature (WBGT). That is a real, measurable warming.

However, here is the catch: The race happens once a year. As the years go by (the calendar), the temperature goes up. At the same time, athletes get faster because of better shoes, better training, and better bikes.

The study found that the race-day heat is 99.5% explained by the calendar year alone. This means the heat and the "era" are locked together like two dancers who never let go. Because the heat always goes up as the years go by, you can't tell if a runner is slower because it's hot, or if they are slower because they are from an older era with less technology.

Think of it like trying to figure out if a car is slower because of the wind or because it's an older model. If every single year you test a newer car in stronger wind, you can't separate the two. The study concludes that in this single-location, single-race-per-year design, the heat effect is unidentifiable. The "sample size" isn't 42,000 athletes; it's really just 23 different race-day climates. The huge number of athletes just makes the confusion more precise, not more clear.

Act 3: The "Superhero" vs. The "Wind"

So, if it's not the heat, what explains why some people run faster after biking?

For decades, people thought the conditions on the bike leg (the first part of the race) somehow "coupled" to the run leg (the second part). They thought, "If the bike leg was hot and windy, the runner must be slower."

The study used advanced machine learning to separate "athlete ability" from "environment." The result? The connection between the bike and the run is 99% about the athlete's natural talent and only about 1% about the conditions.

If an athlete has a fast bike split, they usually have a fast run split. That's because they are just a fast person, not because the weather was nice. The study found that adding weather data to the prediction model improved the accuracy by a tiny, almost invisible amount (about 2.5 seconds over the whole race). The "coupling" is almost entirely the athlete's ability, not the environment.

But there is one hero that survives the heat: The Wind.

While the heat effect vanished into the noise, the wind on the bike leg was a different story. The study found that wind load is a genuine, measurable force that slows cyclists down. In fact, the wind effect on the bike leg could account for a difference of about 50 minutes between the calmest and windiest days.

Crucially, the study also found that you can't measure this wind with just one weather station. A single station near the coast might miss the strong crosswinds in the northern part of the course (near Hawi). To catch the real wind effect, you need a "distributed" map of sensors all along the course. For heat, one station is fine; for wind, you need the whole map.

The Final Verdict

The paper doesn't say heat is safe or that it doesn't hurt your body. It says that this specific dataset cannot prove it.

The Heat: The warming is real (+1.28 °C per decade), but because the race happens once a year, we can't separate the heat from the passing years and better technology. The "heat penalty" seen in past studies was likely a math trick or a confusion with the calendar.
The Wind: This is the only environmental factor that clearly, measurably slows down the cyclists, and it requires a detailed map to see.
The Athlete: The reason some people run fast after biking is that they are just good athletes, not because the weather was perfect.

The researchers suggest that to truly understand how heat affects performance, we need to look at individual athletes over many races (like a "within-subject" study), not just a giant pile of race results from one place. For now, the Kona archive tells us that the wind is the boss of the bike leg, but the heat remains a mystery hidden behind the calendar.

Technical Summary: Heat, Wind, and the Calendar in IRONMAN Performance Analysis

Problem Statement
The IRONMAN World Championship in Kona, Hawaii, serves as a canonical dataset for studying the impact of environmental conditions on ultra-endurance performance. As global temperatures rise, there is increasing interest in quantifying how heat and wind affect race outcomes. However, previous analyses of this dataset have relied on methodological choices that may introduce circularity, confounding, and misattribution of variance. The central problem addressed is whether the apparent environmental effects reported in Kona race archives are genuine physiological signals or statistical artifacts arising from the study design, specifically the collinearity between calendar time, climate trends, and performance improvements.

Methodology
The study utilizes the most comprehensive dataset to date, comprising 42,605 finishers from the IRONMAN World Championship between 2002 and 2025 (excluding 2020 and 2021). The analysis employs a rigorous statistical framework to address four specific research questions regarding identifiability and confounding:

Data Construction & Environmental Reconstruction:
- Cohort: Able-bodied finishers in standard age groups (18–69) and professionals.
- Environmental Indices: The authors reconstructed course-distributed, solar-validated Wet-Bulb Globe Temperature (WBGT) and wind loads using ERA5 reanalysis data sampled at geolocated waypoints along the 180-km bike and 42.2-km run courses.
- Index Revision: The study explicitly rejects the "time-weighted cumulative heat load" (°C·h) used in prior literature, noting it is mechanically proportional to race duration. It replaces this with duration-independent intensity measures (°C) and exogenous fixed-window references.
- Statistical Modeling: Performance is modeled using Generalized Additive Models (GAMs) with thin-plate smooths for environmental indices, sex-specific year smooths to absorb secular trends, and parametric controls for sex, age, and technological eras (specifically carbon-plated running shoes post-2017 and an "Optimization" era for cycling post-2016).
- Validation: Results are triangulated using Extreme Gradient Boosting (XGBoost) with leave-years-out cross-validation and Accumulated Local Effects (ALE) plots to characterize non-linear relationships without extrapolation.
Analytical Strategy:
- The study partitions variance in run times into components attributable to demographics, secular trends, cycling performance carryover, and environmental factors.
- It tests the hypothesis that a single weather station is insufficient compared to distributed sampling.
- It assesses the "identifiability" of heat effects by examining the collinearity between race-day climate variables and the calendar year.

Key Contributions
The paper contributes a methodological critique and a corrected empirical analysis of the Kona dataset:

Identification of Circular Indices: It demonstrates that the cumulative heat-load index (°C·h) is an artifact because it embeds the outcome variable (duration) into the predictor, inflating apparent correlations.
Confounding by Calendar: It establishes that in a single-venue, annual race design, race-day heat is statistically inseparable from the secular trend of the calendar year. Because the climate has warmed over the study period, "heat" is perfectly aliased with "time," making it impossible to isolate a causal heat effect on performance without within-subject variation.
Decomposition of Bike-to-Run Coupling: It quantifies the "bike-to-run" relationship, separating athlete ability (performance carryover) from environmental carryover.
Spatial Sampling Necessity: It clarifies that while distributed sampling is redundant for capturing heat (which is spatially coherent), it is essential for capturing wind load, which varies significantly across the course.

Results

Circularity of Heat Load: The time-weighted cumulative heat load correlated with run duration at $r = 0.93$ . When replaced with duration-independent intensity, the correlation dropped to $r = -0.24$ , and the apparent association with performance reduced significantly (partial $r \approx 0.25$ to $0.33$).
Unidentifiable Heat Effect: Race-day WBGT warmed by $+1.28^\circ\text{C}$ per decade ( $p < 0.001$ ). Consequently, 99.5% of the variance in day-reference WBGT is explained by the calendar year alone. Once the year trend is controlled for, no residual heat variation remains to estimate an effect. The effective sample size for environmental inference is approximately 23 race-day climates, not 42,605 athletes.
Bike-to-Run Coupling: The apparent coupling between cycling conditions and running performance is overwhelmingly driven by athlete ability.
- Adding cycling performance (split time and finish clock) to a run model increased the explained variance ( $\Delta R^2$ ) by 0.379 (99% of the gain).
- Adding cycling environmental conditions and heat-fatigue interactions increased $\Delta R^2$ by only 0.003 (1%).
- The 95% confidence interval for the environmental coupling was $[-0.006, +0.003]$ , indistinguishable from zero.
Identifiable Wind Effect: Unlike heat, wind load on the cycling leg retained genuine explanatory power. The wind index explained a partial effect span of approximately 50 minutes from calm to windy conditions. This effect was identifiable because wind varies within a year across the course and start waves, unlike the monotonic warming trend.
Sampling Resolution: A single weather station captured heat as well as distributed sampling (correlation $r = 0.81$ to $0.87$). However, for wind, single stations (e.g., the pier) failed to capture the course-mean wind (correlation $r = 0.20$ ), whereas distributed sampling was required to capture the specific crosswind regimes driving the identified wind effect.

Significance and Claims
The paper claims that the headline environmental-performance effects previously attributed to the IRONMAN World Championship are largely artifacts of analytical choices and study design limitations.

Methodological Warning: The study serves as a cautionary template for environmental performance research, demonstrating that single-venue annual designs cannot identify heat effects when climate trends with time, regardless of sample size.
Reframing the Narrative: It asserts that the "bike-to-run" coupling is a function of general athlete ability, not environmental conditions. The only robust, identifiable environmental determinant in this dataset is wind load on the bike.
Physiological vs. Statistical: The authors emphasize that their null result regarding heat does not contradict the established physiology of heat impairment (which is well-documented in controlled studies). Rather, it highlights that the specific aggregate, single-venue design of the Kona archive is insufficient to resolve these effects due to collinearity.
Practical Implication: For practitioners, the study suggests that while heat management remains critical for safety, aggregate race records cannot be used to predict specific time penalties for heat. Conversely, wind load is a quantifiable and actionable environmental factor for the cycling leg.

The paper concludes that future research into heat effects on endurance performance requires within-subject, repeated-measures designs or multi-venue comparisons to break the collinearity between climate and calendar time.

Act 1: The "Circular" Stopwatch Trap

Act 2: The Calendar Confusion

Act 3: The "Superhero" vs. The "Wind"

The Final Verdict

Technical Summary: Heat, Wind, and the Calendar in IRONMAN Performance Analysis

More like this