How do scientists decide if a specific marine heatwave was caused by climate change? Inside event attribution, from FAR to counterfactual oceans.
When a marine heatwave strips a coastline of kelp, the first public question is always the hardest one: was this climate change? The honest answer is neither a reflexive yes nor a cautious shrug. Attribution science has spent fifteen years building a rigorous machinery for exactly this question, and it can now deliver a defensible number: how much more likely — or how much hotter — human-caused warming made that specific event. This explainer walks through how marine event attribution actually works, where its numbers come from, and what it can and cannot tell a manager standing on a cleared reef. If you need the basics first, start with what a marine heatwave is and the crucial distinction in marine heatwave vs long-term warming.
Every attribution study begins by conceding two things. First, weather and ocean variability are real: warm-water events happened long before industrial emissions, driven by El Niño, atmospheric blocking, eddies and sheer bad luck. Second, no scientist can look at a single event in a single season and decompose it into "70% climate change, 30% nature" as if it were a recipe.
What attribution can do is compare probabilities across two worlds: the one we live in, and a counterfactual one without human greenhouse forcing. The method is inherited from atmospheric science, where it was developed to answer questions about heatwaves on land, and adapted to the ocean as long, high-quality sea surface temperature records became available.
The logic runs like this. Take an observed marine heatwave — say, thirty days of water four degrees above the seasonal norm off a particular coast. Researchers ask two quantitative questions:
Neither question requires anyone to deny natural variability. They ask only how the dice have been loaded.
The headline number in most attribution papers is the Fraction of Attributable Risk, or FAR. It converts the probability ratio into something readable: the share of the event's likelihood that can be ascribed to human influence. A FAR of 0 means climate change played no detectable role in the odds; a FAR of 0.9 means such an event is ten times more likely in today's climate than it would have been without emissions — nine-tenths of its likelihood is attributable.
FAR answers the question people actually ask — "would this have happened anyway?" — with a probabilistic, honest answer: probably not, at this magnitude, in this era; or possibly, though far more rarely.
Two traps follow. First, FAR is about probability, not causation in the courtroom sense. A FAR of 0.95 doesn't mean 95% of the high temperatures came out of a smokestack; it means a world without human forcing would see such conditions roughly one-twentieth as often. Second, FAR depends heavily on how the event is defined. A thirty-day anomaly of +4 °C has one FAR; a five-day spike of +6 °C in the same bay has another; annual-mean warming has a third. Much of the apparent disagreement between attribution studies is really disagreement about event definition.
Strip away the jargon and every marine heatwave attribution study moves through the same six steps.
1. Define the event precisely. Region, season, depth or satellite sea-surface layer, duration, and the metric — peak anomaly, cumulative degree-heating days, or season-long warmth. Loose definitions make loose answers.
2. Estimate the event's rarity in the observed record. Researchers fit a statistical distribution (often a generalised extreme-value distribution, which is built for tails) to decades of observed temperatures at or near the site. This yields a return period: "the 2016 event was approximately a 1-in-200-year occurrence in the observational climate."
3. Build the counterfactual world. Climate model ensembles simulate thousands of years of ocean conditions both with observed levels of forcing and with pre-industrial forcing. Some studies also use stipulated counterfactuals — for example, subtracting the estimated anthropogenic warming trend from the observed record.
4. Repeat the statistics inside the models. Fit the same distributions to modelled temperatures in both worlds, then compute the probability ratio and intensity shift for the observed event.
5. Test the models against reality. An attribution is only credible if the models reproduce the local variability, seasonal cycle and temperature tails of the actual ocean. An ensemble that can't simulate boundary current behaviour or coastal upwelling around the study site fails this test and its conclusions get discounted.
6. Report with uncertainty. Good studies give a probability ratio with a range — "between 5 and 40 times more likely" — because sensitivity across models, event definitions and statistical choices is itself information.
Marine attribution arrives later than its atmospheric cousin for good reasons.
Memory. The ocean integrates heat over years, not days. Anomalously warm conditions can persist, dissipate and even re-emerge as subsurface layers are mixed back to the surface — behaviour visible in the rise of ocean heat content globally. That memory stretches the concept of an "event" and makes clear start and end dates slippery.
Thin coastal data. Satellite sea-surface temperature is superb over the open ocean but coarse (kilometres to tens of kilometres) in the narrow, cryptic band where kelp lives. The coastal ocean holds sharp fronts, tidal mixing and upwelling cells that global climate models — running at roughly 100-kilometre grid scales — resolve poorly. Attributing a coastal marine heatwave often demands regional downscaling or special-purpose high-resolution models.
Natural modes are strong. El Niño and its cousins swing regional temperatures by degrees on multi-year cycles, and they interact with warming in non-linear ways. Attribution studies must decide whether to condition on the natural state — "given a strong El Niño year, how much hotter did climate change make it?" — a choice that materially changes the answer but, done transparently, is highly informative for managers.
Observed baselines are short. The gold-standard satellite record begins in 1981–82. That's barely forty years against events with nominal return periods of centuries. This is where model ensembles do the heavy lifting: they provide the thousands of simulated years that the observational record cannot.
Across dozens of published marine heatwave attributions, a few robust patterns have emerged:
For coastal councils, fisheries and restoration practitioners, attribution numbers have concrete uses:
Attribution cannot, however, absolve local stressors. Whether climate change loaded the dice has no bearing on whether sewage discharge, overfishing of urchin predators, or sediment plumes made a particular reef more vulnerable to a heatwave when it arrived. Good coastal science holds both truths at once: climate change raised the risk; local management shapes the consequences. Our work at Tidewright, documented on the impact dashboard, assumes both.
Three developments will reshape the field over the coming decade:
Not for a single event, in the strict sense. The scientifically defensible statement is always probabilistic: climate change made this class of event much more likely or more intense. For extreme enough events — with enormous FAR values — scientists do say the event was "virtually impossible without climate change," which is about as close to causation as statistics gets.
Detection asks whether an observed change — say, rising marine heatwave frequency — is distinguishable from natural variability. Attribution asks how much of that change is due to particular causes, like greenhouse gas emissions. Event attribution applies the same logic to individual episodes.
Usually because they define the event differently — different regions, durations, metrics or seasons — or use different model ensembles and baselines. This is a feature, not a failure: each framing answers a different, legitimate question about risk.
Not directly — it looks backward at odds and causes. But its outputs feed forward-looking risk projections. If the odds of damaging events are rising steeply, planning for recurrence is the rational response, whatever any single year brings.
No. Natural variability and human forcing act together: El Niño still happens, but now it happens on a warmer baseline, so regional extremes that used to be rare become common. Attribution studies that condition on natural states make exactly this point rigorously.
Marine heatwaves are weeks of anomalously hot ocean that can deforest entire coastlines. Here's the science behind them and what restoration can do.
Dr. Lena Fischer ·
Marine heatwaves are spikes riding on a rising baseline of ocean warming. How scientists separate the two signals — and why the difference matters for reefs.
Dr. Lena Fischer ·
Over 90 percent of the planet's trapped heat is in the ocean. How we measure ocean heat content, from expendable probes to 4,000 drifting floats — and why it matters.
Dr. Lena Fischer ·