A scientist says global sea level could rise by a certain amount by 2100.
A newspaper reports that sea level will rise by that amount.
Decades later, someone finds the headline and calls it a failed scientific prediction.
Was it?
Maybe.
Maybe not.
It depends on what the original scientific statement actually was.
That distinction is at the heart of almost every attempt to audit historical climate predictions.
The words prediction, projection, scenario, forecast, likely, could, may and will are often treated as though they mean approximately the same thing.
They do not.
And when those distinctions disappear, a conditional scientific statement can be transformed into something its authors never actually predicted.
The reverse problem also exists.
A clear, dated prediction can later be defended as “only a scenario” even when its author presented it as an expected outcome.
So the rule has to work both ways.
PRESERVE THE ORIGINAL WORDING.
Then determine what kind of statement was actually made.
Prediction
The IPCC defines a climate prediction — or climate forecast — as an attempt to estimate the actual future evolution of the climate.
Because the climate system contains uncertainty, these predictions are often probabilistic rather than absolute.
In ordinary language, a prediction answers a question like:
What do we expect to happen?
For example:
Arctic summer sea ice is expected to fall below a specified threshold before a particular date.
If that statement genuinely attempts to describe what will actually occur, it can eventually be compared with observation.
Did the event happen?
Did it happen before the deadline?
Was the predicted magnitude correct?
That is a prediction audit.
But now consider a different statement.
Projection
The IPCC defines a climate projection differently.
A projection is a simulated response of the climate system to assumptions about future greenhouse-gas emissions, concentrations, aerosols, land use or radiative forcing.
The critical word is:
ASSUMPTIONS.
Climate projections depend on conditions that may or may not occur.
A projection therefore answers something closer to:
What would the climate look like if this particular set of assumptions occurred?
That is not quite the same question as:
What will actually happen?
Consider a simplified example.
A climate model might produce:
IF emissions follow pathway A → temperature outcome X
and
IF emissions follow pathway B → temperature outcome Y.
Those are conditional projections.
If pathway A never occurs, it makes little sense decades later to compare the observed climate with outcome X and declare the climate model wrong without mentioning that the input conditions were different.
The condition is part of the projection.
Remove it and you have changed the claim.
Scenario
Now we move one step further back.
A scenario is not principally a climate outcome.
It describes a possible future world.
The IPCC defines a scenario as a plausible description of how the future may develop based on a coherent and internally consistent set of assumptions about important driving forces and relationships.
And the IPCC states explicitly:
scenarios are neither predictions nor forecasts.
A scenario might contain assumptions about:
population,
economic development,
energy use,
technology,
land use,
government policy,
greenhouse-gas emissions,
and other forces.
Those assumptions can then be fed into models.
The model generates projections.
So the chain can look approximately like this:
SCENARIO → EMISSIONS / CONCENTRATIONS → CLIMATE MODEL → PROJECTION
A prediction is conceptually different:
WHAT DO WE EXPECT WILL ACTUALLY HAPPEN?
That distinction sounds technical.
It becomes extremely important when deadlines arrive.
A Simple Example
Imagine three statements.
Statement A
“Global temperature will rise 2°C by 2050.”
That reads like a prediction.
There is an expected outcome.
There is a number.
There is a deadline.
When 2050 arrives, the statement can be tested directly, provided its baseline is known.
Statement B
“Under emissions scenario X, global temperature is projected to rise 2°C by 2050.”
Now we have a projection.
Before judging it, we must ask:
Did emissions actually follow scenario X?
If they did not, comparing the outcome directly with the projection may tell us little about the model's performance.
Statement C
“Scenario X assumes rapid population growth, continued fossil-fuel expansion and high future greenhouse-gas emissions.”
That is a scenario.
It is a possible future used for analysis.
It is not necessarily a claim that this future will occur.
Three sentences.
Three very different things.
Then Comes Probability
Even a prediction may not say that something definitely will happen.
Suppose a scientist says:
“There is an 80% probability that event X will occur before 2030.”
2030 arrives.
The event does not occur.
Was the prediction falsified?
Not in the same sense as:
“Event X will occur before 2030.”
A probability forecast allows for the event not occurring.
An 80% forecast implies a 20% possibility of non-occurrence.
To evaluate probabilistic forecasting properly, one normally needs many forecasts, not just one.
If events assigned an 80% probability occur approximately 80% of the time across a sufficiently large sample, the forecasting system may be well calibrated.
One event alone cannot establish that.
This creates a problem for historical prediction audits.
Some famous statements sound categorical in headlines but were probabilistic in their original source.
The probability cannot simply be deleted.
“Could” Is Not “Will”
Now we reach one of the most important words in climate communication.
COULD
Suppose a study says:
Sea level could rise by up to one meter under a high-emissions scenario.
Four qualifiers are embedded in that sentence:
could
up to
under
high-emissions scenario
Now imagine the headline:
SEA LEVEL WILL RISE ONE METER
Every qualifier has disappeared.
The number survived.
The meaning did not.
That is why this series never audits only the number.
It audits the sentence containing the number.
“Up To” Is Not the Expected Value
The phrase up to creates another common problem.
Suppose a projected range is:
20–60 centimeters.
The upper boundary is 60 centimeters.
That does not automatically mean scientists expect 60 centimeters.
Yet headlines naturally prefer the largest number.
“Sea level may rise 20 centimeters” attracts less attention than:
SEA LEVEL COULD RISE 60 CENTIMETERS.
Then another transformation can occur:
SEA LEVEL WILL RISE 60 CENTIMETERS.
Years later, 60 centimeters becomes remembered as the prediction.
But the original scientific assessment may never have identified it as the most likely outcome.
Likely Range ≠ Maximum ≠ Worst Case
Climate assessments can contain several different numerical categories simultaneously.
There may be:
a central estimate,
a likely range,
a very likely range,
a high-end outcome,
a low-confidence outcome,
and physically possible but highly uncertain outcomes.
They cannot be exchanged freely.
The largest number in a report is not automatically:
THE SCIENTIFIC PREDICTION.
But the opposite rule also applies.
If an assessment explicitly identifies a range as likely, most likely, or otherwise assigns probability to it, that information is part of the historical forecast and can eventually be examined.
The label matters.
The Arctic Shows the Problem
Consider our earlier audit of Arctic sea ice.
Some public statements predicted an effectively ice-free Arctic on very short timelines.
Others described probabilities.
Others discussed model projections.
Others used “nearly ice-free” rather than literally zero ice.
And the scientific convention often defines an effectively ice-free Arctic as less than approximately one million square kilometers of sea ice — not the complete disappearance of every piece of ice.
If all those statements are compressed into:
“Scientists said there would be no Arctic ice by 2013.”
the historical record has been altered.
But if somebody really did attach a specific early date to the disappearance or near-disappearance of summer sea ice, that statement should not be erased either.
The solution is simple:
QUOTE THE ACTUAL CLAIM.
Glacier National Park Shows Another Version
The Glacier National Park story demonstrates how projections can evolve into public deadlines.
Scientific modeling originally examined glacier retreat under warming conditions.
A disappearance estimate around 2030 became associated with the park.
Later, faster observed retreat contributed to an informal revision toward approximately 2020, and the date appeared in public park displays.
The glaciers remained after 2020.
What failed?
Not the observation that glaciers were retreating.
They were.
Not the physical possibility that continued warming would reduce glacier area.
It has.
What failed was the specific public disappearance deadline.
Again:
TREND ≠ DEADLINE
The Maldives Shows Yet Another Problem
The Maldives case involved a dramatic consequence.
Sea-level rise is measurable.
The Maldives are extremely low-lying.
Their exposure to future sea-level rise is real.
But a 1988 report said the islands could be completely covered within approximately 30 years.
That deadline implied roughly 2018.
2018 arrived.
The Maldives remained inhabited.
So the correct audit does not say:
SEA-LEVEL RISE WAS FALSE.
It says:
THE COMPLETE-SUBMERGENCE OUTCOME DID NOT OCCUR BY THE STATED DEADLINE.
The distinction preserves both pieces of evidence.
Polar Bears Demonstrate the Importance of Geography
Our polar-bear audit produced another useful rule.
A projection concerning one subpopulation cannot automatically become a prediction concerning the entire species.
LOCAL POPULATION ≠ GLOBAL SPECIES
A forecast about Western Hudson Bay cannot simply become:
POLAR BEARS WILL BECOME EXTINCT.
Likewise, a projected decline is not identical to extinction.
DECLINE ≠ EXTINCTION
Again, the nouns matter just as much as the numbers.
Sea Level Demonstrates the Importance of Baselines
Suppose a historical report projects:
+30 centimeters by 2025
and modern satellite data report:
+10 centimeters since 1993.
Those numbers cannot be compared honestly until we know the historical projection's baseline.
Was the +30 cm relative to 1980?
1990?
A twentieth-century average?
Another reference period?
A number without its baseline is incomplete.
This is why climate predictions need to be preserved almost like financial records.
You need the number.
But you also need the accounting period.
Conditional Predictions Can Still Fail
There is a danger in taking these distinctions too far.
Imagine a statement:
“If current trends continue, glacier X will disappear by 2020.”
2020 arrives.
The glacier remains.
Someone responds:
“It was conditional, so it cannot be wrong.”
Not necessarily.
The next question is:
Did the stated condition occur?
If the relevant trend did continue approximately as assumed and the glacier nevertheless remained, the conditional prediction can still have failed.
If the condition changed substantially, however, that change belongs in the verdict.
So:
CONDITIONAL ≠ UNTESTABLE
It simply means the condition must be audited too.
Scenarios Can Also Be Audited — Differently
A scenario is not a prediction.
But scenarios are not immune from historical examination.
Suppose researchers modeled climate under a scenario assuming:
Population = X
Coal consumption = Y
CO₂ emissions = Z
Decades later, we can ask:
Did population follow X?
Did coal consumption follow Y?
Did emissions follow Z?
If not, the climate projection associated with that scenario should be interpreted in light of those differences.
This produces two separate tests:
TEST 1 — DID THE SCENARIO OCCUR?
TEST 2 — GIVEN THE FORCING THAT DID OCCUR, HOW DID THE MODEL PERFORM?
Those are much more informative questions than:
“Was the prediction right?”
Models Are Not Crystal Balls
Climate models are mathematical representations of physical processes.
They simulate interactions among the atmosphere, oceans, land, ice and other components of the climate system.
For future projections, scientists have to supply assumptions about external influences.
They cannot know in advance exactly:
how many people will live on Earth,
how much coal will be burned,
how quickly renewable energy will expand,
what climate policies governments will adopt,
how aerosol pollution will change,
or what technologies will exist decades later.
That is one reason scenarios exist.
NASA climate datasets, for example, have explicitly run models under different greenhouse-gas emissions scenarios to examine how different assumed futures produce different climatic outcomes.
The model is answering:
IF THIS, THEN WHAT?
The scenario supplies the IF.
The projection supplies the THEN.
But Sometimes Scientists Really Do Predict
We should not hide behind terminology.
Scientists also make genuine forecasts.
Seasonal climate forecasting is an obvious example.
Scientists may estimate the probability of El Niño conditions.
They may forecast seasonal temperature or precipitation patterns.
Longer-term statements can also be framed as predictions when authors attempt to identify the actual future evolution rather than merely explore conditional futures.
The IPCC glossary itself recognizes the category climate prediction.
So it would be wrong to respond to every failed climate forecast with:
“Science never makes predictions.”
It does.
The challenge is identifying what type of claim we are examining.
The Five Questions Every Prediction Audit Needs
Before calling a climate statement right or wrong, ask:
1. WHAT EXACTLY WAS CLAIMED?
Preserve the original words.
2. WHO MADE THE CLAIM?
A peer-reviewed paper?
A government assessment?
An individual scientist?
A politician?
A journalist?
3. WAS IT CONDITIONAL?
What assumptions had to occur?
4. WHAT WAS THE DEADLINE?
A prediction without an expired deadline cannot receive a final verdict.
5. WHAT WAS THE MEASUREMENT?
Temperature?
Sea-ice extent?
Glacier area?
Population abundance?
Sea level?
Crop yield?
These questions prevent most category errors before they begin.
The Deadline Rule
This may be the simplest rule in the entire project.
NO EXPIRED DEADLINE = NO FINAL VERDICT
A projection for 2100 cannot be declared failed in 2026.
It also cannot be declared confirmed.
The endpoint has not arrived.
We can compare the trajectory with observations.
We can evaluate assumptions.
We can examine whether the scenario remains plausible.
But the final deadline is still in the future.
The same standard must apply to every prediction.
Why This Matters
The distinction between prediction, projection and scenario is not semantic trivia.
It determines what evidence would count as success or failure.
For a prediction:
Did the expected outcome occur?
For a conditional projection:
Did the assumed conditions occur, and what happened under them?
For a scenario:
Did the hypothetical future resemble the future that actually developed?
Those are different tests.
Mix them together and an audit becomes meaningless.
Separate them and historical climate claims become surprisingly easy to classify.
THE VERDICT
PREDICTION
An attempt to estimate the actual future evolution of climate.
TEST: Compare the forecast with the observation when the deadline arrives.
PROJECTION
A potential future climate response conditional on assumptions such as greenhouse-gas emissions or concentrations.
TEST: Preserve the assumptions before comparing the projected outcome with observations.
SCENARIO
A coherent possible future used to explore what might happen under specified assumptions.
TEST: Do not treat it as a prediction unless the source itself presents it as the expected future.
And four rules follow:
COULD ≠ WILL
UP TO ≠ EXPECTED VALUE
CONDITIONAL ≠ UNCONDITIONAL
SCENARIO ≠ FORECAST
But there is one final rule:
A LABEL CANNOT ERASE A PREDICTION.
If an author actually said an event was expected by a specific date, the statement remains auditable simply because someone later calls the surrounding analysis a scenario.
The only reliable method is to return to the original source.
Read the exact words.
Preserve the assumptions.
Preserve the probability.
Preserve the baseline.
Preserve the geography.
Preserve the deadline.
Then — and only then — compare the statement with what actually happened.
From the Book
This article explains the methodology behind Patrick Panteix's CLIMATE: THE PREDICTIONS THAT EXPIRED — What We Were Told. What Actually Happened. Who Profited.
The book does not begin with the assumption that every historical climate claim was wrong.
Nor does it assume that every projection should be treated as a prediction.
Instead, each claim has to survive the same audit:
What was actually said?
Under what conditions?
By what deadline?
And what actually happened?
100 predictions. 100 verdicts.
SOURCES
Intergovernmental Panel on Climate Change (IPCC) — Glossary of Terms
Defines “climate prediction,” “climate projection” and “climate scenario,” including the distinction between an estimate of actual future climate evolution and a projection conditional on future emissions or forcing assumptions.
https://www.ipcc.ch/site/assets/uploads/2018/04/4.9a.html
IPCC — Climate Change 2022: Impacts, Adaptation and Vulnerability — Annex II Glossary
Defines a scenario as a plausible future based on coherent assumptions and explicitly states that scenarios are neither predictions nor forecasts.
https://www.ipcc.ch/report/ar6/wg2/chapter/annex-ii/
IPCC — Climate Change 2021: The Physical Science Basis — Chapter 1
Explains the scenario framework used in AR6 and states that scenarios are not predictions but “what-if” investigations of possible future developments and actions.
https://www.ipcc.ch/report/ar6/wg1/chapter/chapter-1/
IPCC — Special Report on the Ocean and Cryosphere in a Changing Climate — Glossary
Defines a projection as a potential future evolution conditional on assumptions and distinguishes projections from predictions.
https://www.ipcc.ch/srocc/chapter/glossary/
IPCC — Global Warming of 1.5°C — Glossary
Defines climate projections and explains their dependence on emissions, concentration and radiative-forcing scenarios that may or may not be realized.
https://www.ipcc.ch/sr15/chapter/glossary/
NASA — NASA Releases Detailed Global Climate Change Projections
Example of climate simulations run under different greenhouse-gas emissions scenarios to investigate different possible future temperature and precipitation outcomes.
https://www.nasa.gov/news-release/nasa-releases-detailed-global-climate-change-projections/
NOAA Climate.gov — Climate Models
Explains how future climate simulations use scenarios describing possible changes in population, land use, economies and atmospheric forcing.
https://www.climate.gov/maps-data/climate-data-primer/predicting-climate/climate-models
