A book called CLIMATE: THE PREDICTIONS THAT EXPIRED naturally spends a great deal of time examining predictions that failed.
Glaciers that were supposed to disappear but remained.
An Arctic that was supposed to become ice-free by dates that passed.
Islands that were supposed to be covered by the sea.
Mass famines that did not occur on the predicted scale.
Environmental-refugee numbers that were not demonstrated by their deadlines.
But there is a danger in studying failed predictions.
If you look only for failures, you will find failures.
That does not tell you whether other predictions succeeded.
So the final test has to work in the opposite direction.
WHICH CLIMATE PREDICTIONS ACTUALLY CAME TRUE?
Not approximately because the subject was climate.
Not because the underlying problem was real.
Not because something vaguely similar happened somewhere.
But because a historical forecast made a testable claim and subsequent observations substantially matched it.
When that happens, the prediction deserves credit.
The rule cannot change because the result is inconvenient.
The Audit Has to Hurt Both Sides
There is an easy way to investigate predictions:
Collect the mistakes made by people you disagree with.
There is a harder way:
Apply the same test to everyone.
The thermometer does not know who made the prediction.
A satellite does not know whether the person examining its data is an environmentalist or a climate skeptic.
Sea level does not vote.
And a forecast does not become more or less accurate because of the politics surrounding it.
That produces the fundamental rule:
THE AUDIT CUTS BOTH WAYS.
A failed prediction has to be recorded as failed.
A successful prediction has to be recorded as successful.
The manuscript deliberately uses that principle rather than treating every one of its broad 100-case audit corpus as a failure.
So what did climate science get right?
Quite a lot, in some important areas.
1. Global Warming Was Predicted
The strongest example is also the most fundamental.
Long before the warmest years of the twenty-first century, scientists were modeling the effect of increasing atmospheric carbon dioxide on global temperature.
These were not all equally accurate.
Some projected too much warming.
Some projected too little.
Some used assumptions about future greenhouse-gas concentrations that did not match what subsequently happened.
But when researchers later compared historical climate-model projections with actual observations, an important result emerged.
A 2019 study by Zeke Hausfather, Henri Drake, Tristan Abbott and Gavin Schmidt examined model projections published between 1970 and 2007.
They compared 17 historical model projections with subsequent observed global mean surface temperature changes.
Using a straightforward temperature-versus-time comparison, 10 of the 17 projections were consistent with observations.
But there was another complication.
A climate model can contain reasonable climate physics while still producing the wrong future temperature if the modelers assumed future greenhouse-gas concentrations that differ from what actually occurred.
So the researchers also adjusted the comparison for differences between projected and realized external forcing.
Under that test, 14 of the 17 projections were consistent with observations.
That is not perfection.
But it is difficult to reconcile with the claim that old climate models simply failed to predict subsequent warming.
The Difference Between Physics and Assumptions
This is where the article we published on prediction vs. projection vs. scenario becomes important.
A model projection contains at least two major components:
THE CLIMATE RESPONSE
and
THE ASSUMED FUTURE FORCING.
Imagine a model correctly estimates how temperature responds to increasing greenhouse gases.
But its scenario assumes atmospheric CO₂ will rise much faster than it actually does.
The resulting temperature projection may overshoot reality.
Does that prove the climate physics was wrong?
Not necessarily.
Hausfather and colleagues found precisely this issue in several historical projections. Some models assumed stronger future forcing than subsequently occurred; accounting for those mismatches substantially improved their agreement with observations.
That distinction is not an excuse invented after the fact.
The future emissions pathway is an input.
If the input differs from reality, the difference has to be included in the audit.
Just as we preserved conditions when evaluating failed projections, we have to preserve them when evaluating successful ones.
2. The Long-Term Temperature Direction Was Right
Individual models matter.
But there is an even simpler result.
The world warmed.
That sounds obvious in 2026.
It was considerably less obvious when some of the early projections were published.
During the 1970s, the recent observational record included a period of mid-century cooling or reduced warming, and public discussion included the possibility of future cooling.
Yet several early climate models projected substantial warming from increasing greenhouse gases.
Decades of observations followed.
The warming occurred.
The retrospective model study concluded that historical models were generally skillful at projecting subsequent global mean surface-temperature warming and found no evidence that the evaluated models systematically overestimated or underestimated warming over their projection periods.
So on one of climate science's most important long-term predictions:
DIRECTION: HIT.
That does not validate every magnitude.
It does not validate every regional impact.
It does not validate every deadline.
But the basic forecast of greenhouse-gas-driven global warming deserves credit.
3. Ocean Heat Accumulation Was Real
Surface temperature is not the entire climate system.
Most of the excess energy associated with Earth's energy imbalance enters the oceans.
That gives us another powerful observational test.
If greenhouse warming is occurring, increasing ocean heat content is one of the physical responses we expect to observe.
And we do.
NOAA reports that global upper-ocean heat content reached a record high in 2025.
More importantly, this is not the result of one unusual year.
NOAA's observational record shows a sustained upward trend in upper-ocean heat content since around 1970.
2025 was the fifth consecutive year in which ocean heat content reached its highest recorded value.
That matters because ocean heat is much less vulnerable than a single year's surface temperature to the argument:
“It was just unusual weather.”
The ocean integrates enormous amounts of energy over time.
The long-term signal is clear.
The manuscript therefore classifies long-term ocean heat accumulation among the climate outcomes that were substantially confirmed.
4. Global Sea Level Rose
Our previous sea-level article examined several predictions that overshot their deadlines or magnitudes.
But that cannot erase the underlying prediction.
Global mean sea level was expected to rise in a warming climate.
It did.
The physical mechanisms were understood:
warming seawater expands,
mountain glaciers lose mass,
Greenland loses ice,
Antarctica contributes ice mass,
and changes in terrestrial water storage also affect the ocean.
Modern observations show a clear long-term rise in global mean sea level.
So the fair verdict is not:
SEA-LEVEL PREDICTIONS FAILED.
Nor is it:
EVERY SEA-LEVEL PREDICTION WAS CORRECT.
It is:
SEA-LEVEL RISE: CONFIRMED.
SOME SPECIFIC MAGNITUDES AND DEADLINES: MISSED OR OVERSHOT.
Those statements coexist perfectly well.
The manuscript's own scoreboard therefore records mean sea-level rise as a hit on direction, rather than using that success to retroactively validate every historical coastal prediction.
5. Arctic Summer Sea Ice Declined Dramatically
The Arctic is perhaps the clearest example of why prediction audits need more than a simple right/wrong column.
There were highly publicized predictions that summer Arctic sea ice would become effectively ice-free by dates that have passed.
Those deadlines failed.
But something else happened.
Arctic summer sea ice declined enormously.
The manuscript notes that the lowest annual minimum extents in the satellite record are concentrated in the period since 2007.
That is not evidence of an ice-free Arctic.
It is evidence of a major decline.
Therefore:
ARCTIC SEA-ICE DECLINE: HIT.
SOME ICE-FREE DEADLINES: MISS.
This distinction is essential.
If we say the ice-free deadlines were correct because sea ice declined, we move the goalposts.
If we say scientists were entirely wrong about Arctic sea ice because those deadlines failed, we ignore the observations.
Neither is acceptable.
Direction and Timing Are Two Different Bets
This principle appears repeatedly.
Imagine a prediction:
X will decline and disappear by 2020.
By 2020, X has declined 70% but still exists.
How should it be scored?
The answer depends on what we are testing.
DIRECTION → CORRECT
DISAPPEARANCE → INCORRECT
DEADLINE → MISSED
That is not fence-sitting.
It is measurement.
A prediction can contain several claims.
Each claim can receive a different result.
This is why a binary scoreboard can conceal more than it reveals.
6. The Southwest Drying Risk Was Not Imaginary
Climate scientists have also projected increasing aridity risk in the American Southwest.
The twenty-first century subsequently produced an extraordinary drought episode.
Research examining the 2000–2021 southwestern North American megadrought concluded that it was the driest 22-year period since at least the year 800 in the reconstructed record used by the researchers, with anthropogenic climate change substantially increasing its severity.
That does not mean every prediction about Lake Mead, Lake Powell, Colorado River flows or regional water shortages was correct.
Those are separate claims.
But the broader expectation of increased Southwestern aridity risk received substantial observational support.
The manuscript therefore records the Southwestern aridification signal among the predictions that performed well.
Again:
REGIONAL DRYING RISK ≠ EVERY RESERVOIR DEADLINE
The broad trend can be right while a particular countdown is wrong.
7. Exxon’s Historical Climate Projections Were Surprisingly Accurate
One of the more politically charged examples is also scientifically interesting.
Scientists working for Exxon produced climate projections decades ago.
Later researchers reconstructed those projections and compared them with subsequent observations.
The important question for an evidence audit is not what anyone thinks of Exxon as a company.
It is:
How did the projections perform?
The answer is that many tracked subsequent warming reasonably well.
The manuscript therefore includes Exxon's historical global-warming projections among the cases that broadly matched later warming.
That result does not determine what one should think about Exxon's corporate behavior, public communications or climate policy.
Those are separate questions.
It tells us something narrower:
some of its scientists' climate projections performed well.
That is the only conclusion the prediction audit needs.
8. Not Every Correct Trend Is a Successful Prediction
Now we need the opposite warning.
It would be easy to inflate the success column too.
Suppose somebody said in 1990:
“Climate change will cause more extreme weather.”
Thirty years later, we find an extreme storm.
Prediction confirmed?
No.
That is too vague.
Which extreme?
Where?
By how much?
Compared with what baseline?
Over what period?
One hurricane cannot validate a century-scale projection.
One wildfire cannot.
One flood cannot.
One drought cannot.
The manuscript explicitly rejects this kind of retrospective upgrading of vague claims into successful quantitative forecasts.
The standard has to remain demanding even when the evidence seems to support the prediction.
9. Future Predictions Do Not Count Yet
There is another tempting mistake.
Suppose a 2100 temperature projection appears to be tracking observations well through 2026.
Can we call it confirmed?
No.
We can say:
TRACKING WELL SO FAR.
We cannot say:
FINAL PREDICTION CONFIRMED.
Likewise, a 2050 projection that currently looks wrong has not necessarily failed.
The deadline has not arrived.
That is why the manuscript excludes future California snowpack endpoints, for example, from final verdicts simply because their 2050 or 2100 horizons have not expired.
The clock rule applies to successes too.
10. A Successful Prediction Does Not Validate Everything Around It
Suppose a temperature model performed well.
What does that prove?
It provides evidence that the model successfully represented the relevant global temperature response over the tested period.
It does not automatically prove that:
every glacier projection was right,
every hurricane projection was right,
every extinction prediction was right,
every economic damage estimate was right,
every policy response was optimal,
or every dollar spent in response was justified.
Those are separate propositions.
The manuscript states this explicitly:
A GOOD PREDICTION CAN BE SCIENTIFICALLY INFORMATIVE WITHOUT PROVING THAT EVERY INTERVENTION WAS EFFICIENT.
That separation is essential.
Otherwise scientific accuracy becomes a political blank check.
And a Failed Prediction Does Not Invalidate the Science
The mirror image is equally wrong.
A glacier-disappearance forecast fails.
Therefore greenhouse warming is false?
No.
An ice-free-Arctic deadline fails.
Therefore Arctic sea ice has not declined?
No.
A climate-refugee number fails.
Therefore environmental displacement does not exist?
No.
The inference does not follow.
A failed prediction tells us that that prediction failed.
Nothing more should be extracted from it without additional evidence.
This is why the manuscript insists that failed forecasts do not automatically establish fraud, just as successful forecasts do not justify every action taken in their name.
What Did the Successful Predictions Have in Common?
An interesting pattern emerges.
The strongest successes tend to involve large-scale physical variables that can be measured consistently over long periods:
GLOBAL TEMPERATURE
OCEAN HEAT
GLOBAL MEAN SEA LEVEL
ARCTIC SEA-ICE TREND
These are not the same thing as predicting exactly when:
a particular glacier disappears,
a particular island becomes uninhabitable,
a species becomes extinct,
a reservoir empties,
or a particular region experiences famine.
The farther we move from global physical forcing toward local ecological, social or economic consequences, the more variables enter the system.
Adaptation matters.
Technology matters.
Policy matters.
Migration matters.
Trade matters.
Local geography matters.
Human behavior matters.
That does not make impact projections useless.
It makes them harder.
The Best Climate Predictions Were Often the Least Dramatic
There is another lesson here.
A projection such as:
GLOBAL MEAN TEMPERATURE WILL RISE AS GREENHOUSE-GAS FORCING INCREASES
does not have the emotional power of:
THIS GLACIER WILL BE GONE BY 2020.
Or:
THIS COUNTRY WILL BE UNDERWATER.
Or:
50 MILLION PEOPLE WILL BECOME REFUGEES.
The dramatic prediction gets remembered.
The successful physical projection often does not.
That creates a distorted historical memory.
Critics remember the failed countdowns.
Defenders remember the successful trends.
Both can be right about the examples they select.
The question is whether either side is willing to examine the examples selected by the other.
This Is Why Cherry-Picking Works
Imagine two books.
Book A contains only:
failed glacier deadlines,
failed extinction deadlines,
failed famine predictions,
failed sea-ice deadlines.
Its conclusion:
THE EXPERTS WERE WRONG.
Book B contains only:
successful temperature projections,
ocean warming,
sea-level rise,
Arctic sea-ice decline.
Its conclusion:
THE EXPERTS WERE RIGHT.
Both books could contain individually true examples.
Both could still produce a misleading overall impression.
Because selection determines the story.
The solution is not artificial balance.
It is a symmetrical rule.
DO NOT SEARCH FOR AN EQUAL NUMBER OF HITS AND MISSES.
SEARCH FOR THE EVIDENCE.
That principle is also explicit in the manuscript: manufacturing a symmetrical score would itself become another form of cherry-picking.
THE SCOREBOARD
So what can we responsibly say?
Global temperature
Historical model projections often performed substantially better than the caricature suggests.
VERDICT: SUBSTANTIAL SUCCESS
Ocean heat accumulation
Long-term increase observed; upper-ocean heat content reached another record in 2025.
VERDICT: CONFIRMED TREND
Global mean sea level
Long-term rise observed.
VERDICT: DIRECTION CONFIRMED
Arctic summer sea ice
Major long-term decline observed.
VERDICT: TREND CONFIRMED
Specific early ice-free-Arctic dates
Several deadlines passed while summer sea ice remained.
VERDICT: TIMING FAILED
Glacier retreat
Major retreat occurred in numerous regions.
VERDICT: TREND CONFIRMED
Specific glacier-disappearance deadlines
Several glaciers survived beyond advertised dates.
VERDICT: DEADLINES FAILED
Southwestern U.S. aridification
Major multidecadal drying occurred.
VERDICT: SUBSTANTIALLY CONFIRMED
And that is the point.
The scoreboard is not one color.
THE FINAL VERDICT
Which climate predictions came true?
Some of the most important ones.
The world warmed.
The oceans accumulated heat.
Global mean sea level rose.
Arctic summer sea ice declined substantially.
Several historical global-temperature models performed surprisingly well when tested against subsequent observations.
Those successes belong in the record.
So do the failures.
Glaciers survived disappearance deadlines.
The Arctic remained ice-covered in summer after several highly publicized dates.
Dramatic extinction, famine, refugee and inundation forecasts missed some of their advertised endpoints.
Those failures belong in the record too.
Neither set erases the other.
And that may be the most important conclusion of all fifteen articles in this series.
THE EXPERTS WERE NOT ALWAYS RIGHT.
THE EXPERTS WERE NOT ALWAYS WRONG.
THE SKEPTICS WERE NOT ALWAYS WRONG.
THE SKEPTICS WERE NOT ALWAYS RIGHT.
Reality does not care which side benefits from the result.
So neither should an audit.
THE PREDICTION GETS CREDIT FOR WHAT IT ACTUALLY PREDICTED.
NO MORE.
NO LESS.
From the Book
This article concludes the evidence-audit series based on Patrick Panteix's CLIMATE: THE PREDICTIONS THAT EXPIRED — What We Were Told. What Actually Happened. Who Profited.
The purpose of the book is not to replace one climate slogan with another.
It is to restore what historical arguments often lose:
the original source,
the original wording,
the conditions,
the deadline,
and the observation that followed.
Because an honest audit has to record the predictions that failed.
And the predictions that succeeded.
THE AUDIT CUTS BOTH WAYS.
SOURCES
Hausfather, Z.; Drake, H.F.; Abbott, T.; Schmidt, G.A. — “Evaluating the Performance of Past Climate Model Projections” (Geophysical Research Letters, 2020)
Retrospective evaluation of 17 historical climate-model projections. Ten were consistent with observations using temperature versus time; 14 of 17 were consistent when differences between projected and realized external forcing were taken into account.
https://doi.org/10.1029/2019GL085378
NOAA National Centers for Environmental Information — Assessing the Global Temperature and Precipitation Analysis in 2025
NOAA reports that 2025 was the third-warmest year in its global record and that upper-ocean heat content reached a record high for the fifth consecutive year.
https://www.ncei.noaa.gov/news/global-climate-202513
National Snow and Ice Data Center — Sea Ice Index
Satellite-based observational record of Arctic and Antarctic sea-ice extent used to evaluate the long-term decline in Arctic summer sea ice.
https://nsidc.org/data/seaice_index
NOAA Laboratory for Satellite Altimetry — Sea Level Rise
Satellite observations of global mean sea-level change.
https://www.star.nesdis.noaa.gov/socd/lsa/SeaLevelRise/
IPCC — Climate Change 2021: The Physical Science Basis
Assessment of observed global temperature change, ocean warming, cryosphere change and sea-level rise.
https://www.ipcc.ch/report/ar6/wg1/
Williams, A.P. et al. — “Rapid Intensification of the Emerging Southwestern North American Megadrought in 2020–2021” (Nature Climate Change, 2022)
Assessment of the 2000–2021 southwestern North American megadrought and the contribution of anthropogenic climate change.
https://doi.org/10.1038/s41558-022-01290-z
Supran, G.; Rahmstorf, S.; Oreskes, N. — “Assessing ExxonMobil’s Global Warming Projections” (Science, 2023)
Retrospective quantitative evaluation of historical global-warming projections documented by Exxon and ExxonMobil scientists.
https://doi.org/10.1126/science.abk0063
Berkeley Earth — Global Temperature Report
Independent observational global-temperature record used to assess long-term warming and historical claims that global warming had stopped.
https://berkeleyearth.org/global-temperature-report/
