The next century: climate

Foundations

Put the models on trial first

Any lesson about the climate future has to answer one question before it earns the right to say anything else: do these models work?

That is testable, and it has been tested. Take projections published between 1970 and 2007, wait, and compare them with what happened.

Hausfather and colleagues did exactly that for seventeen of them. Most projected the subsequent warming accurately. And the ones that appear to have missed mostly missed for a specific and instructive reason: they assumed emissions that did not occur.

This is the distinction that unlocks the whole subject. A climate projection is a conditional statement: if emissions follow this path, then temperature does this. Judge the physics by asking whether the temperature followed given what emissions actually did. When you correct for that, the physics comes out well.

So the models are not guesses. They are conditional predictions that have been checked.

The worst case got less likely

Here is the first thing that changed, and it went almost entirely unreported.

For years, a scenario called RCP8.5 was described in papers, press releases and headlines as business as usual. It is a very high emissions pathway: coal use expanding enormously through the twenty-first century, little technological change.

It was never a prediction. It was designed as a deliberately high end of the range, so that the span of scenarios would bracket what might happen.

Then something inconvenient happened to it. Coal did not expand the way it assumed. Solar and wind got cheap far faster than the scenario builders expected, which is the subject of the next lesson. Policy, imperfectly and unevenly, did something.

The result is that RCP8.5 is no longer a defensible description of business as usual, and Hausfather and Peters said so directly in Nature in 2020. Papers that used it as a baseline and reported catastrophic outcomes were reporting the consequences of an assumption, not a forecast.

That is the good news, and it deserves to be stated clearly, because doom is demotivating and inaccurate doom is worse.

The best case got impossible

Now the other end, and it moved the other way.

Human-caused warming is currently around 1.2 to 1.3 degrees above the pre-industrial baseline, on recent decadal averages. Individual years have already exceeded 1.5.

Peak warming scales approximately with cumulative emissions — the total ever released, not the annual rate. That is why the remaining allowance for a target is expressed as a budget, a stock to be spent.

The remaining budget for a roughly even chance of holding warming to 1.5 degrees is, at current global emissions of around forty billion tonnes of carbon dioxide a year, a small number of years.

Not a small number of years to start acting. A small number of years of total emissions at the present rate. That target requires the world’s entire emissions to reach approximately zero on a timescale that no plausible physical build-out achieves.

So: the extreme high end became implausible, and the aspirational low end became unreachable. The range of plausible futures narrowed from both directions, and the honest headline is neither the reassuring one nor the apocalyptic one. It is that we are heading somewhere in a middle band, and the width of that band is still ours to set.

The number that was stuck for forty years

One more narrowing, and this one is a genuine scientific achievement that never made the news.

Equilibrium climate sensitivity is how much the planet eventually warms if you double atmospheric carbon dioxide. Everything else scales off it. If it is low, we have room; if it is high, we do not.

The accepted range was about 1.5 to 4.5 degrees. It was published in 1979, and it was still 1.5 to 4.5 in the 2013 assessment. Four decades of satellites, supercomputers and field campaigns had not moved it.

The reason is that the difficulty is not measurement, it is clouds. Warming changes cloud cover, and clouds both reflect sunlight and trap heat. Whether the net cloud feedback amplifies or damps warming was the dominant uncertainty, and models disagreed.

A 2020 assessment attacked it differently. Rather than relying on models, it combined three largely independent lines of evidence — the physics of individual feedbacks, the historical temperature record, and palaeoclimate reconstructions of past warm and cold states — and asked what values of sensitivity are consistent with all three at once.

The likely range came out at roughly 2.6 to 3.9 degrees.

Notice what that does. It removes the comfortable low end, which is the more consequential half. A sensitivity of 1.5 would have meant a manageable problem under almost any emissions path. That is now inconsistent with the evidence.

India is where this stops being abstract

The global average is a convenient number that nobody experiences. What matters is what happens in a place, and for this course the place is India — which is, on several measures, among the most exposed countries in the world.

The monsoon. The Indian Ocean has warmed rapidly, and unusually fast in the western basin. Because the summer monsoon is driven by the temperature difference between a hot land mass and a cooler ocean, warming the ocean weakens the gradient that drives the circulation. This has been linked to reduced rainfall over parts of the subcontinent. And the changes are not simply less rain: India’s own national assessment projects more variable monsoons, with more of the total falling in fewer, heavier events. For agriculture, variability is often worse than a shifted mean.

The Himalaya. Ice loss across the range approximately doubled in rate between the 1975–2000 period and 2000–2016. The measurement itself is a good story: declassified Cold War spy satellite film from the 1970s, digitised and turned into three-dimensional surfaces, compared with modern stereo imagery. Two independent generations of imagery, one conclusion.

The Indus. On an index combining how much water a mountain system supplies with how much the basin below depends on it, the Indus basin ranks among the most important and most vulnerable water towers on Earth. That is a statement about hundreds of millions of people and two countries with a shared river treaty.

Heat. Not air temperature — wet-bulb temperature, the number that matters. Humans shed heat by sweating, and sweat only works if it evaporates. The wet-bulb temperature is the lowest temperature evaporation can get you to. When it approaches body temperature, a healthy person in the shade with unlimited water cannot cool down. Hot dry heat is survivable at remarkable air temperatures. Humid heat is not, at much lower ones. The Indo-Gangetic plain is precisely where hot and humid coincide.

Tipping points, carefully

Some parts of the Earth system may not respond smoothly. Ice sheets, permafrost, monsoon systems, ocean circulation and forests have all been proposed as tipping elements — components that can shift to a different state once a threshold is crossed, and stay there when the forcing is removed.

A 2022 assessment compiled the candidates and estimated thresholds, and concluded that several may lie within the range of warming now plausible.

Two cautions, and they belong together.

The threshold estimates carry wide uncertainty. Many rest on limited palaeoclimate evidence or on models that resolve the relevant processes poorly. A central estimate of 1.5 degrees with a range of 1 to 3 is a very different object from a measurement.

And “tipping point” in public discussion often means something the science does not say: a single global cliff edge beyond which everything is lost. The literature describes multiple, partly independent elements with different thresholds and different timescales, some of them centuries or millennia. That is a worse situation than smooth change and a different one from a cliff.

Measured, modelled, chosen

The promise made at the end of chapter 8 was to keep three categories apart. Here they are for this subject.

Measured. The radiative physics of carbon dioxide, in the laboratory since the nineteenth century. The atmospheric concentration. The temperature record. The ice loss. The ocean heat content and pH. That human activity is the cause. None of this is model output.

Modelled. The magnitude of warming for a given emissions path, with a sensitivity range now narrower but still real. Regional detail, especially rainfall, which is much harder than temperature. Tipping thresholds, which are the softest numbers in the field.

Chosen. Cumulative emissions. Which is to say: the largest single source of spread in the projections for 2100 is not a gap in the physics. It is that nobody has decided yet.

That is not a comforting statement and it is not meant as one. It is the difference between a forecast and a decision, and it is why the next lesson is about energy.

Deeper

How past projections were scored

Two questions have to be separated. First, did the model get the physics right, meaning the warming per unit of forcing? Second, did the scenario get the emissions right? Hausfather and colleagues scored both, using an implied transient climate response derived from each projection and comparing projected against observed warming both directly and after adjusting for the difference between assumed and realised forcing. Ten of the seventeen were consistent with observations on the direct comparison, and fourteen after adjusting for forcing. The improvement on adjustment is the informative result: the residual errors were mostly in the scenarios, not the physics.

Why three independent lines constrain sensitivity better than models

The Sherwood assessment uses a Bayesian framework in which each line of evidence contributes a likelihood. Process evidence constrains individual feedbacks — water vapour, lapse rate, surface albedo, and the difficult cloud terms — from theory and observation. The historical record constrains sensitivity through the observed warming, but weakly, because aerosol forcing is poorly known and the pattern of past warming differs from the equilibrium pattern. Palaeoclimate evidence, principally the Last Glacial Maximum and past warm periods, constrains the response to large forcings. Because their weaknesses are different, combining them narrows the posterior far more than any one does alone. The result depends on the assumed independence of the lines and on the priors, and the paper is explicit about testing both.

Why regional precipitation is harder than global temperature

Global mean temperature is close to a straightforward energy balance problem, and errors in one region partly cancel against another. Precipitation is set by circulation, which depends on gradients, on land surface properties, on aerosols, and on processes such as convection that occur below the model grid scale and must be parameterised. For the Indian monsoon specifically, models have long-standing biases in simulating the mean state, and aerosol effects over South Asia are both large and uncertain. This is why the Indian assessment expresses rainfall projections with wider confidence intervals than temperature projections, and why any confident regional rainfall number should be treated with suspicion.

What the wet-bulb limit really means

The commonly quoted survivability threshold of a 35 degree wet-bulb temperature derives from a thermodynamic argument: at that point, skin at 35 degrees cannot shed metabolic heat to the environment by any mechanism, so core temperature rises regardless of behaviour. Recent laboratory work on human subjects suggests the practical limit for uncompensable heat stress is lower than 35 for most people and most conditions, which makes the situation worse rather than better. Observed wet-bulb temperatures approaching these values have already been recorded, briefly and locally, in parts of South Asia and the Gulf.

Frontier

Whether the sensitivity range narrows further

The remaining uncertainty is dominated by cloud feedbacks, particularly in low marine clouds. High-resolution global models that resolve convection explicitly, now becoming feasible, are the main hope for reducing it, along with satellite observations of cloud responses to interannual variability.

Tipping element thresholds

Improving these estimates requires better palaeoclimate constraints and models that resolve ice sheet dynamics and ocean overturning properly. The West Antarctic Ice Sheet and the Atlantic overturning circulation are the two where the gap between concern and constraint is widest, and both are areas of active and contested work.

Transcript and shot list

Target runtime about 10 minutes. Built as test the models first, then narrow the range from both ends, then separate what is settled from what is chosen. Shot types: ANIM is rendered from code with Manim, AI is a generated shot from Higgsfield with the shared style reference, REAL is agency imagery with credit, CAM is the narrator.

[00:00] S01 · CAM · Narrator direct to camera.

Before I tell you anything about the climate future, I owe you an answer to one question. Do these models actually work? That is testable. So let us test them first, and everything else in this video depends on what we find.

[00:19] S02 · ANIM · Seventeen projection curves published between 1970 and 2007, drawn forward from their publication dates. Then the observed temperature record overlaid.

Take projections published between 1970 and 2007. Wait. Compare them with what happened. Hausfather and colleagues did exactly that for seventeen of them.

[00:30] S03 · ANIM · Most curves matching. Then the mismatched ones isolated, and their assumed emissions compared against actual emissions — the gap resolving the mismatch.

Most projected the warming accurately. And the ones that look like misses mostly missed for a specific and very instructive reason: they assumed emissions that did not happen. That distinction unlocks the entire subject. A climate projection is a conditional statement. If emissions follow this path, then temperature does this. So you judge the physics by asking whether the temperature followed, given what the emissions actually did. Correct for that, and the physics comes out well. These are not guesses. They are conditional predictions that have been checked.

[01:11] S04 · ANIM · The RCP8.5 pathway drawn as the top edge of a scenario fan, labelled as designed to bracket the range. Then media headlines attaching the phrase “business as usual” to it.

Now the first thing that changed, and it went almost entirely unreported. For years a scenario called RCP8.5 got described in papers, in press releases and in headlines as business as usual. It is a very high emissions pathway. Coal expanding enormously through this century, very little technological change. It was never a prediction. It was built as a deliberately high end so the range of scenarios would bracket what might happen.

[01:44] S05 · ANIM · Coal use failing to follow the RCP8.5 trajectory; solar and wind costs falling faster than the scenario assumed; the pathway detaching from reality.

Then something inconvenient happened to it. Coal did not expand the way it assumed. Solar and wind got cheap far faster than the scenario builders expected, which is the whole of the next video. Policy, unevenly and imperfectly, did something. So RCP8.5 is no longer a defensible description of business as usual, and Hausfather and Peters said exactly that in Nature in 2020. Papers that used it as a baseline and reported catastrophic outcomes were reporting the consequences of an assumption. That is genuinely good news and it deserves saying plainly, because doom is demotivating and inaccurate doom is worse.

[02:30] S06 · ANIM · Warming at 1.2 to 1.3 degrees on decadal average, with individual years already crossing 1.5. Then cumulative emissions as a filling budget, at roughly 40 billion tonnes a year, with the 1.5-degree allowance running out in a few years’ worth.

Now the other end, which moved the other way. Human-caused warming is around 1.2 to 1.3 degrees above the pre-industrial baseline on recent decadal averages, and individual years have already gone past 1.5. Peak warming scales with cumulative emissions — the total ever released, not the rate. So the remaining allowance for any target is a budget, a stock to spend. And the remaining budget for a coin-flip chance at 1.5 degrees is, at about forty billion tonnes of carbon dioxide a year, a small number of years. Not a small number of years to start acting. A small number of years of total emissions. That target now needs global emissions at approximately zero on a timescale no plausible build-out achieves.

[03:26] S07 · ANIM · The scenario fan collapsing inward from both edges: the top made implausible, the bottom made unreachable, leaving a narrower middle band whose width is left undetermined.

So the extreme high end became implausible and the aspirational low end became unreachable. The range of plausible futures narrowed from both directions at once. The honest headline is neither the reassuring one nor the apocalyptic one. We are heading somewhere in a middle band, and the width of that band is still ours to set.

[03:52] S08 · ANIM · Climate sensitivity as a range: 1.5 to 4.5 degrees, published in 1979 and unchanged in 2013, with four decades of satellites and supercomputers passing over it. Then clouds isolated as the reason — reflecting sunlight and trapping heat, with models disagreeing on the net.

One more narrowing, and this one is a real scientific achievement that never made the news. Climate sensitivity is how much the planet eventually warms if you double carbon dioxide. Everything scales off it. The accepted range was about 1.5 to 4.5 degrees. That was published in 1979 and it was still 1.5 to 4.5 in the 2013 assessment. Four decades of satellites and supercomputers had not moved it. And the reason was not measurement. It was clouds. Warming changes cloud cover, and clouds both reflect sunlight and trap heat, and whether the net effect amplifies or damps warming was the dominant uncertainty.

[04:39] S09 · ANIM · Three independent evidence streams — feedback physics, the historical record, palaeoclimate — converging on an overlap at roughly 2.6 to 3.9 degrees. The comfortable low end shown being excluded.

A 2020 assessment attacked it differently. Instead of leaning on models, it combined three largely independent lines of evidence: the physics of individual feedbacks, the historical temperature record, and palaeoclimate reconstructions of past warm and cold states. Then asked which sensitivity values are consistent with all three at once. The likely range came out around 2.6 to 3.9 degrees. And notice what that does. It removes the comfortable low end, which is the more consequential half. A sensitivity of 1.5 would have meant a manageable problem under almost any emissions path. That is now inconsistent with the evidence.

[05:24] S10 · ANIM · The Indian Ocean warming, the land-sea gradient weakening, the monsoon circulation slackening; then rainfall becoming more variable — the same annual total delivered in fewer, heavier events.

The global average is a number nobody experiences. What matters is what happens in a place, and for this course that place is India, which is among the most exposed countries in the world on several measures. The Indian Ocean has warmed fast, especially in the west. The summer monsoon runs on the temperature difference between hot land and cooler ocean, so warming the ocean weakens the gradient that drives it, and that has been linked to reduced rainfall over parts of the subcontinent. And it is not simply less rain. India’s own national assessment projects more variable monsoons, with more of the total arriving in fewer, heavier events. For agriculture, variability is often worse than a shifted average.

[06:19] S11 · ANIM · Declassified 1970s spy satellite film digitised into three-dimensional glacier surfaces, differenced against modern stereo imagery; the loss rate doubling between 1975–2000 and 2000–2016. Then the Indus water tower index, with hundreds of millions of people downstream.

The Himalaya. Ice loss across the range roughly doubled in rate between 1975 to 2000 and 2000 to 2016. And the measurement is a good story in itself: declassified Cold War spy satellite film from the 1970s, digitised, turned into three-dimensional surfaces and compared against modern imagery. Two entirely separate generations of technology, one conclusion. And on an index combining how much water a mountain system supplies against how much the basin below depends on it, the Indus ranks among the most important and most vulnerable water towers on Earth. That is a sentence about hundreds of millions of people and two countries sharing a river treaty.

[07:09] S12 · ANIM · Wet-bulb explained: sweat evaporating, cooling failing as wet-bulb approaches body temperature. Dry heat survivable at high air temperature; humid heat lethal at much lower. The Indo-Gangetic plain highlighted where both coincide.

And heat. Not air temperature — wet-bulb temperature, which is the number that actually matters. Humans shed heat by sweating, and sweat only works if it evaporates. The wet-bulb temperature is the lowest temperature evaporation can get you down to. When it approaches body temperature, a healthy person, in the shade, with unlimited water, cannot cool down. Dry heat is survivable at remarkable air temperatures. Humid heat is not, at much lower ones. And the Indo-Gangetic plain is exactly where hot and humid coincide.

[07:48] S13 · ANIM · Tipping elements listed — ice sheets, permafrost, monsoon, ocean overturning, forests — each with a threshold estimate and a wide error bar drawn honestly. Then the popular image of one global cliff edge shown and struck out, replaced by multiple partly independent elements on different timescales.

Some parts of the system may not respond smoothly. Ice sheets, permafrost, monsoons, ocean circulation and forests have all been proposed as tipping elements: things that shift to a different state once a threshold is crossed and do not shift back when you remove the forcing. A 2022 assessment concluded several may lie within the range of warming now plausible. Two cautions, and they go together. Those threshold estimates carry very wide uncertainty — a central estimate of 1.5 with a range of 1 to 3 is a completely different object from a measurement. And tipping point in public usage usually means a single global cliff edge past which everything is lost. The literature does not say that. It describes multiple partly independent elements, with different thresholds and timescales, some of them centuries long. That is worse than smooth change, and it is not a cliff.

[08:55] S14 · ANIM · Three columns filling: MEASURED — radiative physics, concentration, temperature record, ice loss, ocean heat and pH, attribution. MODELLED — warming per path, regional rainfall, tipping thresholds. CHOSEN — cumulative emissions. The third column widening to dominate the 2100 spread. Calendar returns.

I promised at the end of chapter eight to keep three things apart, so here they are. Measured: the radiative physics of carbon dioxide, in the lab since the nineteenth century. The concentration. The temperature record. The ice loss. Ocean heat and pH. That we caused it. None of that is model output. Modelled: how much warming for a given path, with a range that is narrower now but still real. Regional detail, especially rainfall, which is much harder than temperature. Tipping thresholds, which are the softest numbers in the field. And chosen: cumulative emissions. Which means the largest single source of spread in the projections for 2100 is not a gap in the physics. It is that nobody has decided yet. That is not comforting and it is not meant to be. It is the difference between a forecast and a decision. Which is why the next video is about energy.

Shot list

Zero AI shots, thirteen code-rendered shots, one narrator shot. Every visual in this lesson is data, and the data has to be drawn accurately with its error bars — this is the lesson where invented imagery would do the most damage.