A sky thinned by what warms the ground

Everything below about a thousand kilometres eventually comes down, and the reason is the outermost wisp of the atmosphere. Air at 400 kilometres is a trillion times thinner than at sea level, and a satellite passing through it at nearly eight kilometres a second is slowed in a way that makes it speed up as it falls. That drag is a nuisance to anyone trying to keep a satellite up. It is also the only natural clean-up process in low Earth orbit. A dead spacecraft and the fragments of a broken one are cleared by the same air, and the whole economics of a shell of dead objects depends on how fast that happens.

The air up there has been treated by orbit designers as a variable of known habit. It swells with the solar cycle and with geomagnetic storms, and it is hard to forecast to better than a factor of two on any particular day. Averaged over cycles, it had been assumed to stay about the same. It has not been staying the same. Since at least the 1970s, the air at these heights has been getting steadily thinner, and the cause is found down near the ground.

The connection seems bizarre, and it is firmly established in principle. Carbon dioxide warms the lower atmosphere and cools the upper. A pair of opposite effects from one gas comes from a single piece of radiative physics that is seldom described beyond the first half of it, and what comes of the second half is a change to the space environment that nobody designed for.

The same molecule on both sides #

A molecule of carbon dioxide can be excited into a bending vibration that radiates at a wavelength of 15 microns. In the dense lower atmosphere this radiation is absorbed again almost immediately by another molecule of the same kind. Energy can leave only from a height where the air above is thin enough to let it through, so more of the gas raises that height and slows the escape of heat from the surface: the familiar blanket.

Now go to 200 or 300 kilometres. Carbon dioxide is a tiny trace there, but atomic oxygen is abundant and moving fast, and a collision with an oxygen atom readily excites the bending mode. The photon the molecule then emits has essentially no overlying gas to stop it. It leaves. Each such collision removes a little of the kinetic energy of the gas, and it does so with no chance of the energy being returned. So in thin air carbon dioxide is purely a coolant, and adding more of it provides more ways for the gas to lose heat. Raymond Roble and Robert Dickinson showed in 1989 that a doubling of carbon dioxide would be expected to cool the thermosphere by some tens of kelvin. This was an unusually clear prediction from a model, made before anyone had measured the effect.

The same radiative behaviour on another world has been used to explain why some atmospheres keep their upper layers cool, and it matters a good deal for what a planet can hold on to. In this case it reaches into the economics of satellites.

Contraction at a fixed height #

Air in a planet’s atmosphere is stacked up in pressure balance, with a scale height set by its temperature. Cool the column and the scale height shrinks. The base of the thermosphere stays about where it was, so the whole upper structure sinks down towards it.

Satellites sit at fixed heights, not fixed pressures. As the column sinks, each height finds itself further up the profile, in thinner air. And since density falls exponentially with the number of scale heights passed, a small change of scale height produces a change in density that accumulates with altitude. The effect is therefore larger the higher one goes, and a small fall in temperature can become a large fall in density at the top of the used range.

A three per cent change in temperature thus changes the air at 800 kilometres by about thirty per cent. Density is the quantity orbital decay responds to directly, so a change that looks marginal to a meteorologist is substantial for a mission designer, in exactly the region where the debris problem is most acute. Those altitudes are where the air is thinnest and removal is slowest even now.

The same object falls later #

The decay rate of a circular orbit is proportional to density, and an object has a ballistic coefficient setting how much the air slows it. For a given object the time from any starting height is an integral over the density it passes through.

The shape of the falling curve is what makes this sensitive. An orbit decays very slowly at first and then quickly near the end, so that nearly all of its life is spent near its starting height. The lifetime is therefore controlled by the density at the start, and a change in density at that height passes almost straight through to the lifetime. A fifty-degree cooling lengthens the life of this object by about half.

A common guideline in the industry requires a satellite to come down within twenty-five years of the end of its mission, or five under newer rules. A satellite is designed to meet this at some altitude, using a density model that assumes the upper atmosphere will stay as it has been. Yet the figure shows a sizeable margin of error at altitudes near a guideline’s limit. A plan to rely on natural decay at 600 kilometres depends on the air at 600 kilometres, and if the air thins by a third over a satellite’s lifetime and the decades after, the guideline’s meaning has moved without anyone having changed the rule.

Two worlds where carbon dioxide is most of the air #

The direct test of the idea that this molecule cools thin air is to look at a planet that is nearly all carbon dioxide. Venus has a surface hot enough to melt lead, under about ninety times the Earth’s surface pressure, and its greenhouse is the largest in the Solar System. Its upper atmosphere is another matter. The temperature at the top of the Venusian thermosphere on the day side is about 300 kelvin, and on the night side it falls to about 100 or 130 kelvin, so that region has become known as the cryosphere. That is against a thousand kelvin or so at the same level above the Earth, even though Venus is nearer the Sun and receives about twice as much ultraviolet.

Mars makes the same case more modestly. Its upper atmosphere, also mostly carbon dioxide, is far cooler than the Earth’s, at some 200 to 300 kelvin in spite of having little in the lower atmosphere to warm it. Spacecraft sent there have profited, since aerobraking at Mars works within a thin, cool thermosphere, though one that shifts readily with dust storms in the air below.

So the planet with the most extreme greenhouse in the lower atmosphere has the coldest upper atmosphere of the three, and the sign of the effect is therefore not a fine detail of a computer model. A small addition of the gas to the Earth’s thermosphere moves it a tiny way along a line that those two worlds occupy at the far end. It would be wrong to read the comparison as a calibration, since the other gases, the distances and the rotation rates all differ. But it does rule out any suspicion that the sign of the effect might be an artefact.

The opposite effect on a bad day #

The trend is slow, and it should not be confused with the air’s short-term behaviour, which is large and in the other direction. A geomagnetic storm deposits energy in the thermosphere at high latitudes, heats it by hundreds of kelvin in a day, and raises density at satellite height by a factor of two or more. In February 2022, a modest storm caught a batch of 49 newly launched Starlink satellites at a low insertion altitude of about 210 kilometres. Increased drag prevented most from raising their orbits, and 38 of them came down.

That event is a useful reminder of proportion, and it has been cited as evidence that nothing in a decades-long trend of a few per cent could matter next to a storm. The two questions are different. A storm matters for a satellite’s first hours, or for its collision forecast over the next few days, and it lasts a few days. The long-term trend matters for the integrated drag over decades and for objects with no means of manoeuvring. In a calculation of a fragment’s life, the storms add up to a part of the removal rate, and the air between storms, which is what the trend alters, carries most of the rest. Nor do the two act independently, since a cooler background responds more strongly to a given heat input. How that affects forecasting is an active area of work, because a density model calibrated on one era is no longer quite right for the next.

A trend under a cycle ten times its size #

Measuring this is hard for an obvious reason. The Sun’s ultraviolet output changes by a large factor over its eleven-year cycle, and it dominates the heating of the thermosphere.

The change sought is a few per cent a decade under a swing of a factor of ten. The instrument that detected it is the catalogue of tracked objects, with decades of two-line elements for hundreds of objects. Each object’s orbit decay gives a measure of the air it passed through. By the turn of the century, analyses of orbital decay by Gary Keating and his colleagues, and then John Emmert and his, were finding that thermospheric density near 400 kilometres had been falling by several per cent per decade, more or less as Roble and Dickinson’s calculation implied, after removing the solar contribution. These measurements rest on the same ballistic coefficients that belong to the surface and carry the same uncertainties, but a coefficient error is a constant bias over an object’s life, and this is a trend.

There are complications to remove. Solar activity has its own long-term changes, and the recent solar cycles have been rather weaker than those of the middle of the last century, which would thin the air too. Geomagnetic activity has varied, as have other trace gases. Recent work has separated those contributions and found that a signature of carbon dioxide remains and is broadly as expected, with some discrepancy in size. The measured rate is also larger at solar minimum, because carbon dioxide’s cooling matters more when there is less other heating. That is one of the fingerprints that points to the cause. A similar contraction has been inferred at lower altitudes from the ionosphere’s height measured with radio soundings, and it adds weight to a real change in the column.

What a longer life costs #

The largest consequence comes through the population of debris. Fragments in low orbit are created by collisions that need no rate of launches to continue, and they are removed by drag. If drag weakens, fragments persist for longer, and each one has more time to hit something else.

That published calculation, by William Parker, Matthew Brown and Richard Linares in 2025, introduces a useful quantity, the largest number of satellites an orbital shell can hold before fragments are produced faster than drag can remove them. They found, for the emission scenarios examined over the present century, a reduction of about half to two-thirds in that number. The simple figure above gets a comparable answer because the main lever is the same, and that kind of agreement should be read with care: a cruder model with a single altitude cannot capture how one shell feeds the next, as debris decays down through lower shells. Their conclusion is less about the exact number than about the sign. Low Earth orbit’s capacity is not a fixed resource. It is set in part by the surface climate.

The relationship carries an awkward time asymmetry. Satellites launched in the coming years will fly in air that is thinner than the data sets that model decay were fitted to, and their debris, should it form, will persist into a period when the air is thinner still. Since a small number of very massive objects holds most of the future risk of fragments in a given shell, an object now at 800 kilometres with a lifetime of two centuries is effectively permanent. It will not come down until well after the climate question is settled either way.

The unexpected reach of a greenhouse gas #

The argument connects two parts of Earth science that are usually kept entirely apart. The emission of carbon dioxide from a power station has an effect, measurable and growing, on a satellite’s orbit hundreds of kilometres above it, by way of molecular vibrations and the hydrostatic stacking of thin air. The people doing orbital mechanics had not thought of the composition of the lower atmosphere as a parameter, and those doing climate work had not much cause to think about orbits.

There is also a neat reversal in what the orbits are for. A satellite’s decay has long been treated as a nuisance to be predicted, and the catalogue of tracked debris as a hazard list. Both have now become a climate instrument, among the longest continuous records of the upper atmosphere that exists. The data that revealed the trend come from objects whose only remaining use is as involuntary probes. An orbit is a measurement of whatever disturbs it, and this one measures something happening a long way below.

What the picture cannot show #

The model atmosphere is a one-dimensional column, with a fixed lower boundary and a simple temperature shape chosen to give reasonable densities at a solar-minimum-like condition. Its composition changes smoothly with height, and helium and hydrogen, which dominate above about 600 or 800 kilometres at low solar activity, are left out. That matters, since a lighter gas falls off more slowly, so the true density at 800 kilometres is less sensitive than the figures suggest and the absolute lifetimes differ. The model has no day-night difference, no latitude structure and no winds.

Each decay in these figures takes place at a constant exospheric temperature. Real objects experience the solar cycle, which dominates their decay at each maximum, and storms, so that the lifetimes above should be read as comparisons, not forecasts. Objects with a higher or lower ballistic coefficient scale accordingly. The steady-state room for fragments is a one-altitude cartoon, with its numerator fixed. It simply shows how the removal rate enters.

The temperature falls used are round numbers, not projections. Connecting a stated rise in carbon dioxide to a fall in exospheric temperature takes a full chemistry-climate model. Its answer depends on the height, the solar conditions and the concentration of atomic oxygen, which supplies the collisions and is itself changing.

Still open: how the trend depends on the Sun #

The cooling effect is expected to be larger at solar minimum than at maximum, and the comparison across different cycles is complicated by there being so few of them. The cycles now being observed include a very low one and a stronger one, with far better drag data from accelerometers on board low-altitude satellites and with many more objects in the catalogue. These allow, for the first time, a separation of the cooling trend by altitude and by solar phase, using the same instrument over the same period. If the measured rate continues to match predictions in both height and solar dependence, the thermosphere’s response can be treated as understood, and it can be built into operational density models, which at present do not include it. If it does not match, then something about the composition of the upper air is changing that the model lacks, most probably in atomic oxygen, on which both the coolant and the drag depend.