We Only Track Orbital Debris Down to Ten Centimetres

The public catalogue of Earth orbit runs to tens of thousands of objects, and it reads like a complete inventory. It is an inventory of what radar can see on repeated passes, fit an orbit to, and publish, and the size at which that becomes possible sits just above the size at which shielding stops helping. The debris that actually shapes mission design lives in the gap between those two numbers, and every warning about it is a probability estimate with a propellant bill attached.

A Catalogue Is a Sensor Product, Not a Census

Object tracking is radar and optics doing very different jobs. Radar owns low orbit because it supplies its own illumination and works through cloud, but its sensitivity collapses with distance: return power falls as the inverse fourth power of range, and for a target smaller than the wavelength, radar cross-section falls roughly with the sixth power of diameter. Halve a fragment's diameter and the return drops by roughly sixty-four times. That is why a catalogue has a floor rather than a gentle slope, and why the floor sits near ten centimetres in low Earth orbit.

Detecting an object once is not the same as cataloguing it. A usable entry needs a state vector good enough to predict where the thing will be next week, which means repeated observations across multiple passes, each refined against the last. Something seen once may as well not exist; nothing can be computed from one position. That requirement sets the practical threshold, and it is why a fresh breakup is the worst moment to ask the catalogue a question: hundreds of fragments appear at once, all uncatalogued.

Optics pick up the rest, but only under conditions nobody can schedule. A telescope sees an object when it is sunlit and the observer is in darkness, which restricts low-orbit work to twilight windows and makes optics far more useful at geostationary altitude, where objects are large, slow and lit for most of the night. Weather removes nights. Moonlight removes contrast. The constraint is geometry, and no purchase order fixes geometry.

The Gap Between Shielded and Seen

Every spacecraft carries a defence against the small stuff: a Whipple shield, a thin bumper spaced off the pressure wall so an impactor vaporises and disperses before it reaches the structure. It is mass-efficient and it works, up to roughly a centimetre. Above that, the energy density outruns what a bumper a few millimetres thick can spread, and the shield becomes a liability that adds mass without adding protection.

The arithmetic explains the design pressure. A one-centimetre aluminium sphere masses about 1.4 grams, and at a 10 km/s closing speed it carries roughly 70 kilojoules. That is not a puncture; it is an explosion at the impact point, and nothing built to fly is armoured against it. So the answer is avoidance, which requires tracking, which returns the problem to the ten-centimetre floor. Below a centimetre you shield. Above ten you dodge. Between the two you hope — and the objects in that band are estimated in the hundreds of thousands, with the millimetre population running into the millions. Those are modelled populations, not a sensor count, which is worth remembering when someone quotes them to three significant figures.

Risk from that band accumulates as flux multiplied by cross-section multiplied by time, and each term is doing work. Flux is why altitude matters: a few hundred kilometres of extra height changes the density of the environment substantially. Cross-section is why a solar array is the exposed organ, since a large flat panel presented to the ram direction sweeps an enormous volume every orbit. Time is why a satellite designed for five years and still flying at fifteen has a different risk profile than its builders modelled. And the population is not static. A breakup injects fragments into a shell that stays populated for decades, which is the mechanism Kessler described in 1978 and which nobody has yet had to test at scale.

Every Warning Is a Probability With a Price

A conjunction message does not say two objects will collide. It says two predicted states, each with an error volume, overlap in a way that produces some probability of contact. The calculation convolves the two covariances into a combined uncertainty and integrates the chance that the true positions fall inside a hard-body radius. When the covariance is wider than the predicted miss distance, the miss distance is not the interesting number and the probability is not very informative either, which is why the message carries the covariance and not just a figure.

Those error volumes grow with time, and fastest where the physics is least predictable. In low orbit the dominant force is drag, drag depends on atmospheric density, and density depends on solar activity in ways that are modelled far better than they are forecast. A prediction three days out has a tighter error ellipsoid than one ten days out, which is why screening windows are short and warnings arrive with hours rather than weeks to spare. Lengthening the window buys planning time and pays for it in false alarms, because every marginal conjunction becomes a decision.

The decision has a price. An avoidance burn costs velocity, velocity costs propellant, and propellant is mission life, so a manoeuvre is a withdrawal from a budget allocated years earlier. A station that reboosts routinely absorbs it without noticing. A small satellite with cold gas sized for station-keeping cannot, and the threshold it sets is a policy choice dressed as an engineering one. Set it low and you spend the mission dodging; set it high and you accept a small chance of losing everything.

What I Think

The tracking network is the cheapest input to the whole problem and it is funded like an afterthought. Improving it does not close the gap, because the gap is a sensor-physics problem at one end and a mass problem at the other, but it buys warning time, and warning time is the only currency that converts into options. The specific thing I would spend on is not a bigger dish. It is more covariance data published in machine-readable form, earlier, so that operators can make the trade themselves instead of receiving a verdict.

The deeper problem is that debris risk is an unpriced externality. Whoever creates a fragment cloud pays nothing for the avoidance burns it forces on everyone else, and the operators least able to manoeuvre are the ones least able to afford the tracking they depend on. I would make manoeuvrability and identification a condition of licensing rather than a virtue, because a satellite that cannot move and cannot be found is not a participant in the traffic system, it is a hazard with a radio. Until then, every conjunction warning is a subsidy paid by the careful to the careless.