Polar research station operations.
A polar station generates terabytes of science in a place with exactly one channel to the rest of the world — a metered satellite link whose budget the data would exhaust many times over. This mission is about refusing to push everything through that needle: bulk lands locally, only the products fly, and the aircraft doubles as the courier the island never had.
Polar programmes and the institutes operating remote research stations — a summer field team, a winter crew, one satellite terminal.
No ground infrastructure exists at all. The only channel south is metered satellite, and one season’s survey data would exhaust its budget many times over.
A local bond at the station where bulk imagery lands at once; quota discipline on the satellite; the aircraft as courier to the instrument sites.
The station server, the aircraft’s companion computer and the instrument-site loggers. One Linux binary, one configuration file — fully offline.
The mission
A summer field season at a coastal polar station. Drones map the glacier front and the breeding colonies on a flight schedule the weather dictates; automated instrument sites — GNSS monuments, weather masts, seismometers — sit scattered across the island, logging quietly between the rare visits anyone can spare. The science team on the continent wants to steer the campaign while it is still running: which transects to refly, which colony counts look wrong, which instrument has drifted. Between them and the station stands one satellite terminal, a hard monthly budget, and weather that closes the sky for a week at a time.
What breaks
The arithmetic is brutal and permanent. A single mapping flight produces tens of gigabytes; the satellite link is metered by the megabyte and priced accordingly; and there is no second channel — no cellular, no fibre, no neighbour to borrow from. The classical answers each concede something the programme cannot afford. Buying more satellite capacity moves the ceiling without removing it, at a cost that eats the science budget. Carrying disks home at season’s end delivers everything — months too late to steer the campaign that produced it. And the instrument sites are worse off still: their readings accumulate on loggers that someone must physically visit, so a failed sensor is discovered on the visit after the data it ruined.
There is also the quieter operational failure: when every byte competes for the same thin link, the station’s ordinary traffic — logistics, medical, the crew’s contact with home — competes with the science, and whichever matters most at the moment is decided by whoever is quickest to the terminal.
The architecture on this mission
The station is the hub. It runs a node that every aircraft bonds to locally the moment it is in range — a short, fast, free hop — so a mapping flight’s full product lands on the station’s storage minutes after landing, not weeks later on a disk in a crate. The satellite terminal joins the same node as one more link, but a deliberately different kind of link: metered, budgeted, and spent only on what has been explicitly classed to travel it. Selected products — the flagged orthophoto tile, the colony count, the anomaly the glaciologist needs to see today — go south under quota; everything else stays local until the ship does the heavy lifting at season’s end.
The instrument sites become first-class members of the same network. Each runs a small node that queues its readings — for days or for months, store-carry-forward is the routine transport here, not the exception. When the survey aircraft overflies a site on its way to the glacier, the site’s queue drains into the aircraft in priority order over the brief contact, rides home, and lands at the station with the rest of the day’s take. No landing, no visit, no logger swap. Mesh addressing means the site is simply an address: whether its packets travel by relay or by courier is a routing detail, not a procedure.
Traffic classes keep the thin link honest. The station’s operational and safety traffic holds reserved floors on the satellite; science products travel by explicit priority; and nothing bulk can ever crowd the channel by accident. The spend is a decision made once, in configuration — not renegotiated at the terminal every evening.
How the season unfolds
- Fly-in. The field team lands with the aircraft and the season’s configuration. The station node is already up from winter; the aircraft and the instrument-site nodes authenticate by key and join. The satellite quota for the season is set in one file.
- A survey day. Two glacier transects and a colony pass. On return, the aircraft bonds to the station on approach and the full imagery set — tens of gigabytes — drains over the local hop while the batteries charge. Nothing has touched the satellite.
- The courier overflight. The outbound leg crosses two instrument sites; each drains its queue into the aircraft during the seconds of contact, priority first. A seismometer’s fault flag, queued since Tuesday, reaches the station the same afternoon — and the continent the same evening.
- The evening window. The science lead flags the products that should fly: a handful of tiles, the counts, the anomaly crops. They go south under quota as high-class traffic; the journal records the spend to the byte.
- The storm. Ten days of weather. Nothing flies, and for two of those days the satellite is down too. Queues hold everywhere — at the sites, on the aircraft, at the station — in priority order. When the sky reopens, they drain in that order, automatically. Nothing is lost; nothing needed a decision.
- Season’s end. The full-resolution dataset travels home with the ship, as it always would. What is different is the journal beside it: a per-link, per-day record of what the satellite actually carried, what moved locally for free, and what each site delivered by courier — the artefact that plans next season’s quota.
What each mechanism contributes
- Store-carry-forward — the routine transport, not the fallback: queues at the instrument sites, on the aircraft and at the station, tolerant of the months when nothing moves at all. Shipped.
- Quota discipline on the metered link — the satellite spent deliberately on classed products and operational floors, never accidentally on bulk. Shipped.
- Mesh addressing — every site and aircraft an authenticated, addressable node; courier delivery is routing, not procedure. Shipped.
- Traffic classes — station safety and operations traffic reserved ahead of science bulk on the one link that is always scarce. Shipped, opt-in.
- Continuous measurement and the journal — a season-long record of what moved, where, and what it cost. Shipped.
The honest boundary: Atlas cannot make the satellite wider or cheaper, and the season’s full-resolution dataset still travels home as cargo — the architecture changes what must fly, not the physics of the link. The aircraft, their endurance in the cold, the station’s power and logistics, and the science itself are the programme’s.
What a pilot should prove
- The split, from the journal: bytes moved locally for free versus bytes sent by satellite, against the same campaign run the old way.
- Courier delivery: instrument-site queues measurably drained per overflight, timestamped end to end, with no site visit.
- Product latency: a flagged science product on the continent the same day it was flown, under quota.
- Queue integrity across a real multi-day no-contact interval — everything delivered, in priority order, after reconnection.
One station, one aircraft, one month of the season. The evaluation format covers the structure.