Data Transit & Collection Planning

The Earth observation industry has largely solved collection. Commercial electro-optical (EO) constellations image the planet daily, synthetic aperture radar (SAR) sensors work through cloud and darkness, and radio frequency (RF) payloads geolocate emitters from orbit. The unsolved half of the problem is data transit. A SAR satellite can image a port at midnight through solid overcast, and that image can still spend two hours in an onboard buffer waiting for a ground station to appear over the horizon. The delay is an architecture decision that was made years before the tasking request, and it restricts every collection that satellite will ever make.

Collection planners tend to treat data transit as someone else’s problem, that is, a ground segment detail that begins after the interesting work ends. But the time from photons hitting the imager to an image in an analyst’s hands is governed as much by contact geometry, relay availability, and fiber backhaul as by the sensor itself, and for time-critical missions the data transit legs are usually the limiting factor. Every collection plan is therefore also a data transit plan, whether the planner writes it down or not.

This piece details the transit problem end to end. It covers the overhead segment (the strategies for getting data off the spacecraft), the ground segment (the paths from antenna to analyst), and then details what those architectures do to time-critical collection and to multi-sensor, multi-vendor operations.

The Problem

Three physics facts define the data transit issue.

First, contact between a remote sensor and a communication system is generally scarce. A satellite in Low Earth Orbit (LEO) at roughly 500 kilometers circles the planet every 90 minutes and holds line of sight to any single ground station for about 10 minutes per pass, often less at low elevation angles. Outside those windows the spacecraft is mute. Where it can image and where it can talk are two different parts of the schedule, and for most overhead architectures, they rarely overlap when it matters.

Second, collection outruns communication. A modern SAR or hyperspectral payload generates data faster than its downlink can drain it. An X-band downlink delivers hundreds of megabits per second, or gigabit-class at the high end, against collections that run to hundreds of gigabytes per orbit. The onboard solid-state recorder fills faster than any single pass can empty it, and when it fills, the satellite must either wipe past collection or decline tasking regardless of how favorable the imaging geometry looks.

Third, delivery is a chain, and every leg adds time. For most system architectures, the path generally has five legs:

1. Collection to contact. The wait between imaging and the next usable ground station pass.
2. Downlink. The minutes of actual transmission during the pass, shared with everything else in the buffer.
3. Station to processor. Backhaul from the antenna site into a processing center or cloud region.
4. Processing. Radiometric correction, geolocation, product formation, and exploitation.
5. Dissemination. Delivery across networks, and for defense users across security domains, to the person or machine that acts.

Leg 1 is the one collection planning controls, and it is routinely the longest. The architectures below are all attempts to shorten or eliminate it.

The Overhead Segment

The space vehicle holds data the user needs, and something must carry that data toward the ground. Three strategies exist, and they form a maturity ladder from cheap and slow to expensive and immediate.

Store and Foreward

Store and forward is the default architecture of Earth observation. The satellite collects wherever the tasking sends it, persists it to the onboard recorder, holds the data until its orbit carries it over a contracted ground station, then downlinks the backlog and repeats.

Store and forward requires no relay contracts, no crosslink hardware, and no coordination beyond the operator’s own station schedule. It decouples imaging from communications completely, so the sensor collects at full capacity anywhere on Earth. For archive-building missions, broad-area mapping, and any product measured in days, it remains the correct engineering answer.

The weaknesses compound for time-critical work. Delivery latency is set by orbital geometry, not by need. A mid-latitude collection may wait 20 to 90 minutes for the next station contact, and a satellite with sparse station contracts can wait several orbits. Priority inversion appears inside the buffer, because a routine archive collection recorded before an urgent one can sit ahead of it in the downlink queue unless the operator actively reorders. And the recorder itself becomes a tasking constraint, since a full buffer converts directly into declined collections.

Imaging in Contact (aka Relays)

The second strategy removes the wait by transmitting while collecting, which requires the satellite to be in contact with something at the moment of imaging.

The simplest form is in-contact imaging against a ground station. The planner accepts only those collection opportunities where the target and a downlink terminal sit inside the same pass, so the image streams to the ground seconds after capture. The latency win is total, and the cost is geographic. Targets near well-served regions qualify constantly. Targets in the central Pacific, the polar oceans, or hostile interiors rarely qualify at all, and constraining feasibility to station-visible passes discards most of a constellation’s theoretical capacity.

The general form is the relay, which replaces the ground station with a higher satellite. A relay in Geostationary Orbit (GEO) sees nearly a full hemisphere of LEO at once, so a LEO imager in view of the relay is effectively always in contact. The National Aeronautics and Space Administration (NASA) has operated this architecture since the 1980s through the Tracking and Data Relay Satellite System (TDRSS). Europe’s European Data Relay System (EDRS) modernized it with lasers, receiving optical crosslinks from the Copernicus Sentinel satellites at 1.8 gigabits per second and dropping the data to Europe in near real time regardless of where the Sentinel was flying.

Relays buy immediacy without waiting for station geometry, and they carry their own frictions. Relay capacity is contended, so access must be scheduled in advance, which reinserts a planning problem one layer up. Terminal hardware adds mass, power, and pointing complexity to the imaging spacecraft. And a commercial EO operator using a government or third-party relay inherits that party’s priorities, availability, and pricing.

Constant Contact

Instead of one relay overhead, the constellation itself becomes the network. Optical Inter-Satellite Links (ISLs) connect satellites into a mesh, and data hops node to node until it reaches any satellite that currently sees a gateway. The imaging satellite’s own geography stops mattering. If any node in the mesh can reach the ground, every node effectively can.

An optical terminal with 10 to 20 microradians of beam divergence paints a spot only tens of meters wide at a range of 3,000 kilometers, so nearly all transmitted power arrives at the receiver, and links of 10 to 100 gigabits per second close on small-satellite power budgets. The same tight beam is also, in practice, immune to jamming and interception from outside the line of sight. The cost of that focus is Pointing, Acquisition, and Tracking (PAT). Holding a microradian-class beam on a target moving at 7.5 kilometers per second, through platform jitter and thermal flexing, while aiming at where the receiver will be when the light arrives, is precision engineering that only recently became affordable at scale.

The architecture is operational in communications and still maturing for Earth observation. Starlink runs thousands of optical crosslinks moving petabytes per day across its mesh, and the United States Space Development Agency (SDA) is fielding its Transport Layer, a proliferated LEO mesh with standardized optical terminals built to move sensor data and tasking at machine speed. Several commercial providers now sell orbital data relay as a service, so an EO operator can buy constant contact instead of building it. For collection planning, the promise is exact. Constant contact removes leg 1 entirely, and it removes the same bottleneck in the other direction, because a tasking uplink no longer waits for the next command window.

The Ground Segment

However the data comes down, it arrives at an antenna, and the antenna is the start of the terrestrial half of the problem.

Downlink Terminals & Spaceports

The classical ground segment is a network of parabolic antennas at fixed sites, increasingly called spaceports when clustered. Geography sets their value. A polar-orbiting satellite passes near the poles on every orbit, so high-latitude stations such as the Kongsberg Satellite Services (KSAT) sites at Svalbard and Troll see that satellite roughly every 90 minutes, while a single mid-latitude station may see it only a few times per day. A handful of Arctic and Antarctic sites therefore carry a disproportionate share of the world’s EO downlink.

Ground-station-as-a-service networks, such as Amazon Web Services (AWS) Ground Station and the independent antenna operators, let satellite owners rent contacts by the minute instead of building sites, and they terminate the data directly into cloud regions where processing already lives. And optical direct-to-earth downlink is arriving, with NASA’s TeraByte InfraRed Delivery (TBIRD) demonstration having pushed 200 gigabits per second from orbit. Optical downlink multiplies throughput per pass by an order of magnitude, and it introduces a new weather dependence, because clouds block lasers. An optical ground network needs site diversity for exactly the reason optical imaging needs revisit.

Direct Downlink to Consumer

The second ground strategy bypasses the network entirely. The satellite downlinks straight to a terminal at the point of use, during the same pass in which it collected, and the whole five-leg chain collapses into one.

Landsat has supported international direct-readout stations for decades, letting national agencies receive imagery of their own territory as the satellite passes overhead. Commercial operators sell regional receiving stations on the same logic. Defense users push it furthest, with transportable and even vehicle-mounted terminals that pull imagery in-theater, keeping the data off congested reachback networks and inside the tactical timeline. A collection manager supporting an operation can, under this architecture, hold a fresh SAR scene minutes after collection with no dependency on infrastructure outside the theater.

Direct downlink serves only targets near the terminal, since the satellite must see both in one pass. The terminal owner receives only what that satellite collects locally, not the global catalog. Licensing, encryption, and compatibility must be negotiated per vendor. Direct downlink is therefore a point solution for the highest-urgency geography, not a general architecture.

Opportunistic Downlink

Between the fixed network and the dedicated terminal sits opportunistic downlink. The operator contracts across multiple shared antenna networks and drains the buffer through whichever compatible antenna appears next, buying contacts the way cloud workloads buy spot compute. This shortens the wait on average, and it converts scheduling into a market problem, because shared antennas are contended across every customer of the network and the highest-value pass may already be sold.

Data landing at Svalbard still has to reach the analyst, and it travels through a small number of undersea fiber cables, then across public or private networks, then often across a security boundary. Remote stations exist where orbits favor them, not where fiber is rich, so backhaul capacity and redundancy at the station can gate the whole chain. For government missions, cross-domain transfer, the controlled movement of data from unclassified to classified networks, frequently costs more minutes than the downlink itself. A transit plan that ends at the antenna has solved only half of the ground segment.

What Data Transit Time Does to Time-Critical Collect

Time-critical collection is where these architectures stop being background engineering and start deciding mission outcomes.

For example, a maritime watch floor receives an RF tip on a dark vessel and cues a store-and-forward SAR satellite. The satellite images the position 25 minutes later, which orbital access geometry made unavoidable. The image then waits 40 minutes for the next station contact, downlinks over 6 minutes, processes in 12, and clears dissemination and cross-domain transfer in 15. Total time from tip to analyst approaches 100 minutes, and a vessel making 12 knots has moved roughly 20 nautical miles from the imaged position. The collection succeeded by every sensor metric and failed the mission, and two thirds of the failure was transit, not access.

Three planning consequences follow.

Delivery time must be a feasibility parameter. Classical feasibility asks whether the sensor can see the target within the window. Time-critical feasibility must also ask when the data can reach the user, which depends on the satellite’s buffer state, its contact schedule, relay availability, and the downstream processing path. Two satellites with identical imaging access can differ by hours in delivery, and the planner who scores them as equal has optimized the wrong number.

Tasking latency is transit latency in reverse. The order to collect rides the same scarce links as the data. A store-and-forward satellite that passed its last command uplink five minutes before an urgent tip cannot learn about the tip until the next contact, an orbit away. Constant-contact architectures shorten both directions at once, which is why they matter more for dynamic tasking than raw downlink numbers suggest.

Weather attacks the link twice. Rain fade degrades Ka-band downlinks, and cloud blocks optical ones, so the same front that spoils an electro-optical collection can also delay delivery of the radar collection that was tasked to defeat it. Transit-aware planning treats link weather and imaging weather as separate feasibility layers.

The strongest compression available today sits on the spacecraft. Onboard processing that reduces a multi-gigabyte scene to a kilobyte-scale detection changes what must transit at all, and a detection message fits through links that a scene never could. Edge processing does not replace transit architecture, and it does buy back most of the timeline for the specific questions it has been trained to answer.

What Data Transit Time Does to Multi-Vehicle Collect

Modern collection decks span optical, SAR, RF, and hyperspectral sensors owned by different vendors, and each vendor arrives with its own transit architecture. The result is a timing problem that no single provider can see.

Each vendor operates its own station contracts, relay arrangements, processing pipeline, and delivery interface, so each has a distinct and variable time-to-deliver. A tip-and-cue chain inherits the sum. The RF vendor’s tip may deliver in 15 minutes while the SAR vendor’s cue takes 90 end to end, and the chain’s value depends on the slowest link plus the tasking latency between them. Fusion has the same exposure, because an analytic that correlates a radar pass with an optical pass must wait for whichever product lands last, and confidence in the fused answer decays while it waits.

Sovereignty and security constraints then prune the options. Some customers require that their data touch only in-country ground stations, which disqualifies certain vendors or forces them onto slower paths. Encrypted delivery, licensing boundaries, and cross-domain rules differ per provider. A multi-vendor collection plan is therefore not one transit plan but several interacting ones, and the interactions are where deadlines die.

The planning answer is orchestration that treats transit as data. A transit-aware orchestrator holds each vendor’s contact schedules, buffer behavior, relay windows, processing benchmarks, and delivery history, and it scores every candidate collection by expected time-to-answer rather than time-to-collect. It chooses the SAR satellite that images four minutes later but delivers an hour sooner. It routes urgent products to direct-downlink terminals and archive products through cheap store-and-forward paths. No human planner can hold that state for dozens of satellites across five vendors, which is exactly why it belongs in software.

Why This Matters for the Mission

Intelligence, Surveillance, and Reconnaissance (ISR) outcomes are measured at the analyst’s screen, not at the sensor aperture, and the clock the mission cares about runs from need to answer. Collection planning that ignores transit optimizes a fraction of that clock and routinely loses the rest in buffers, contact gaps, and handoffs. Kestrel builds Artificial Intelligence (AI)-native software for ISR collection and mission management, and treats delivery time as a first-class feasibility parameter alongside access geometry and sensor quality. If your team tasks across vendors whose data arrives on five different timelines, we should talk.

References

• Kaushal, H., & Kaddoum, G. (2017). Optical Communication in Space, Challenges and Mitigation Techniques. IEEE Communications Surveys & Tutorials, 19(1), 57-96.
• Caplan, D. O. (2007). Laser communication transmitter and receiver design. Journal of Optical and Fiber Communications Reports, 4, 225-362.
• Akyildiz, I. F., Kak, A., & Nie, S. (2020). Inter-Satellite Link (ISL) Networks for Emerging Space Systems. IEEE InfoCom 2020 Workshops, Toronto, ON, Canada.
• Maral, G., & Bousquet, M. (2009). Satellite Communications Systems, Techniques and Technology (5th ed.). Wiley.