Predicting Satellite Maneuvers
A new approach to Space Situational Awareness through automated satellite maneuver prediction algorithms
When Director of National Intelligence, the Honorable James Clapper, challenged the space community to look outside its field for new approaches to Space Situational Awareness (Clapper, 2016), it raised an interesting question: Could the predictive analytics used to track and understand the behavior of assets on land, sea, and air be applied to the domain of space? Certainly the physics and phenomenology of shipping channels and vehicle tracks are different than the movement of satellites in the voids of space, but the challenges are similar. If algorithms can learn and recognize the normal or expected activities of satellites, they could also quickly identify deviations, helping avoid or perhaps even prevent mishaps and threats in the increasingly congested and contested environment of space. Automated satellite maneuver prediction algorithms can be a way to predict when and where future maneuvers will occur.
Considering the essential roles satellites play in US defense and intelligence—providing ISR, GPS, and SATCOM—analysts need to know where an object will be in the future, its intent, and what relationships it has to other space objects. Traditional space operations that rely on a ‘detect-track-characterize-catalog’ process can’t keep pace with the hundreds to thousands of additional satellites and smallsats envisioned, along with more debris and other space objects. As in other domains, Space Situational Awareness has become a big data challenge that requires a more computationally robust, automated, and predictive approach.
Borrowing from GEOINT tradecraft and building on a strong background of automated activity recognition (Levchuk and Shabarekh, 2015), the Air Force Research Laboratory sponsored this novel approach which analyzes the activities or ‘patterns of life’ the satellite is engaging in, rather than focusing on the satellite in isolation. Patterns of life are repeatable, predictable behaviors that satellites exhibit within a certain context and constraints. For example, a satellite’s station-keeping maneuvers form mostly predictable patterns as the satellite re-positions itself to account for orbital perturbations caused by non-uniform gravitational pull, radiation pressure, and atmospheric drag. Aptima developed an unsupervised machine learning algorithm to analyze these historical patterns, and accurately predict future maneuvers. Critically, from an Indications and Warnings (I&W) perspective, deviations from those expected patterns or maneuvers can be rapidly detected, flagged, and passed to space operators.
In preliminary studies, the algorithm was able to successfully predict satellite maneuvers with high confidence 8 days in advance of the actual maneuver and it was able to detect deviations from expected behavior within hours of the predicted maneuver (Shabarekh, et al 2016).
If the role of intelligence is to foretell when and where an unexpected event will occur with enough advanced notice to execute a course of action, this game-changing technology could be used to instruct a satellite to thrust away from a hazard, or provide early warning indicators to space operators, alerting them of potential threats. The predictive analytics could fill in the gaps or blind periods when maneuvers go undetected during intermittent sensor coverage, or used in data association to gain credible custody of uncorrelated targets.
While the US has historically enjoyed unchallenged dominion in space, that scenario is fast changing, which is why defense and intelligence leaders are seeking new and novel approaches to proactively and preemptively protect our satellite and space-based capabilities.
This work was sponsored by the Air Force Research Laboratory/Space Vehicles Directorate and presented at the National Space Symposium, GEOINT 2016, and the INSA Innovators Showcase, and will be presented at the Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference in September, 2016. More than halfway through the research effort, the algorithms are being validated on operational data, with a planned demonstration of results to space operators at various agencies.