Fortunately, modern analytical techniques have improved our ability to find extremely rare signals within a sea of clutter, whether that is one Higgs event in 1010 collisions with the Large Hadron Collider, or a small number of photons from an exoplanet hiding in a billion stellar background photons. If the background cannot be minimized, it has to be characterized in detail and completely; detailed knowledge of the signatures (morphological, spectroscopic, kinematic) of all known airborne events need to be incorporated to eliminate spurious detections of known phenomena. This requires an extensive study of known events with accurately calibrated instruments.
There are numerous balloons and drones in the air at any moment. Observers may report some of these conventional objects as anomalies. The DoD already has the responsibility of alert response to unexplained aircraft in U.S. airspace. NASA could be a partner in the search for aerospatial events by enabling cross-identification with anomalies in the Earth-space environment. Since NASA data are already public and offered to the world in well-curated repositories accessible programmatically, the Agency's portfolio is set up to enable cross-referencing with NASA data and contribute to this characterization.
A database that supports the characterization of background signals should include information about the launch rate of balloons (weather, scientific, commercial, hobbyist, and military—where allowed by national security considerations); number of aircraft in the sky across the United States and the globe; daily drone launch rate within U.S. airspace; as well as characteristics of the appearance and motion capabilities of these items.
There are two approaches to detecting anomalies in large datasets. If you are looking for a needle in a haystack, one approach is to have a detailed model of the properties of needles and look for anything that looks like a needle. The other approach is to have an accurate model of the properties of hay and look for anything that looks different from hay.
In the first approach, if one knows the signal to expect, a model (or simulations) can be developed to look for that signal in large datasets. While we may be able to anticipate the sorts of signals produced by physical systems that adhere to known laws of physics, we cannot comprehensively envision all possible signals that could explain UAP, or that come from new technology or new physics (were it adversarial, extraterrestrial, or a naturally occurring but as-of-yet unknown phenomena).
The alternative approach to detecting anomalies requires a deep and thorough knowledge of what is normal and known, which can subsequently be separated from what is anomalous and unknown.
Machine learning has emerged as a powerful tool for the search for rare events, such as the creation of a Higgs Boson at an accelerator, the detection of rare cancer types, or the detection of fraudulent credit card charges to intrusions in cyber infrastructure. Machine learning and AI can play a role in the study of UAP, but not until the data both meet the standards described above and enable an extensive characterization of known and anomalous signals.
A recommendation about which methodologies specifically should be applied to this problem cannot be given at this time, as that selection depends on the nature of the data to be analyzed. Thus this question should be asked after (or ideally together with) the questions pertaining to UAP observing platforms and curated repositories for UAP data. Once the nature of the data is established, selecting algorithms for their analysis can be completed.
However, in the broad and lively domain of anomaly detection it is likely that methodologies for studying UAP already exist or can be adapted from analytical methods developed in other fields. Developing entirely new methodologies will likely be unnecessary and even a waste of resources, though adapting existing methods will still require some amount of dedicated effort. NASA could leverage its name, broad reach, and popularity to encourage and support an extensive review of existing methods for anomaly detection in the context of multidisciplinary conferences, workshops, and data challenges with mock datasets.
32
