With the aim of creating a safer space environment, researchers from the Risk Institute at the University of Liverpool are collaborating with UCL and Dstl to develop advanced statistical methods to track and predict the movements of objects in space.

There are currently 128.9 million artificial pieces of space debris in Earth's orbit.Source

Since the beginning of the space age in 1957, Lower Earth Orbit (LEO) has become increasingly congested due to mission activities, launches, collisions and breakup events, and its growth is showing no signs of slowing down. As early as 1978, the NASA scientist Donald Kessler first discovered the growth of a debris belt surround the Earth and suggested that the volume of debris could get so great that it could continuously collide with itself. This phenomenon is now known as Kessler syndrome, and whilst it has served as a plot line for various science fiction works, such as the 2013 film Gravity and the 2018 novel Tom Clancy: Oath of Office, a report from the National Academy of Science in 2011 stated that the current population of orbital debris was already substantial enough for this situation to arise.

In January 2019 the European Space Agency estimated that the mass of objects in Earth's orbit was over 8.5 million kgs, of which there was only 1950 active satellites, and over 128.9 million pieces of artificial debris. This debris is a collection of nonfunctional spacecraft, abandoned launch vehicle stages, mission-related debris and fragmentation debris, all of which threatens to collide with and damage, other space objects, such as, debris, satellites and spacecrafts. These collisions place communication, national security, space exploration and even astronaut's lives at risk. Therefore, a strong space surveillance network is required to ensure a safe space environment by identifying any accidental or malicious threats to assets in space or on the ground. This network should have the ability to; detect, track, catalogue, identify, and predict the anticipated orbital paths of artificial objects in space. Current space surveillance networks are regularly tracking and cataloguing only 22,300 pieces of debris, of which, 58% was created due to previous collisions.

Small pieces of debris (measuring between 1mm and 1cm) account for 99% of the total volume of debris. Detecting this is particularly challenging, since depending on it's altitude, it can be too small for ground-based sensors to detect, and it is prone to deorbiting. This limits the majority of catalogued debris in LEO to 5-10cm in size, yet, due to their high travelling speeds, small pieces of debris still have the potential to cause severe damages to unprotected surfaces. For example, whilst the manned satellite, the International Space Station (ISS) is extensively shielded, in 2016 NASA/ESA released an image of a crack in the window of an observation window, which may have been caused by a small paint flake or metal fragment, no bigger than a few thousandths of a millimetre across. This was lucky to have caused only minor damage, as objects up to 1 cm in size have the potential to disable a satellite's critical instruments or systems, objects over 1 cm can penetrate protective shields on crew modules, and objects which are larger than 10 cm could cause a satellite to shatter into pieces.

Impact chip on the ISS due to a small paint flake or metal fragment. Source

Active Debris Removal (ADR) is an innovative and active area of research that aims to guide debris into Earth's atmosphere where it can break up safely. Methods have been developed using, nets, lasers, harpoons, and sails. RemoveDEBRIS is an active satellite mission testing the efficiency of these methods on mock targets in LEO. However, these techniques create potential challenges to the policies and treaties forming international space law, which need to be addressed before ADR can become a viable option. Alternatively collision prediction analysis can be performed: if a collision is predicted with high probability and there is the opportunity to remotely alter the position of an object, a collision avoidance manoeuvre can be performed. However, adjusting the trajectories of objects in orbit is not a simple task, thus any inaccuracies in predictive methods can lead to expensive and unnecessary manoeuvres, or in expensive damages. To advance both of these methods, further improvements to the orbit prediction method are required. The ultimate goal is to accurately determine an object's position at a given time.

A REmoveDEBRIS net experiment successfully capturing a moek piece of debris. Source

This has provided the motivation for us to begin the project with research question: Given that we have detected an object at a specific location, is it possible to predict it's position at a future time steps in order to obtain additional detections of the object, and ultimately to determine it's full orbit?

We are taking an interdisciplinary approach to answer this question by combining Dynamical force models from physics with statistical Bayesian methods. The dynamical force models are based on Newton's laws and describe the motion of an object in orbit, and are used to make a prediction of an objects position at some future time step. However, the accuracy of this prediction relies on knowing the forces acting upon it, and in complex environments such as LEO accurate knowledge of these forces is near impossible, and therefore, in practice, approximations are commonly made. Additional inaccuracies and uncertainties in the models also arise due to, the inability to explicitly solve these complex equations, and statistical noise being present in the detection images. To incorporate these uncertainties into the model, statistical methods are being utilised to allow a prediction to be made probabilistically, resulting in a set of predictions, each with a corresponding probability. These predictions are made using a recursive filter, which estimates the predictions and their probabilities by using the incoming observations and the dynamical model. First, a prediction is made using the dynamical model, and then, at the future time step when the object is (or is not) identified, the prediction is updated using the received information. This updated prediction is then used to make a prediction for the next time step, and this continues recursively in a two step (prediction and update) process. Sequentially processing the observations and using these to update the predictions, results in them becoming increasingly more reliable over time, and the prediction of the objects orbit can be refined given more observations. Answering this research question is just the beginning, and will provide a strong foundation on which we can continue developing a stronger space surveillance network.