Precise and Efficient Orbit Prediction in LEO with Machine Learning using Exogenous Variables

Aug 4, 2026

The growing population of objects in Earth orbit creates a major demand for Space Situational Awareness, where accurate orbit prediction is essential for collision avoidance and space-debris mitigation. Orbit Propagation must incorporate uncertain non-conservative effects, including atmospheric drag and gravitational perturbations, which limit both conventional SGP4-like propagators and high-cost numerical approaches. The work proposes a machine-learning orbit prediction method that forecasts spacecraft state vectors from past positions and exogenous environmental variables (e.g., atmospheric density). Using precision ephemeris from ILRS over nearly one year, time-series learning yields low positioning errors with very low computational cost, improving orbit determination speed and reliability as object counts rise.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Related Work

Precise and Efficient Orbit Prediction in LEO with Machine Learning using Exogenous Variables 1st Francisco Caldas NOVA LINCS NOVA School of Science and Technology Caparica, Portugal [f.caldas@campus.fct.unl.pt](f.caldas@campus.fct.unl.pt) 2nd Cludia Soares NOVA LINCS NOVA School of Science and Technology Caparica, Portugal [claudia.soares@fct.unl.pt](claudia.soares@fct.unl.pt) arXiv :2407 . 1 1026v2 [ cs .LG] 27 Jul 2024 Abstract—The increasing volume of space objects in Earth’s orbit presents a significant challenge for Space Situational Awareness (SSA). And in particular, accurate orbit prediction is crucial to anticipate the position and velocity of space objects, for collision avoidance and space debris mitigation. When performing Orbit Prediction (OP), it is necessary to consider the impact of non-conservative forces, such as atmospheric drag and gravitational perturbations, that contribute to uncertainty around the future position of spacecraft and space debris alike. Conventional propagator methods like the SGP4 inadequately account for these forces, while numerical propagators are able to model the forces at a high computational cost. To address these limitations, we propose an orbit prediction algorithm utilizing machine learning. This algorithm forecasts state vectors on a spacecraft using past positions and environmental variables like atmospheric density from external sources. The orbital data used in the paper is gathered from precision ephemeris data from the International Laser Ranging Service (ILRS), for the period of almost a year. We show how the use of machine learning and time-series techniques can produce low positioning errors at avery low computational cost, thus significantly improving SSA capabilities by providing faster and reliable orbit determination for an ever increasing number of space objects. Index Terms—Orbit Prediction, Propagation, Orbit Determination, Deep Learning, Forecasting I. INTRODUCTION Since the late 1950s, when the first artificial satellite was launched, the number of resident space objects (RSOs) has steadily increased. Presently, Earth’s orbit is host to an estimated one million objects larger than 1 cm, with only 35 ,000 objects exceeding 10 cm regularly tracked [1] . To add to this, an additional 10 ,000 satellites are expected tobe launched by the year 2030, as per existing estimates and licensing arrangements [2] . To protect the space environment in Low-Earth Orbit, and due to the compounding effects of space pollution that can create a chain reaction of collisions, known as Kessler Syndrome, it is indispensable to accurately track and predict space debris and satellites’ orbits. The predominant methods to predict the future position and velocity of a satellite (state vectors) are physics-based methods that can be divided into analytical and numerical approaches. Numerical methods obtain the future state of the space objects by integrating the equations of motion of the satellite or debris, and taking into consideration the conservative and nonconservative forces applied on the space object. These methods are highly precise, at the expense of being computationally costly. Common propagators include Dormand-Prince 8(7)(RKDP8) and Runge-Kutta-Nystrom 12(10) (RKN12) [3], [4] and predictor-corrector methods, namely Adams-BashforthMoulton (ABM) and Gauss-Jackson (GJ) [5] . Analytical methods leverage closed-form equations derived from simplified models of orbital dynamics, offering expedited computations. However, such approaches, exemplified by the well-known Simplified General Perturbations 4 (SGP4) method [6], exhibit limitations in long-term orbit predictions due to their reliance on overly simplified force models. Orbit Propagation is a key component in any space surveillance framework, as it allows to determine possible close approaches between space objects in the future. When two space objects are in a close approach, this is called a conjunction

FAQs

Why is accurate orbit prediction critical for Low-Earth Orbit (LEO)?

Accurate orbit prediction supports collision avoidance and space-debris mitigation within Space Situational Awareness, especially as the number of tracked objects grows and conjunction opportunities increase.

What limitations do traditional orbit propagators have?

SGP4-like analytical methods rely on simplified force models and degrade in long-term prediction accuracy, while numerical propagators can model complex forces only at a high computational cost.

How does the proposed machine-learning approach perform orbit prediction?

It forecasts spacecraft state vectors using historical positions together with environmental exogenous variables such as atmospheric density, based on time-series machine learning techniques trained on precision ephemeris data from ILRS.

Aug 4, 2026

Upload by Genevieve