How AI and ML Technologies Are Transforming Satellite Operations

Many of the latest satellites launched are incorporating artificial intelligence (AI) and machine learning (ML) algorithms into their operations. These technologies enable satellites to filter out irrelevant data from earth observations, improve weather forecasting, detect natural disasters, steer the satellites clear of space debris, manage mechanical faults, and so much more.

Though AI and ML systems are transforming satellite operations in many different ways, experts warn that deploying these technologies aboard satellites can be challenging because of limited computing resources and power. So, is it a worthwhile endeavour? We find out.

Understanding AI and ML for satellites

Before we dive into the latest AI and ML satellite projects and use cases, it’s worth understanding the difference between AI and ML. Although it’s easy to think of AI and ML as the same thing, they are distinct technologies. AI is a broader term describing a machine’s ability to mimic human traits like reasoning, problem-solving, and learning. More narrowly, ML is a type of AI concerned with teaching computers to perform complex tasks independently by feeding them large datasets.

Given these differences, it’s a fair assumption that AI and ML use cases will vary for satellite operations. Dr. Rickbir S. Bahia, space data uptake lead at the UK Space Agency, explains: “AI is all about enabling autonomy on satellites, allowing them to make real-time decisions such as avoiding collisions or optimising data transmission. Machine learning focuses on analysing the vast amounts of data collected, such as detecting patterns in environmental changes or identifying anomalies.”

Exciting AI projects in space

The biggest space agencies are using the latest AI and ML technologies to improve the efficiency and effectiveness of their satellite operations. The European Space Agency (ESA), for example, began experimenting with this technology for earth observation aboard the Ф-sat-1 satellite in 2020. This project demonstrated how AI can remove poor-quality samples from Earth Observation imagery, ensuring only fit-for-purpose images are sent to specialists on the ground.

Building on the success of this launch, ESA announced Φsat-2 in 2024. Using six different AI applications, this satellite can create maps from Earth imagery, identify and classify clouds and vessels in images, and detect wildfires and marine ecosystem anomalies. Moreover, because the satellite uses AI to compress images in space and reconstruct them when they reach Earth, data is quicker to download.

In the U.S., NASA has developed a wide range of AI use cases for satellites. These include detecting aircraft and ships from satellite imagery, predicting algae blooms, and monitoring extreme weather events like floods and wildfires. In May 2024, the American space agency appointed David Salvagnini as its first Chief Artificial Intelligence Officer – a sign that the organisation sees a bright future for AI in space.

Along with pioneering AI projects happening at global space agencies, tech giants like Google are also developing exciting AI and ML satellite applications. Working with wildfire authorities and other partners, Google’s research arm has helped to create a satellite constellation for identifying and monitoring classroom-sized wildfires within minutes.

As part of the aptly named FireSat project, the satellites will be equipped with custom infrared sensors to identify small fires more effectively than with traditional methods. Meanwhile, the AI software decides whether there is a wildfire by comparing any sizeable gaps in observation imagery with previous data while taking into account the local infrastructure and weather. The first of these satellites is set to launch in 2025.

UK-based small satellite manufacturer Surrey Satellite Technology Limited is another company applying ML technology to satellites. Its chief technology officer, Andrew Haslehurst, explains that the company’s latest Earth Observation satellites use ML algorithms to identify image-based metrics and communicate this information in almost real-time over Inter Satellite Link (ISL). He tells Orbital Today: “This is very useful in the modern era when satellites can collect thousands of square kilometers of coverage every day.”

Big benefits with AI and ML onboard

While AI and ML technologies are still undergoing rapid growth, many experts are excited about the opportunities they could unlock for satellite operations and the broader space industry over the coming years and decades.

Bahia of the UK Space Agency explains that these technologies can process large volumes of data to make real-time earth observation possible. This allows scientists and researchers to track weather, the environment, and natural disasters, he adds.

As well as huge benefits for earth observation, he says using AI and ML aboard satellites can help improve global communications, stop satellites from colliding with space debris, detect and handle hardware issues, and omit irrelevant data to conserve satellite power and bandwidth.

Bahia tells Orbital Today: “The use of AI and machine learning onboard satellites provides further exciting opportunities to develop Earth Observation technologies which will deliver better outcomes for science, our communities, and our planet.”

With AI and ML algorithms, satellites can even make autonomous decisions. Rajvardhan Oak, a computer science researcher at UC Davis and applied scientist at Microsoft, explains that this would remove “the need for constant human intervention” in satellite operations and decrease the time it takes to perform essential actions like “changing orbits, adjusting instruments and avoiding debris”.

Oak also expects AI and ML to improve the efficiency of communications between satellites and their operators based on Earth. By processing large volumes of onboard data, Oak says these technologies make it quicker for raw data to travel from satellites to recipients on the ground. He adds: “This minimises bandwidth usage and speeds up the delivery of actionable insights.”

Challenges to overcome

Despite offering a plethora of use cases for satellite operations, AI and ML technologies aren’t easy to deploy on satellites. In fact, Haslehust of Surrey Satellite Technology Limited describes it as something “not for the light-hearted” due to the many risks involved. For example, he says satellites that make in-orbit maneuvres without human intervention could potentially crash into other spacecraft.

According to Bahia of the UK Space Agency, the main challenges of using AI and ML algorithms on satellites are their “limited computing resources and power”, potential hardware faults caused by the “harsh space environment,” and “the difficulty of updating software once in orbit”.

He adds: “Managing these constraints while ensuring real-time decision-making, especially for critical tasks like collision avoidance, is a constant challenge for AI-driven satellite operations, but is a challenge that the UK space sector is well-positioned to tackle.”

Echoing similar thoughts, Oak says satellites have computational power, memory, and storage constraints. Such limitations can make deploying AI and ML algorithms infeasible on “small or resource-limited satellites” as these technologies require “significant processing capacity”.

Oak continues: “Similarly, power has to be utilised judiciously for other essential satellite operations like communication, imagery, and navigation, so constantly running models may not be practical.”

When satellites utilising AI and ML algorithms are sent into space, Oak says updating or reconfiguring them can be challenging due to poor communication capabilities. Consequently, satellite operators may struggle to issue bug fixes and other critical updates, meaning that AI and ML algorithms could potentially falter in orbit.

For many space agencies and companies, deploying AI and ML algorithms on their satellites is a great way to increase efficiencies through autonomous decision-making, improve earth observation, and unlock many other new capabilities. But clearly, these efforts will require significant expertise and funding, which may make AI and ML less attractive for smaller satellite companies and launches.