7 minute read
AI, EO, and you
By Shane Keating, University of New South Wales
Artificial Intelligence (AI) is seemingly everywhere these days, disrupting industries from web search to healthcare and finance to fine arts. The space industry is no different, and a host of innovative companies are finding novel ways to exploit the AI revolution.
Broadly speaking, artificial intelligence aims to create machines that mimic human intelligence by performing tasks that normally require “thought” --- tasks like recognising a face, understanding speech, or deciding whether to bring an umbrella on your morning commute. A neural network is a machine that mimics the brain itself, taking in information (“let’s look outside”) and passing it through a network of artificial neurons or nodes. Machine learning is way of developing artificial intelligence by looking for patterns in data (“It’s cloudy; it might rain.”). And deep learning is a type of machine learning that uses many neural networks stacked on top of each other to produce an output (“better bring a brolly”).
AI research started in the 1950’s, but progress surged in the mid-2010’s with the increased availability of powerful computers, more sophisticated algorithms, and, above all, data. Over the same period, there has been an explosion in the availability of Earth Observation (EO) data --- NASA estimates that their constellation of spacecraft alone produces 2 Gb of data every 15 seconds, with only a fraction of that amount being analysed. AI is now a crucial tool for making sense of the deluge of EO data from space.
“We're using AI to get fast results from our satellites,” says Jessica Cartwright, a senior scientific programmer and remote sensing expert at Spire Global. Spire is a leading global provider of space-based data, analytics and space services, operating a constellation of more than 100 multipurpose satellites for monitoring maritime and aviation data, soil moisture and weather patterns. “AI enables us to create models by leveraging the massive amounts of data we have from these satellites,” says Cartwright.
Cartwright is leading a project at Spire that uses AI to measure sea ice from space. Sea ice is a crucial component of the climate system and improved measurements will help us better understand the impact of climate change. But measuring sea ice is challenging, given the geographic remoteness and vast scale of the Arctic and Antarctic and the costs and challenges of polar research. “If we can tell what's happening with the ice, we can understand not only how the climate is changing, but we can also understand who it might affect when it does change,” Cartwright told me.
Cartwright uses a technique called GNSS-Reflectometry, which uses reflected signals from Global Navigation Satellite Systems (such as GPS in the US and Galileo in Europe) to costeffectively collect large quantities of information about Earth’s surface. As the GNSS signal is bounced off Earth’s surface, it carries with it a signature of the surface that is measured by Spire’s satellites. Using that data to measure sea ice requires a model for how the signal is modified by the roughness and conductivity of the ice. That’s where AI comes in.
Spire partnered with our group at the UNSW Data Science Hub as well as polar researchers from the UK and Australia to develop a machine learning algorithm that could “learn” the signature of different ice types. The IceCube project was funded through the UK-Australia SpaceBridge scheme, which is managed and led by SmartSat CRC and is supported by Austrade, the UK Government and the UK and Australian Space Agencies. We hope that the new AI techniques we have developed will provide critical insights into Antarctic sea ice change to enable enhanced climate forecasting and modelling.
AI has great potential for finding patterns in the vast amounts of Earth Observation data that is becoming increasingly available. But all that data can only be transmitted back to Earth within narrow windows of time when the satellite is passing over a ground station. “It’s like you are in the outback and going down some highway and there’s very spotty cell reception”, says Taofiq Huq, founder and CEO of Spiral Blue, a space start-up based in Sydney.
Spiral Blue is using AI to solve the downlink bottleneck using an onboard computing system that gives satellites the ability to process images on the satellite itself. The satellite can then decide which images to send back to Earth, dramatically increasing the capacity of the satellite, especially over regions with few ground stations. One application of Spiral Blue’s technology is to detect illegal fishing vessels from space. Simply downloading satellite images of the ocean to find one ship is not cost-effective, so Spiral Blue developed an AI algorithm that could identify which images contained a ship, reducing the number of images that needed to be transmitted back to Earth. “We're able to do that automatically, not just on the ground, but on the on the satellite itself”, Huq told me.
A practical challenge for developing AI-enabled satellites is the need to process data at the point where the data is collected – in space. In the tech world, this type of application is called edge computing. Huq realized that the computer chips onboard satellites did not have the processing power they needed. “Running heavy processing workloads on a device that's constrained by power availability, constrained by access and constrained by communications --- it’s an edge computing problem, dialled up to 100,” says Huq. Together with co-founders James Buttenshaw and Henry Zhong, Huq developed the Space Edge One (SE-1) computer that has the processing horsepower needed to run AI algorithms but doesn’t draw too much power from the satellite and is shielded from the hostile environment of space. Their prototype, which was launched into orbit in January 2023 and is currently undergoing flight testing, has the potential to increase the capacity of EO satellites by as much as twenty times.
Improved efficiency and increased availability are helping bring down the cost of EO data, placing it within reach of more industry players. “The opportunities are immense and across so many sectors”, says Moira Smith, CTO of D-CAT, a UK and Australia-based space data analytics company that provides monitoring, reporting and verification services to businesses worldwide. “As we’re getting more good quality sensors up there, the data will become cheaper”, she told me. “We want to bring affordable opportunities as the price point comes down.”
D-CAT are using AI and advanced sensor processing technology to solve problems across a wide range of industries, from mining and energy to forestry and insurance. In Australia, D-CAT are working closely with the agriculture sector to provide farmers and agronomists with intelligence from space data to manage their crops and livestock, irrigate more efficiently, and monitor for frost damage. “Frost damage is increasingly becoming a problem here,” says Smith. “As the climate is changing and the weather is becoming more extreme, you get hotter summers, colder winters, and so parts of Australia that never used to get frost are starting to experience frost.”
Frost can be a big problem for farmers, damaging crops and significantly reducing yield. Knowing if and where crops have been damaged is vital so farmers can reduce the impact of frost. Remote sensing, including infrared imagery, can help detect frost damage early. “Farmers want to know as early as possible,” Smith told me. “Then they can make an informed decision about what they're going to do.”
For Smith, AI opens new opportunities to tackle big challenges, like climate change. “In the past, these things weren’t possible”, she told me. But with new AI processing capabilities and cheaper Earth Observation data, individual users can access insights from space. “All of that together helps us to solve humanity’s problems.”