With more than half the Earth covered by clouds at any given moment, Earth observation satellites often struggle to deliver clear, usable imagery. Ubotica is using AI-driven on-orbit processing to help detect and remove clouds, improve data throughput and help make EO satellites more efficient.
Read our top four takeaways from our conversation with Aubrey Dunne, co-founder and CTO of Ubotica, or listen to the full episode.
Takeaway 1: AI Helps Satellites See Through the Clouds
Clouds represent a fundamental efficiency problem for Earth observation. When satellites downlink images full of cloud-covered pixels, operators waste bandwidth and drive up data costs.
“You cannot see the ground in the way that you would expect to,” Dunne said. “If you are downlinking images that are full of pixels with clouds in them, your overall efficiency of valuable data is being reduced.”
Instead of waiting until the imagery reaches Earth, Ubotica uses AI onboard the satellite to detect and discard cloudy pixels before transmission.
“We process it directly on the satellite, detect where the cloudy pixels are, and then attempt to remove them — only downlinking the non-cloudy pixels,” Dunne said.
By filtering out clouds before transmission, satellites can focus limited downlink capacity on valuable data.
“You can increase the value of your usable data by about 20x by dynamically imaging only where there’s useful, non-cloudy data,” Dunne said.
Takeaway 2: Dynamic targeting increases efficiency of imaging
Working with NASA’s Jet Propulsion Laboratory (JPL), Ubotica demonstrated a process called dynamic targeting, in which the satellite uses AI to preview, assess and retarget imaging in real time.
“We look forward with the satellite about 500 kilometers ahead, capture an image, process that with an AI algorithm to detect clouds, and find the least cloudy portion,” Dunne said. “Then we use that as a control signal to autonomously roll and image the area with the least cloud.”
By allowing satellites to decide where to look based on the likelihood of clear skies, this approach reduces wasted imaging time and improves overall productivity.
“It’s dynamically imaging where there’s the highest value to be attained,” Dunne said.
Takeaway 3: AI Is making CubeSats smarter
AI in space faces strict constraints: limited power, small form factors and limited bandwidth. Ubotica has tailored its technology for CubeSats and small satellites, running deep-learning algorithms on hardware that consumes just a few watts.
“The satellites we target are relatively small, so power becomes an issue,” Dunne said. “We address this by using very power-efficient hardware that specifically accelerates inference in a very power-efficient way.”
Ubotica’s AI models run quickly without draining a spacecraft’s limited energy. This enables satellites to process data onboard instead of relying on ground infrastructure.
By removing unusable imagery before it’s transmitted, operators triple the amount of useful data that can be downlinked, Dunne said.
“If two-thirds of the world is cloudy and you remove that cloud on board, you can downlink three times the amount of useful data,” Dunne explained. “For a typical Earth observation CubeSat, that can save up to $150,000 per year in downlink costs.”
Takeaway 4: Smarter satellites are becoming more autonomous
As satellites grow more autonomous, managing data privacy, governance, and reliability becomes increasingly critical. Dunne said AI supervision and encryption are essential to ensuring that autonomy doesn’t come at the cost of trust.
“We encrypt both at the payload level and the communications level,” he said. “So you have a double layer of encryption.”
Dunne said Ubotica has developed a “neural network supervisor” to monitor how the AI model performs in orbit.
“It looks at how the neurons in the network are firing and checks if they fired that way during training,” he said. “If they didn’t, it means the model may be working outside its trained domain, and you should be more suspicious of that result.”
This built-in AI oversight helps ensure that the model’s decisions are verifiable and consistent, building confidence in autonomous mission operations. For more, listen to the full episode.
Explore More:
SmallSats, Big Data: The State of Earth Observation
Can Earth Observation Innovate Without the Defense Market?
Earth Observation Market Trends to More Growth and Value From Data
