EarthDaily Analytics is in an interesting position: they’re a self-described software company that’s aiming to launch a constellation of satellites early next year. While the company has had several key successes of late, including a Canadian Space Agency contract under the Space Technology Development Program for data analytics tools focused on cleaning up signals from Synthetic Aperture Radar satellites, their big focus next year is on their own upcoming EarthDaily Constellation.
SpaceQ spoke with Chris Rampersad, VP Engineering at EarthDaily Analytics, about the constellation and the progress they’ve made on it.
“Oven-sized” satellites that last a decade
The key thing Rampersad emphasized early and often was that the company set out to build a Earth Observation constellation that sidestepped a lot of the trends commonly seen among other Earth observation (EO) focused private enterprises.
For example, where many companies are focused on building CubeSats, or often even smaller nanosats, the satellites in the EarthDaily constellation are larger, “around the size of a large oven” according to Rampersad, with a mass of about 225 kilograms and large solar panels on their sides.
He explained that there were several reasons for the larger-sized satellites. The first is longevity. A larger satellite may or may not be more robust, but it can definitely carry larger solar panels, as well as carrying more fuel for maneuvers. This will end up being critically important for EarthDaily, as their plan is to have their constellation of ten satellites (nine active, one spare) last for seven to ten years. Ultimately, this is a business decision: Rampersad said that they “didn’t want to have to do continuous capex reinvestment every year to launch new satellites,” and that “we wanted to launch once and then not have to think about launching [until]…year seven or something like that.”
That stability and predictability will be a key part of their business model: they will “have satellites that are fixed…we can build commercial terms, we can build product analytics around them… [and customers] can rely on it being there for a long time.”
Calibration across a constellation
The other reason for their size, though, is tied to their constellation’s unique approach to imaging, which also breaks from the norm.
Almost all other satellite constellations are “tasked”; commanded from the ground (or by edge computing AI) to focus on particular parts of the Earth and take images of it. Whether the satellite is taking multispectral, hyperspectral, or synthetic aperture rdar (SAR) imagery, tasking is the common factor. Anything that happens outside of that tasking is missed entirely.
The EarthDaily Constellation is different. It’ll be imaging the entire planet, across 22 spectral bands ranging from visible to long wave thermal infrared, looking for those daily changes. In fact, their constellation and its polar orbit were carefully designed so that they could image the entire planet, every day, at the exact same time at each location: 10:30 am local time.
Rampersad said that that focus on consistent timing was “really important to our mission.” Their goal is primarily to detect changes over time anywhere and everywhere across the globe. In order to understand those changes, and for EarthDaily to analyze them with machine learning AI algorithms, every extrinsic or confounding factor (like, say, time of day) needs to be removed as much as possible. Therefore, they all do their imaging at 10:30 local time.
But because of that desire for consistency, they also need to address differences that might crop up in the imagery made by different satellites, or even by different parts of the imaging hardware. While software can work to correct that, it simply won’t be enough; they’re aiming to achieve the kinds of standards seen in scientific research by government organizations like NASA and the ESA.
That’s why the satellites are comparatively large: the satellites imaging equipment (especially including their lenses) needs to be large enough to be able to provide that kind of imagery, and to undergo the intense and delicate calibration of the optics that would be difficult-to-impossible to perform on the comparatively smaller and simpler cameras found on an EO CubeSat or nanosat.
Rampersad said that their calibration partners are using tunable lasers “that sweep across the detectors at very small increments…like one nanometer at a time” to ensure that even the tiniest of detectable daily changes won’t get lost in the noise. He said it was “a very fine-grained scientific approach” that will ensure that “once we get into orbit we will understand very precisely how our sensor behaves,” and work to achieve that scientific quality they’re seeking.
Still, they’ll need to perform in-orbit calibration as well, to account for everything from weather to orbital changes. Rampersad said that this was another reason why they chose to image at 10:30 am; because it’s the same time that imaging is done by many scientific satellites from NASA or the ESA, like NASA’s Sentinel Two. That will mean that they can work with these government agencies to enhance their satellites’ calibration, checking for differences and similarities in order for EarthDaily to correct any miscalibration in their satellites’ hardware and software, and keep it up to scientific standards.
Using imagery and AI to analyze and predict change
When all the calibration is done, what will EarthDaily be doing with their satellites that’s different and unique? Rampersad emphasized that the big difference is, again, their focus on continuous global scans that aid them in discovering all the minute changes that others miss.
Rampersad said, “that’s important to our mission, because we are effectively delivering a system that can do change detection on a global or very broad area basis.” They will, he said, “see the change with very high quality; same viewing geometries, always the same way.”
Rampersad went on to note, however, that the imagery is “not designed for visual analysis.” Instead, it is “really designed for AI-driven information change detection…and then as you get to understand those changes [you can] move into prediction.” The system “isn’t really designed for taking a high resolution view of [a] city,” he said, but is instead “designed to see changes that are happening in that city every day—but not just one city, every city across the globe.” He said that the constant stream of change data, analyzed by their AI tools, will let clients “be able to effectively understand the pulse of the Earth.”.Rampersad added that “to understand the state of the Earth…you need to understand what’s changing.”
While the value of this kind of imagery for the national security market is straightforward enough to understand, Rampersad didn’t discuss it; instead he said that the constellation and the AI models that it generates will “help us to accurately identify methane emissions, detect changes in biodiversity, understand carbon sequestration [and] identify wildfire risk.”
He also said that a major beneficiary of the constellation will be their subsidiary EarthDaily Agro, which serves a clientele of agricultural companies and agriculture-related organizations. The constellation and the AI models it is used to build will “provide a suite of agriculture services” for the subsidiary’s clients.
Software company and its hardware partners
Rampersad emphasized that, constellation aside, EarthDaily is a software company. He said that “we have a little over 50 software engineers in our Vancouver office; a lot of them are PhDs who specialize in math or atmospheric physics, some are pure software engineers,” many of them “working on the software side of how to get high-quality data.” This is critical to the calibration efforts; they need to develop the software that will create the exacting consistency between the various satellites’ imagery that the AI models will need to do their job.
Rampersad said that “we have been building and optimizing our ground software to automatically calibrate, process, and QA satellite imagery to scientific quality for the past ten years.”
A major focus is “digital twinning,” which is creating software models of “digital twin of the Earth, satellite, and sensor to model how the satellite behaves, the effects of the atmosphere and terrain, how the sensor responds to different frequencies and intensities of light, understanding optical distortion, and the satellite’s dynamics.” He said that “true scientific quality that can reliably drive AI applications requires combining high-quality satellites with a high signal-to-noise ratio, high dynamic range, and precise spectral filters with a rigorous digital twin for calibration.”
As to the question of why the company focuses so heavily on AI, Rampersad responded at length. “We are at a point in time where there are far too many pixels being generated by satellites and not enough eyes to manually analyze them,” he said, ”so AI is really the only practical tool to help us understand and interpret the massive amounts of data.” But because of the issue of false positives, which serve as “a major impediment for automation,” there’s been a need for much higher-quality AI built on higher-quality data.
He added that “we are beginning to see an explosion in research of much larger satellite-based AI models” to solve these issues. These new foundation models, “created from enormous amounts of high-quality data” like the daily change data generated by the EarthDaily Constellation, can create “higher AI model accuracy, better generalizability of solutions across different regions, [while requiring] less training data than past AI models.”
“We believe the EarthDaily Constellation is uniquely positioned to power these emerging foundation models.”
Production, integration, launch, and beyond
Since EarthDaily is a software company, however, building and launching a satellite constellation is a major challenge. EarthDaily has built relationships with a variety of partners to help them with the project. Loft Orbital will be in charge of system integration, launch, and operations, and the satellite bus will be the “Longbow” bus made by Airbus OneWeb Satellites for Loft Orbital. The imaging payload itself, as detailed in recent SpaceQ coverage, is the result of a partnership between EarthDaily and global technology company ABB. The payloads’ electronics are supplied by Montreal’s Xiphos, and INO is providing the thermal sensors.
Rampersad also called out a company called L1 as a key vendor; L1 is responsible for the pre-launch calibration using the tunable lasers, and has experience calibrating scientific payloads up to National Institute of Standards and Technology (NIST) standards.
When asked about progress, Rampersad said that it’s happening on schedule. “The system is in the flight build phase,” he said, with ABB currently integrating the first flight payload of ten. They will be delivering the payload to Loft Orbital when it’s finished, who will integrate it with the bus.
As to launch, Rampersad said that it’s still on schedule for 2025. The first launch will be Q1 and is scheduled for February with SpaceX, and the second will be early Q2. After that, that’s it; the entire constellation will be in space, maneuvering to its final polar orbit. After they’re ready to go, Rampersad said, EarthDaily will “begin our data sales to our customers.” “The other part is that we start serving our own analytics,” he said, adding that “right now it’s mainly agricultural analytics, so that data stream will be fed directly to [EarthDaily Agro] to deliver ag-based solutions to our customers.”
After launch, they’ll also plunge into their AI-training work. Rampersad said, “after we launch our constellation, we will be exploring the development of large-scale AI models” which will “help build a deeper understanding of the dynamic changes across the Earth.”
This may even eventually involve the use of Large Language Model (LLM) generative AI. Rampersad said that people outside EarthDaily are working “to build applications that take natural language to trigger machine learning (ML) models,” where “the applications are asking an LLM to call another large scale satellite AI model.” In this situation, he said, “LLMs are a helper to make the complex satellite data easier to work with.” The EarthDaily constellation wouldn’t necessarily be used to train the LLM, but it would be used to improve and refine the satellite data models that it’s relying on.
“Our belief is that since we’re focussing on data quality and volumes, this will lead to more accurate ML models.”