Welcome to Issue 13, where From calibrated pixels to solar forecasts
Last week SpaceX and Nvidia named Starmind AI1, and a European space-policy institute mapped the continent's orbital-compute gap. This week, nothing notable landed on the Space4AI side — so we're doubling down on AI4Space.
Earth observation spent the week hardening its middle layer: the same analysis-ready pixels now feed commercial platforms, government awareness programs, and the foundation models that sit on top of them.
The clearest example is EarthDaily. In consecutive days, its constellation won a multi-year Canadian government award to test satellite data and AI for border monitoring with federal research and police partners, and separately signed a subscription that pipes the same daily, calibrated imagery into SynMax's maritime-intelligence product. The model layer came next: Ai2 turned its open OlmoEarth foundation models into exportable embeddings, and a NASA-tied effort pushed observation into prediction — a machine-learning model that forecasts solar active regions up to 12 hours before they surface.
Specifics below.
Did you know?
We have recently launched the Bits & Orbits Insights
Check out the growing set of independent dashboards, trackers, and maps on the
AI × Space frontier:
AI4Space
EarthDaily constellation wins Canadian border AI pilot and feeds SynMax maritime intelligence
On August 12, EarthDaily Analytics announced it had been awarded a contract by Defence Research and Development Canada (DRDC) to lead the Space-based Monitoring, Alerts and Tactical Awareness Knowledge (SMATAK) project. The multi-year research and development initiative is funded through the Canadian Safety and Security Program (CSSP) and is being conducted in collaboration with DRDC and the Royal Canadian Mounted Police (RCMP). The project will evaluate how commercially available satellite data, artificial intelligence and geospatial technologies can enhance situational awareness for Canadian border monitoring.
EarthDaily will design, develop and demonstrate an end-to-end technology pilot integrating daily satellite data, AI-powered change detection, an analyst Alert Centre, ArcGIS and the Tactical Awareness Kit (TAK) ecosystem, built to flag potential changes or activities of interest through near-real-time alerts. SMATAK builds on nearly a decade of prior collaboration between EarthDaily, DRDC and the RCMP, including the Maritime Cueing of Optical Satellites (MarCOS) and Satellite Border Surveillance (SBS) projects. The CSSP is managed by DRDC's Centre for Security Science, which funds science and technology work addressing public safety and national security challenges.
The demonstration platform is designed to present satellite-derived intelligence through tools already familiar to public safety personnel, supporting faster interpretation across national, regional and field-level workflows. The project is scheduled to conclude with a final end-to-end demonstration in 2027, after which EarthDaily will provide technical recommendations and a feasibility assessment to inform the RCMP's consideration of potential future border-monitoring capabilities.
A day later, on August 13, SynMax, a multi-domain geospatial intelligence company, announced a subscription agreement with EarthDaily Analytics to integrate imagery from the EarthDaily Constellation across the SynMax platform. Under the subscription agreement, SynMax receives access to tens of millions of square kilometers of daily, analysis-ready imagery that feeds its maritime product SynMax Theia. The EarthDaily Constellation is designed to image the planet every day under consistent radiometry and geometry, producing calibrated, analysis-ready data across 22 spectral bands.
That consistency gives SynMax frequent, directly comparable coverage for broad-area change detection at global scale, and fused with SynMax's other data sources, the daily imagery sharpens what SynMax can detect and measure for maritime domain awareness. The imagery is being integrated across SynMax products on a rollout beginning this year.
Eric Anderson, CEO and co-founder of SynMax, framed the value in timing terms, saying "the edge goes to whoever sees the truth first" and that daily, calibrated imagery at scale lets customers act on proof rather than guesswork. Don Osborne, CEO of EarthDaily, said organizations responsible for critical infrastructure and maritime security need trusted, daily intelligence they can act on, and positioned EarthDaily as the AI-ready data foundation that turns daily observations into operational intelligence.
Allen Institute launches OlmoEarth embeddings export from OlmoEarth Studio
On August 12, the Allen Institute for AI (Ai2) added custom embedding exports to OlmoEarth Studio, its platform for building Earth-observation models. The feature lets users compute and export embedding vectors — compact numerical representations of EO data produced by the open-source OlmoEarth foundation models — as Cloud-Optimized GeoTIFFs for downstream analysis. The source code, model weights, and research paper are all publicly available, so the generation process can be inspected.
Embeddings are computed on demand rather than pulled from a pre-computed global archive, so outputs reflect the exact area, time range, and conditions the user specifies. Through the Studio UI or API, users select an area of interest, a time span of 1 to 12 monthly periods, spatial resolution of 10, 20, 40, or 80 meters per pixel, and imagery from Sentinel-2 L2A, Sentinel-1 RTC, or both. Three encoder variants are offered: Nano (128-dim, 1.4M params), Tiny (192-dim, 6.2M params), and Base (768-dim, 89M params). The delivered COG carries one band per embedding dimension, stored as signed 8-bit integers ranging from -127 to +127, with -128 reserved for nodata.
Ai2 documents four frozen-embedding workflows that run in seconds on standard raster data. In a few-shot segmentation test over the coastal mangrove region of Ca Mau, Vietnam, a logistic regression trained on just 60 labeled pixels — 20 each for mangrove, water, and other, sampled from ESA WorldCover 2021 — produced a wall-to-wall land-cover map at a weighted F1 of 0.84, with accuracy barely changing between 30 and 300 labels. For change detection, monthly Sentinel-2 embeddings of Butte County, California from September 2023 and September 2024 were differenced by per-pixel cosine distance, surfacing the Park Fire burn scar without labels or training. Similarity search and PCA false-color exploration round out the examples.
Custom embedding exports are available now to OlmoEarth Studio users, who can request access from Ai2. The exported GeoTIFF works with QGIS, GDAL, rasterio, or custom scripts, and Ai2 provides tutorial code plus a Colab notebook reproducing the examples. For higher performance, Studio also supports supervised fine-tuning, training a task-specific head on user labels, which typically outperforms linear probes on frozen features.
NASA COFFIES team unveils AI model to predict storm-causing active regions on the Sun
On August 14, a team with NASA's COFFIES (Consequence Of Fields and Flows in the Interior and Exterior of the Sun) reported a machine-learning model that predicts the emergence of solar active regions up to 12 hours before they appear on the Sun's surface. Active regions are the main engines behind severe space weather. Space weather forecasters number and track sunspots because they are visible manifestations of active regions, which serve as the main engines behind severe space weather events such as solar flares and coronal mass ejections.
These eruptions send waves of high-energy radiation and charged particles across space, creating storms that can threaten astronauts, disable satellites, and disrupt radio communications on Earth.
COFFIES is a NASA DRIVE Science Center, and the work drew researchers from the New Jersey Institute of Technology, Princeton University, and NASA's Ames Research Center in California. By analyzing data captured by the agency's Solar Dynamics Observatory and using NASA Ames' supercomputing resources, this new approach, published in the Journal of Geophysical Research: Machine Learning and Computation, looks at fluctuations in acoustic waves caused by sunspot regions when the regions form beneath the solar surface and begin the journey upward to emerge on the surface. The model uses a sliding-window transformer architecture built to handle very long sequences of data. Instead of looking at all activity on the solar surface at once, like earlier deep learning approaches have done, this new model moves a fixed-size "viewing window" across a long timeline of the Sun's activity to focus on recent data while remembering overall patterns.
The signal is faint by design. "We cannot directly see the magnetic structure while it is still rising through the solar interior. Instead, we must look for indirect effects — very small changes in the magnetic field and in the pattern of acoustic waves continually traveling through the Sun," said Alexander Kosovichev, a COFFIES co-investigator at NJIT. The developed technique identifies precursors associated with an emerging active region in slight changes of the Sun's acoustic power — more like a slight change in rhythm within a very noisy orchestra.
The approach departs from current operational practice. To develop current operational forecasts, the National Oceanic and Atmospheric Administration's Space Weather Prediction Center and the United States Air Force monitor active regions that are already visible on the Sun to analyze the regions' characteristics and estimate the probability of solar flares. The COFFIES model instead predicts approximate locations of emerging sunspots before they become visible. While the model is not ready for operational real-time forecasting, the team plans to validate the approach across many more known solar events to fine-tune the model.
The predictive lead time is aimed at protecting crews and hardware as NASA plans Artemis lunar missions and later crewed flights to Mars. "The COFFIES AI model is exciting to our team because it could provide us with new capabilities towards predicting potential flaring locations ahead of time," said Michelangelo Romano, deputy director of NASA's Moon to Mars Space Weather Analysis Office. Sunspot region emergence prediction capabilities, especially of the Sun's far side, could provide new information that supplements current models used by these teams.
Till next time,
Meta-beat Column of this week
Read also about the AI Pipeline that sits at the core, producing this Newsletter, including its ups and downs of this week: