Earthmover

🌽 Launching today on the Earthmover Data Marketplace: agricultural field boundaries mapped at global scale by Taylor Geospatial: 𝗧𝗵𝗲 𝗙𝗶𝗲𝗹𝗱𝘀 𝗼𝗳 𝘁𝗵𝗲 𝗪𝗼𝗿𝗹𝗱! 🌾 This is Taylor Geospatial's third dataset released on the Earthmover Data Marketplace and together they signal something bigger than a collection: 𝗮 𝗰𝗼𝗺𝗽𝗹𝗲𝘁𝗲 𝗘𝗢 𝘀𝘁𝗮𝗰𝗸 𝗳𝗼𝗿 𝗴𝗹𝗼𝗯𝗮𝗹 𝗮𝗴𝗿𝗶𝗰𝘂𝗹𝘁𝘂𝗿𝗮𝗹 𝗔𝗜. 🛰️ First input: Taylor's global 𝗦𝗲𝗻𝘁𝗶𝗻𝗲𝗹-𝟮 𝗺𝗲𝗱𝗶𝗮𝗻 𝗺𝗼𝘀𝗮𝗶𝗰𝘀 — 𝗽𝗹𝗮𝗻𝘁𝗶𝗻𝗴 𝗮𝗻𝗱 𝗵𝗮𝗿𝘃𝗲𝘀𝘁 𝘀𝗲𝗮𝘀𝗼𝗻 𝗰𝗼𝗺𝗽𝗼𝘀𝗶𝘁𝗲𝘀, 𝗿𝗲𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝗴𝗹𝗼𝗯𝗮𝗹 𝗴𝗿𝗶𝗱, 𝗱𝗲𝗹𝗶𝘃𝗲𝗿𝗲𝗱 𝗮𝘀 𝗭𝗮𝗿𝗿 𝗩𝟯 — are already on the marketplace. Raw, cloud-optimized, analysis-ready imagery at planetary scale. 🧠 Layer on the embeddings. The 𝗔𝗹𝗽𝗵𝗮𝗘𝗮𝗿𝘁𝗵 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀 𝗱𝗮𝘁𝗮𝘀𝗲𝘁, also on the marketplace, provides Google DeepMind's satellite embeddings from 2017 to 2025, providing a rich, pretrained representation of the Earth's surface to build on. 🌾 𝗠𝗮𝗸𝗲 𝘁𝗵𝗲 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀. The Global Fields of The World (FTW) dataset delivers 2024–2025 field boundary predictions worldwide, in Zarr V3, GeoParquet, and PMTiles, ready to plug into your pipeline. Researchers from Arizona State University, Washington University in St. Louis, Clark University, and the Microsoft AI for Good Lab from Microsoft Research had to build a novel architecture specifically designed for global-scale field boundary inference AND produce the training dataset needed to get there. Global inference was made possible by Wherobots RasterFlow, with Taylor Geospatial and NASA Harvest ensuring the work stayed grounded in real-world user needs from the start. 😣 This is harder than it sounds because smallholder plots in Ethiopia look nothing like Iowa corn fields or Brazilian soy. 𝗚𝗹𝗼𝗯𝗮𝗹 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗺𝗲𝗮𝗻𝘀 𝘀𝗼𝗹𝘃𝗶𝗻𝗴 𝗳𝗼𝗿 𝗱𝗮𝘁𝗮 𝗱𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆, 𝗰𝗼𝗺𝗽𝘂𝘁𝗲 𝘀𝗰𝗮𝗹𝗲, 𝗮𝗻𝗱 𝗺𝗼𝗱𝗲𝗹 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 across every agricultural landscape on Earth simultaneously. The model had to learn them all. 🇪🇹🇺🇸🇧🇷 𝗙𝗼𝗿 𝗮𝗻𝘆𝗼𝗻𝗲 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹𝘀 𝗼𝗻 𝗘𝗮𝗿𝘁𝗵 𝗼𝗯𝘀𝗲𝗿𝘃𝗮𝘁𝗶𝗼𝗻 𝗱𝗮𝘁𝗮, 𝘁𝗵𝗲 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲 𝗻𝗼𝘄 𝗴𝗶𝘃𝗲𝘀 𝘆𝗼𝘂 𝘁𝗵𝗲 𝗶𝗻𝗽𝘂𝘁𝘀, 𝘁𝗵𝗲 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀, 𝗮𝗻𝗱 𝘁𝗵𝗲 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 — 𝗮𝗹𝗹 𝗶𝗻 𝗰𝗹𝗼𝘂𝗱-𝗻𝗮𝘁𝗶𝘃𝗲, 𝗭𝗮𝗿𝗿-𝗻𝗮𝘁𝗶𝘃𝗲 𝗳𝗼𝗿𝗺𝗮𝘁𝘀 𝗱𝗲𝘀𝗶𝗴𝗻𝗲𝗱 𝗳𝗼𝗿 𝘁𝗵𝗲 𝘀𝗰𝗮𝗹𝗲 𝘁𝗵𝗲𝘀𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺𝘀 𝗱𝗲𝗺𝗮𝗻𝗱. We acknowledge the hard work of as many contributors as we can fit here: especially Juan M. Lavista Ferres, Caleb Robinson, Hannah Kerner, Nathan Jacobs, Isaac Corley, Lyndon Estes, Chris Holmes, Matthias Mohr, Jed Sundwall, and Jennifer Marcus. Powered by the PRUE model (CVPR 2026), trained on the FTW benchmark (AAAI 2025), and currently available on the Earthmover Data Marketplace. Links in the comments. 🖇️

𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱 𝗮 𝗯𝗿𝗮𝗻𝗱 𝗻𝗲𝘄 𝗮𝗿𝗿𝗮𝘆 𝗳𝗼𝗿𝗺𝗮𝘁 𝗳𝗿𝗼𝗺 𝘀𝗰𝗿𝗮𝘁𝗰𝗵 — 𝗜𝗰𝗲𝗰𝗵𝘂𝗻𝗸 — in a few months — that can be swapped in without breaking the workflows a whole community depends on? That was our core challenge and unit tests alone were not enough. Deepak Cherian and Sebastian Galkin share the high-leverage techniques that let us move fast and ship confidently: ⚙️ 𝗣𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝘁𝗲𝘀𝘁𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗛𝘆𝗽𝗼𝘁𝗵𝗲𝘀𝗶𝘀 using test case generators producing arbitrary inputs and asserting that simple invariants held across all of them. 🎲 𝗦𝘁𝗮𝘁𝗲𝗳𝘂𝗹 𝘁𝗲𝘀𝘁𝗶𝗻𝗴: arbitrary sequences of valid operations with invariants checked at every step. 🔌 𝗪𝗶𝗿𝗶𝗻𝗴 𝗶𝗻𝘁𝗼 𝘂𝗽𝘀𝘁𝗿𝗲𝗮𝗺 𝗭𝗮𝗿𝗿 𝗮𝗻𝗱 𝗫𝗮𝗿𝗿𝗮𝘆 𝘁𝗲𝘀𝘁 𝘀𝘂𝗶𝘁𝗲𝘀 even when not intended for external use. 😈 (We used it anyway. It immediately caught a gnarly bug.) There's also a cautionary tale about how property testing can breed overconfidence — and what actually makes a good generator. Come read about how we’ve 𝗯𝘂𝗶𝗹𝘁 𝗿𝗶𝗴𝗼𝗿 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀 𝗼𝗳 𝗼𝘂𝗿 𝗮𝗿𝗿𝗮𝘆 𝗰𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 𝘀𝘁𝗮𝗰𝗸: Link in the comments. 🔗

🪩 New data sets from a 𝗵𝘆𝗯𝗿𝗶𝗱 𝗽𝗵𝘆𝘀𝗶𝗰𝘀 / 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹? A major new 𝗱𝗼𝘄𝗻𝘀𝗰𝗮𝗹𝗲𝗱 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝗶𝗼𝗻 𝗲𝗻𝘀𝗲𝗺𝗯𝗹𝗲 from the Columbia Climate School? 🪩 🗓️ Join us for tomorrow's webinar with Planette AI CEO Hansi Singh, PhD and Kevin Schwarzwald, PhD, hosted by Ryan Abernathey: you don't want to miss that Q&A. 💬 🔗 Registration link in the comments 🔗

🇨🇭 𝗘𝗻𝗲𝗿𝗴𝘆 𝗮𝗻𝗱 𝗰𝗼𝗺𝗺𝗼𝗱𝗶𝘁𝘆 𝘁𝗿𝗮𝗱𝗶𝗻𝗴 𝗳𝗶𝗿𝗺𝘀 attending the Commodity Trading Summit in Switzerland this week: This tall guy holding down the 𝗟𝗲𝘀 𝗕𝗲𝗿𝗹𝗶𝗻𝗴𝗼𝘁𝘀 𝘁𝗿𝗮𝗰𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗕𝗲𝗮𝘂 𝗥𝗶𝘃𝗮𝗴𝗲 𝗶𝘀 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗖𝗧𝗢 𝗗𝗿. Joe Hamman. If you spot him, ask him if your weather prediction / satellite imaging / physical world AI/ ML teams are already working with AI-ready weather data via the Earthmover platform. It's going to be a pleasant development either way: either your team is 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝗯𝗹𝘆 𝘀𝗼𝗽𝗵𝗶𝘀𝘁𝗶𝗰𝗮𝘁𝗲𝗱 𝗿e 𝘀𝗰𝗶𝗲𝗻𝘁𝗶𝗳𝗶𝗰 𝗔𝗟/𝗠𝗟 𝗺𝗼𝗱𝗲𝗹𝗶𝗻𝗴- or they're about to be! 🇨🇭

This April newsletter is PACKED with updates: upcoming events with Hansi Singh, PhD, Kevin Schwarzwald, PhD, and Tom Gowan, PhD; new Earth Observation data sets on the Earthmover Data Marketplace from Spire Weather & Climate, Sylvera, and CTrees, the release of Icechunk V2; and case studies with Kettle and Eoliann. Read on!

🔥 How does a cutting-edge wildfire insurer manage more than 100 TB of satellite, weather, and geospatial data to power AI models that simulate 10,000 wildfire scenarios a year? 🔥 Kettle, a technically sophisticated climate risk company, uses Earthmover to power their wildfire underwriting platform. Kettle's challenge was a common one: the data was complex (massive multidimensional raster grids spanning space, time, and dozens of variables), their pipelines worked, but they were painful to build and maintain. Pre-computing spatial query outputs for every possible slice of data. Hand-rolling versioning logic to avoid breaking model runs. Building elaborate safeguards just to update a dataset. As Maxime Dion put it: “Tedious.” How Kettle now benefits from the Earthmover platform:  🗂️ Array-native architecture built for multidimensional raster data 🔁 Real data versioning: writing new dataset versions without disrupting concurrent reads or downstream model runs 🔑 Open-source foundation: Zarr & Icechunk, compatible with existing Python workflows Today Earthmover sits at the heart of Kettle's data pipeline. Monthly vegetation index (EVI) updates from NASA are versioned and served automatically. Arbitrary spatial slices - any region, time window, or variable - are fetched directly without pre-computation. Their three proprietary AI models (ignition, spread, and vulnerability) run on data that is stable, queryable, and manageable. In Blake Haugen's words: “The developer experience has improved dramatically." 💌 🌐Array-native workloads demand array-native tooling. Read the full case study on the Earthmover blog. 👇

🌳 𝗔𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗺𝗲𝗻𝘁: 𝗧𝗼𝗱𝗮𝘆 𝘄𝗲 𝘄𝗲𝗹𝗰𝗼𝗺𝗲 𝗖𝗧𝗿𝗲𝗲𝘀 𝘁𝗼 𝘁𝗵𝗲 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗗𝗮𝘁𝗮 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲 🌴 CTrees 𝗶𝘀 𝗼𝗳𝗳𝗲𝗿𝗶𝗻𝗴 𝘁𝗵𝗲𝗶𝗿 𝗔𝗯𝗼𝘃𝗲𝗴𝗿𝗼𝘂𝗻𝗱 𝗕𝗶𝗼𝗺𝗮𝘀𝘀 (𝗔𝗚𝗕) 𝗱𝗮𝘁𝗮𝘀𝗲𝘁 - 𝟭𝟬𝟬-𝗺𝗲𝘁𝗲𝗿 𝗿𝗲𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻, 𝗴𝗹𝗼𝗯𝗮𝗹𝗹𝘆, 𝗮𝗻𝗻𝘂𝗮𝗹𝗹𝘆 𝗳𝗿𝗼𝗺 𝟮𝟬𝟬𝟬 𝘁𝗼 𝗽𝗿𝗲𝘀𝗲𝗻𝘁 - 𝘃𝗶𝗮 𝘁𝗵𝗲 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗗𝗮𝘁𝗮 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲.  This is exactly the kind of data set we built it for: decades of rigorous science, instantly accessible as cloud-native, analysis-ready data. 🌏 Global coverage at 100m resolution, updated annually since 2000 🛰️ Built on a multiscale ML framework fusing satellite imagery with airborne LiDAR 📊 Every pixel comes with uncertainty quantification 🔬 Grounded in 20+ years of peer-reviewed research Use cases for this AI-ready data span GHG inventory and reporting, carbon accounting, nature-based climate solutions, and forest carbon markets. CTrees’ entrance marks another milestone for the Earthmover Data Marketplace as we continue to build out the central access point for high-quality environmental data we started with Sylvera and Spire Weather & Climate.  We see more and more 𝘁𝗲𝗮𝗺𝘀 𝗶𝗻 𝗲𝗻𝗲𝗿𝗴𝘆, 𝗶𝗻𝘀𝘂𝗿𝗮𝗻𝗰𝗲, 𝗰𝗼𝗺𝗺𝗼𝗱𝗶𝘁𝗶𝗲𝘀 𝗮𝗻𝗱 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝘁𝗲𝗰𝗵 𝗺𝗼𝘃𝗶𝗻𝗴 𝗳𝗮𝘀𝘁𝗲𝗿 𝘁𝗵𝗮𝗻 𝗲𝘃𝗲𝗿 𝘄𝗶𝘁𝗵 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 because the data they subscribe to is already structured, queryable, and ready to use. 🌲You can learn more about CTrees systematic approach to mapping biomass at all scales in by checking out the recent webinar on YouTube from CEO and Chief Scientist Sassan Saatchi and Carbon Scientist Yan Yang, exploring the Aboveground Biomass data set and Google Colab tutorial demonstrating how to access, visualize, and analyze CTrees data using Arraylake and Python, or 𝗴𝗼 𝘀𝘁𝗿𝗮𝗶𝗴𝗵𝘁 𝘁𝗼 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮 𝗼𝗻 𝘁𝗵𝗲 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗗𝗮𝘁𝗮 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲 𝘁𝗼𝗱𝗮𝘆. 🌿👇

🌍 Two fresh datasets from two significant scientists: 𝗝𝗼𝗶𝗻 𝘂𝘀 𝗳𝗼𝗿 𝗮 𝘄𝗲𝗯𝗶𝗻𝗮𝗿  with Planette AI CEO Hansi Singh, PhD and Kevin Schwarzwald, PhD from the Columbia Climate School next Thursday, April 23. 🗓️ 🤖🌦️ Planette's Seasonal Forecast: A 𝗵𝘆𝗯𝗿𝗶𝗱 𝗽𝗵𝘆𝘀𝗶𝗰𝘀 / 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹 drives Planette's flagship seasonal forecast product. 📊🌡️ Climate Uncertainty Lab's Massive Ensemble: A brand-new bias-corrected, downscaled climate projection ensemble from Columbia University represents a 𝗺𝗮𝗷𝗼𝗿 𝗹𝗲𝗮𝗽 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝗶𝗻 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝗿𝗶𝘀𝗸 𝘂𝗻𝗰𝗲𝗿𝘁𝗮𝗶𝗻𝘁𝘆 at the resolution that actually matters for decision-making. 𝗕𝗼𝘁𝗵 𝗱𝗮𝘁𝗮𝘀𝗲𝘁𝘀 𝗮𝗿𝗲 𝗹𝗶𝘃𝗲 𝗼𝗻 𝘁𝗵𝗲 𝗘𝗮𝗿𝘁𝗵𝗺𝗼𝘃𝗲𝗿 𝗗𝗮𝘁𝗮 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲 — analysis-ready, cloud-native, no pipelines required. 🚀 Join Earthmover CEO Ryan Abernathey and these two dataset creators for a 45-minute deep dive into what makes these data sets special and how to start using them today. 🔗 𝗥𝗲𝗴𝗶𝘀𝘁𝗮𝘁𝗶𝗼𝗻 𝗹𝗶𝗻𝗸 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 🔗

👀 There’s a lot that goes into building a platform like Airis, and most of it remains invisible to the final user. The article offers a glimpse into some of the technical work behind it, developed together with Earthmover: read it if you’d like to discover a bit more of Airis’ technical backstage!

✨ Congratulations to Eoliann on the recent launch of Airis, the commercial climate risk platform in use at major infrastructure operators in Europe. We're proud to be playing a core role in their stack! ✨ 🗄️ Their original stack was built around GeoTIFF files and did not support necessary analytic operations across enormous, multi-dimensional raster datasets covering European geographies at 30-meter resolution, across dozens of climate scenarios and time horizons. So Giovanni Luddeni, Paolo Melissari and the team found Earthmover, validated the technical solution (Zarr + Icechunk + Flux) in a day, and continues to benefit from using the platform:  ✅ 1000x dataset growth handled in days ✅ Interactive hazard maps served via Flux (before ingestion was even complete!)  ✅ 2 additional data engineering hires avoided ✅ Commercial launch of the Airis platform in March 2026 ✅ Climate scientists pulling ERA5 data faster through Earthmover's marketplace than through internal tooling 🌍 Read the full case study on the Earthmover blog: From GeoTIFF Chaos to Cloud-Native Climate Risk: How Eoliann Built Airis on Earthmover 👇