How Google Is Building Geospatial AI Agents

In this episode I caught up with Shravya Shetty, who leads research and engineering for Earth AI at Google Research, to discuss how AI agents are changing geospatial analysis and predictive modelling. We begin by unpacking the difference between language models, fixed workflows, and agents that can plan, select tools, and iterate on results. Shravya explains how Earth AI brings together foundation models for remote sensing, population dynamics, and weather, and how its agentic capabilities can combine these with other geospatial datasets and users’ own data.We explore the Planetary Prediction Engine, which automates data discovery, preparation, feature engineering, model training, and evaluation from natural-language requests. Shravya also shares examples from work with WHO AFRO and partners in the Democratic Republic of the Congo to analyse Ebola risk and predict potential spread. Finally, we discuss how these systems are evaluated, why domain expertise remains essential, and how interested organisations can apply for early access.* 📺 Video of this conversation on YouTube* 🖥️ Google Earth AI* 🖥️ Geospatial Reasoning project and early access* 🖥️ Google Research: Geospatial agents* 📖 Planetary Prediction Engine paper* 📖 Geospatial Reasoning Agent technical report* 🏋️ Geospatial Reasoning benchmarksBio: Shravya Shetty is a Distinguished Engineer and Research Lead for Google’s Earth AI—driving research at the intersection of AI and geospatial understanding to build foundation models and agentic systems across satellite imagery, climate, and population dynamics to address critical global challenges. Sitting at the nexus of planetary and human well-being, she also leads research in global public health—spanning climate and health, infectious diseases, and open-source initiatives such as MedGemma and Open Health Stack to advance global health equity.Over her 18+ years leading large-scale machine learning at Google, Shravya previously spearheaded open-domain Question Answering for Google Search, a foundational precursor to modern LLM-driven search experiences. Prior to joining Google, she was a researcher at the Center of Excellence for Document Analysis and Recognition (CEDAR), where she developed machine learning algorithms for handwriting and fingerprint recognition. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I caught up with Justin Guthrie to discuss Infra-Bench CLS, a global benchmark testing seven foundation model backbones on critical infrastructure classification using satellite imagery. We explore how the dataset combines OpenStreetMap labels with Sentinel-1 and Sentinel-2 imagery, and what it takes to compare models with different input requirements using linear probing and fine-tuning.Justin explains the surprising result that DINOv3, pretrained on natural images, achieved the best overall performance in the benchmark, ahead of the Earth observation models tested. We also discuss why airports, train stations, and data centers are easier to classify, while power infrastructure remains challenging at 10-meter resolution. Finally, we cover lessons for designing fair benchmarks, experiments using just 30% of the training labels, and next steps towards higher-resolution imagery and segmentation.* 📺 Video of this conversation on YouTube* 📖 Paper* 💽 Dataset* 🖥️ Code* 🖥️ Website* 👤 Justin on LinkedIn* 📧 Academic inquiries: [email protected]* 📧 Business inquiries: [email protected]: Justin Guthrie is a data scientist who applies AI and Earth observation to understand critical infrastructure and how it holds up under stress. He is a Research Data Analyst at Northeastern University’s Defense Industrial Base Institute (DIBI), a researcher in Northeastern’s Sustainability and Data Sciences (SDS) Lab and at George Mason University, and is a co-founder and GIS Lead of Enodia, an infrastructure resilience startup in Portland, Maine. His research focuses on how infrastructure fails and how those failures spread. His work includes InfraBench CLS, a global benchmark of seven Earth observation foundation models for infrastructure mapping (under review at Big Earth Data) and an investigation of Google AlphaEarth Foundation’s precomputed embeddings in a second-author work which is in publication at IEEE TGRS. He is also a co-first author on a nationwide transmission tower detection study (in progress). With the NASA Disasters group, he uses satellite nighttime lights data to verify coordinated utility power shutoffs, testing whether satellites can track outages independently of utility reports. When not working, you can find Justin at a metal show, relaxing in a cafe, or walking around in Maine’s nature.🚀 TIMELINE00:13 – About Justin: Data science, Earth observation, and infrastructure research.01:05 – Infra-Bench CLS: Why benchmark foundation models on critical infrastructure?03:13 – Building the dataset: OpenStreetMap labels and Sentinel-1 and Sentinel-2 imagery.06:03 – Why start with classification rather than detection or segmentation?07:13 – Selecting foundation models for the benchmark.08:55 – Comparing models: Input modalities, linear probing, and fine-tuning.10:37 – Engineering and compute: Google Colab and research clusters.11:56 – Benchmark results: DINOv3’s surprising performance.15:55 – Class-level results: Airports, train stations, and data centers.17:30 – Power infrastructure: Resolution limits and ambiguous substation labels.21:49 – Next steps: Segmentation, higher resolution, and better experimental design.25:08 – Practical advice for researchers and infrastructure practitioners.27:22 – Limited labels: Experiments using 30% of the training labels.28:56 – Accessing the paper and dataset.29:25 – Following Justin’s work and getting in touch. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Petya Kangalova and Kshitij Sharma from the Humanitarian OpenStreetMap Team (HOT) to discuss how GeoAI and community-driven mapping are being combined to support disaster response and humanitarian action. We explore how HOT works with local communities to generate training data, fine-tune AI models for specific geographic regions, and rapidly identify areas requiring attention following disasters. Petya and Kshitij explain why human validation remains essential, particularly in regions where existing training data is limited, and how open-source collaboration can help address these challenges. We also dig into the new version of HOT’s fAIr platform, which aims to make GeoAI models easier to discover, deploy, and reuse through standardised pipelines and the STAC-MLM specification. Finally, we discuss their open call for community-contributed models, the technical requirements for participation, and opportunities for researchers and developers to get involved, including available funding.* 📺 Video of this conversation on YouTube* 📺 Demo video of fAIr platform on YouTube* 🗣️ Open call for GeoAI models* 🖥️ HOTOSM on Github* 🖥️ FAIR Platform * 👥 HOTOSM Working Groups* 📢 HOTOSM Community DiscussionBio: Petya Kangalova is the Senior Technology Partnership and Engagement Manager at the Humanitarian OpenStreetMap Team (HOT), with over 13 years of experience in international development, humanitarian action, and technology for social good. She leads global partnerships and initiatives advancing the use of open mapping technologies, geospatial data, and AI for disaster response and sustainable development. Previously, she spent five years at the International Aid Transparency Initiative (IATI), supporting the adoption of open data and open-source technologies across international organisations. A passionate advocate for open-source innovation and community collaboration, Petya is committed to making geospatial technologies more accessible, inclusive, and impactful.Bio: Kshitij Raj Sharma is a Senior AI Engineer at the Humanitarian OpenStreetMap Team (HOT), where he develops open-source GeoAI tools for humanitarian mapping and disaster response. He holds a master's degree in Geodata Science and AI for Earth Observation through the Copernicus Master in Digital Earth programme. His research includes VHR-Bench, a benchmark evaluating geospatial foundation models on very-high-resolution imagery. An OpenStreetMap contributor since 2016, he joined HOT in 2021 and is now a core developer and maintainer of fAIr, HOT's AI-assisted mapping platform. He is also a founding member of OSGeo Nepal and an active contributor to the open-source geospatial community. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I caught up with Ash Alex and Jimin Seo, co-founders of Hyades, to discuss BlazerOps, an agentic platform for building machine-learning models and pipelines for geospatial data. They explain how the platform aims to make spatial ML faster and more accessible, helping engineers automate complex workflows while enabling analysts to define objectives in plain English. We discuss how geospatial agents differ from conventional LLM and coding-agent workflows, why spatial data requires specialist tools, and how multiple agents can coordinate data inspection, preprocessing, model development, and evaluation.A major focus is how BlazerOps combines general-purpose language models with specialist geospatial models and curated technical knowledge. Ash and Jimin describe their in-house foundation model for spectral imagery and explain why geospatial applications may require a toolkit of specialist models rather than a single universal “supermodel.” We also discuss embeddings, fine-tuning, bespoke architectures, organisational knowledge, and how agentic development could free ML engineers and researchers to focus on more advanced work.* BlazerOps website* Video of this conversation on YouTube* Demo of BlazerOpsJimin Seo leads operations and global partnerships at Hyades, with a background in commerce and law, and passion for designing and delivering software solutions. He works closely with the technical team to translate organization-level data, privacy, IP and trustability requirements into a trustable, personal and powerful platform to enable spatial ML development for everyone.Ash Alex is the CEO of Hyades. Coming from a physics background, Ash has experience designing optical communications systems for commercial satellites at the University of Auckland, Cornell, and NASA’s Decadal Survey Program, where he was exposed to the complexities of working with geospatial data. His mission is to embed spatial insights everywhere. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Joe Redmond from the Allen Institute for AI (AI2) to discuss OlmoEarth, AI2's open geospatial foundation model for Earth observation. Joe explains how the project emerged from AI2's environmental and climate initiatives, where partners needed practical tools for analysing satellite imagery across applications such as agriculture, wildfire risk, ecosystem mapping, and conservation. We discuss the unique challenges of remote sensing data, including its temporal and multispectral nature, why geospatial machine learning differs from traditional computer vision, and AI2's philosophy of building open models and tools that can be adapted to real-world environmental problems.A major focus of the conversation is Latent MIM Lite, OlmoEarth's self-supervised pretraining approach. Joe explains how the method strikes a balance between masked autoencoders, which reconstruct pixels and train reliably but often learn weaker representations, and latent-space methods such as I-JEPA and Latent MIM, which can produce stronger features but are notoriously unstable. By replacing the target encoder with a frozen random linear projection in token space, Latent MIM Lite achieves stable training while preserving the benefits of latent-space prediction. We also discuss the broader challenges of evaluating geospatial foundation models, the trade-offs between embeddings and fine-tuning, and why practical performance on partner applications often matters more than leaderboard results.* 📺 Video of this conversation on YouTube* 🖥️ OlmoEarth on Github* 🖥️ OlmoEarth Platform* 👤 Joe’s websiteBio: Joseph Redmon is a research scientist at Ai2 building multimodal foundation models for geospatial data. As part of the OlmoEarth team he’s working to bring cutting edge AI research to non profits and NGOs working on conservation, ecological, and environmental problems. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Lakshay Sharma, a machine learning scientist at Instacart and former member of Microsoft’s geospatial AI team, to discuss self-supervised learning for remote sensing and his recent research on efficient pretraining for semantic segmentation. Lakshay explains the evolution of self-supervised learning, covering predictive, generative, and contrastive approaches, and discusses how foundation models such as DINO have transformed computer vision and geospatial machine learning. We explore the unique challenges of applying these techniques to remote sensing imagery, where assumptions that work for natural images often break down.We then dive into Lakshay’s recent paper, Sub-Image Overlap Prediction: Task-Aligned Self-Supervised Pretraining for Semantic Segmentation in Remote Sensing Imagery, presented at the Computer Vision for Earth Observation Workshop at WACV 2026. He walks through the intuition behind the method, which trains models to localize extracted sub-images within larger scenes as a proxy task for semantic segmentation. We discuss the experimental setup, comparisons against established self-supervised learning approaches, and the surprising finding that the method achieves competitive or superior results using only thousands of pretraining images rather than millions. Along the way, we explore transfer learning across datasets, the growing importance of data efficiency, and why targeted pretraining may offer a compelling alternative to increasingly resource-intensive foundation model development for niche geospatial applications.* 📺 Video of this conversation on YouTube* 👤 Lakshay on LinkedIn* 🖥️ Personal website of Lakshay* 📖 PaperBio: Lakshay Sharma is a Senior Machine Learning Scientist / Engineer at Instacart. His research spans Computer Vision (CV) and Vision-Language Models (VLMs) with a focus on Self-Supervised and Semi-Supervised Learning. He has previously worked at Microsoft on multi-modal representation learning, and using aerial/satellite and streetside imagery for maps and geospatial applications. He has also worked at Amazon where he was focused on representation learning for videos. Based in New York City, Lakshay is an avid fan of soccer, snowboarding, and cricket. He often daydreams of some day applying his computer vision chops to sports. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Jennifer Marcus and Isaac Corley from Taylor Geospatial to explore Fields of the World - an open initiative to create globally consistent agricultural field boundary datasets from satellite imagery using AI and cloud-native geospatial infrastructure. Taylor Geospatial, a newly formed research organization, is building openly licensed global datasets as foundational public goods. Jen and Isaac explain the motivation behind the project, the challenges of scaling machine learning beyond well-labelled regions, and why openness in datasets, tooling, and intermediate model outputs, is central to their approach.We dive into the technical details behind the first global release: assembling noisy and uneven benchmark datasets from around the world, training models that generalise across diverse agricultural systems, and releasing everything from Sentinel-2 mosaics and raw segmentation probabilities to polygonised field boundaries through Source Cooperative. Along the way, we discuss community-driven improvement loops inspired by OpenStreetMap, the limitations of 10 m imagery for smallholder agriculture, and the importance of pairing academic researchers with engineering teams to rapidly operationalise new methods. Finally, we look ahead to Taylor Geospatial’s next phase - richer agricultural datasets, “Features of the World,” and a benchmarking initiative aimed at improving evaluation standards and reproducibility across geospatial foundation models.* 📺 Video of this conversation on YouTube* 🖥️ Taylor Geospatial website* 🖥️ FTW websiteBio: Jennifer Marcus is Vice President of Strategic Innovation Programs at Taylor Geospatial, where she advances partnerships and programs that translate breakthrough geospatial AI research into real-world impact. With deep experience across defence, federal government, and open-source geospatial ecosystems, Jennifer brings decades of expertise translating emerging technologies into mission-critical impact. She previously served as the inaugural Executive Director of Taylor Geospatial Engine, which in 2024, launched what would become Fields of The World, and has held leadership roles at Planet, Boundless Spatial, and Northrop Grumman.Bio: Isaac Corley is Director of AI/ML Research at Taylor Geospatial, where he leads a team to build the models behind earth observation research and to create open data products that elevate the geospatial market and community as a whole. Isaac builds and publishes geospatial AI from research through production, including the RasterFlow platform at Wherobots, which was used to run Fields of The World. He has served as PI on the IARPA SMART program at BlackSky and maintains widely-used open-source projects, including TorchGeo and SMP. Check out his blog with Caleb Robinson at geospatialml.com. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Mikolaj (Miko) Czerkawski from Asterisk Labs to explore BetaEarth, an experimental open-source emulator trained on AlphaEarth Foundations' public embedding archive. AEF — released by Google and Google DeepMind as a global 10 m embedding product derived from a wide range of Earth-observation modalities — is what makes BetaEarth possible: its openness lets the community build lightweight independent emulators that approximate AEF's pixelwise outputs from standard Sentinel inputs, and use them to probe how much of a model's behaviour is captured in its public embeddings. Miko walks through BetaEarth's design — compact architectures based on SegFormer-B2 with separate per-modality encoders, and a shared DINOv3 backbone over 3-band spectral primitives — and the surprising finding that reasonably strong approximations can be achieved even from simple RGB inputs.We then dive into a live demo: generating BetaEarth embeddings for arbitrary regions and time ranges using Sentinel-1, Sentinel-2, and COP-DEM data. Along the way, we cover practical considerations such as cloud contamination, modality trade-offs, tiling artefacts, and strategies for merging multi-temporal signals. Finally, we discuss what this complementary tooling enables for the geospatial ML community — embeddings as pretraining or regularisation signals, lightweight local inference alongside AEF's global annual rasters, and what the combination of large proprietary archives and open emulator-style tools could unlock next.* 📺 Video of this conversation & demo on YouTube* 🖥️ BetaEarth Github page* 🖥️ BetaEarth demo on HuggingfaceBio: Miko is a researcher specialising in AI, computer vision, signal processing and Earth observation. Before co-founding Asterisk Labs he was a postdoctoral research fellow at the European Space Agency. His research interests include data-centric analyses of large-scale Earth observation data, dataset curation, generative modelling, and restoration tasks for satellite imagery. He is a co-founder of the Major TOM community project, a platform for collaborating and reusing Earth observation datasets designed specifically for AI pipelines. He received the B.Eng. degree in electronic and electrical engineering in 2019 from the University of Strathclyde in Glasgow, United Kingdom, and the Ph.D. degree in 2023 at the same institution, specialising in applications of computer vision to Earth observation data. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Kentaro Wada, a computer vision engineer at Mujin and creator of LabelMe, to explore the evolution of image annotation workflows. We discuss how his need to label data for a robotics challenge led to building one of the most widely used open-source annotation tools, and how it has evolved alongside the shift from traditional computer vision to deep learning. Kentaro explains the impact of foundation models like Segment Anything (SAM), and how annotation is rapidly moving toward a prompt-and-verify paradigm where models do the heavy lifting and humans focus on quality control. We also dive into his recent work integrating SAM into LabelMe, the challenges of applying these models to satellite imagery, and why approaches like bounding-box prompting outperform text in that domain. Finally, we cover new support for large, multi-channel geospatial data, practical deployment considerations, and what this means for scaling annotation in real-world machine learning systems. Note that a recording of this conversation, along with a demonstration of geospatial annotation using LabelMe, is available on YouTube via the links below:* 🖥️ LabelMe website* 🖥️ Kentaro’s personal website* 📺 Video of this conversation on YouTube* 📺 Demo video on YouTube Bio: Kentaro Wada was born in Japan in 1994. He received his B.Sc. (2016) and M.Sc. (2018) from Mechanical Engineering and Computer Science Department in The University of Tokyo (UTokyo). In his research at UTokyo, he was working on learning-based scene understanding for robotic manipulation at JSK Laboratory supervised by Prof. Masayuki Inaba and Prof. Kei Okada. He received his PhD in 2022, at Dyson Robotics Laboratory in Imperial College London supervised by Prof. Andrew Davison. During his PhD, he worked on object-level semantic scene understanding, a general scene representation useful for robotic manipulation, and showed several novel capabilities of robots. He joined Mujin, Inc. in 2022 as a computer vision engineer, and is working on advancing robots' capabilities in the real-world environment. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com

In this episode I sat down with Hannah Kerner and Tristan Grupp to discuss Fields of The World (FTW), an open-source benchmark and ecosystem for global field boundary segmentation from satellite imagery. We explore the core challenge of building models that generalise across vastly different agricultural systems, and why data diversity, rather than model architecture, is often the limiting factor. Hannah and Tristan explain how targeted annotation in underperforming regions can dramatically improve results, how combining global and local training data avoids catastrophic forgetting, and what they learned from large-scale model experimentation. We also dig into practical evaluation beyond standard IOU metrics, including consistency and throughput, and how small modelling choices like boundary loss weighting can have outsized impact on usability. Finally, we cover the growing tooling ecosystem, real-world user feedback, and what’s coming next, including improved models and a global map of predicted field boundaries.* 🖥️ FTW website* 🖥️ Trazo South American extension to FTW* 📺 Recording of this conversation on YouTubeBio Hannah: Hannah Kerner is an Assistant Professor in the School of Computing and Augmented Intelligence at Arizona State University. Her research focuses on advancing the foundations and applications of machine learning to foster a more sustainable, responsible, and fair future for all. Her lab’s research topics include machine learning for remote sensing, algorithmic bias, and machine learning theory. She translates research advances to real-world impact through her roles as the AI/Machine Learning Lead for NASA Harvest and NASA Acres, Center Faculty for the ASU Center for Global Discovery and Conservation Science (GDCS), and Research Director for Taylor Geospatial. She has been recognised by multiple research awards including NSF CAREER (2025), Schmidt Sciences AI2050 Early Career Fellowship (2025), and Forbes 30 Under 30 in Science (2021). Bio Tristan: Tristan Grupp is an Agricultural Data Scientist in the Food, Land, and Water Program and Data Lab at the World Resources Institute. He collaborates closely with Land and Carbon Lab. His current research focuses on applying remote sensing and machine learning to monitor deforestation and natural land conversion driven by agricultural supply chains, supporting commodity traceability and corporate sustainability compliance, including under the EU Deforestation Regulation (EUDR). His work spans forest change monitoring, climate adaptation, and the intersections of food systems and natural landscapes. Beyond WRI, Grupp has contributed to research on climate change adaptation tracking in support of national adaptation planning under the UNFCCC, protected area policy evaluation in the EU, and tropical forest dynamics in the Peruvian Amazon. He has presented his work at international venues including AGU, COP, and the UN National Adaptation Planning Conference. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.satellite-image-deep-learning.com