A new generation of AI models is learning to understand Earth's geography directly from satellite imagery, maps, and demographic data, rather than being trained for single tasks. Esri, a leader in geographic information systems (GIS), announced the release of geospatial foundation models that bring this capability to professionals working with location intelligence and Earth observation. Unlike traditional AI models that require extensive labeled data for each specific task, these foundation models learn rich, reusable representations of geographic information that can be adapted to many workflows with minimal additional training.

What Are Geospatial Foundation Models and Why Do They Matter?

Foundation models represent a fundamental shift in how AI approaches geospatial analysis. Rather than building separate models for every application, organizations can start with models that already understand geographic information and adapt them to a wide variety of GIS workflows. At the core of many of these models are embeddings, which are compact numerical representations that capture the essential characteristics of an image, location, or geographic feature. Similar places produce similar embeddings, enabling workflows such as similarity search, clustering, prediction, and retrieval.

The development of geospatial foundation models required more than advances in AI alone. It demanded deep understanding of geographic data and the workflows used to analyze it. By combining decades of geospatial expertise, authoritative Esri datasets, and flexible deployment patterns in ArcGIS, Esri is helping shape the next generation of AI for location intelligence and Earth observation.

What Three Types of Geospatial Models Is Esri Releasing?

Esri's new capabilities span three complementary areas, each designed to address different aspects of geospatial analysis and Earth observation:

• Location Encoder Models: These learn representations of places rather than individual images or features, capturing geographic, environmental, and socioeconomic characteristics that define each location. Esri has developed two complementary location encoder models: the Global Location Encoder, which learns from globally available Sentinel-2 satellite imagery to produce embeddings that capture characteristics of locations around the world, and the Geodemographic Foundation Model, which learns from thousands of authoritative demographic, socioeconomic, housing, and environmental variables including data from the U.S. Census and American Community Survey.
• Geospatial Vision Language Models: GeoVLM brings the power of multimodal AI to Earth observation by connecting satellite imagery with natural language. Instead of building separate models for every remote sensing task, users can interact with imagery using prompts. The model can support a wide range of remote sensing tasks through natural language prompting, including object detection, pixel classification and segmentation, image captioning, object counting, visual question answering, and image or region classification.
• Remote Sensing Foundation Models: Unlike traditional computer vision models pretrained on everyday photographs, these models are trained directly on satellite and aerial imagery from sensors such as Sentinel, Landsat, and NAIP. They provide stronger starting points for many Earth observation tasks, often requiring less labeled training data while delivering improved accuracy. Esri has integrated ArcGIS with several leading open-source remote sensing foundation models including TerraMind, Prithvi EO 2.0, Clay, DOFA, and DINO.

The Global Location Encoder (Sentinel-2) is available through ArcGIS Living Atlas as a Deep Learning Package (DLPK), allowing users to generate embeddings for locations virtually anywhere on Earth. The USA Geodemographic Embeddings are being released as a beta feature layer through ArcGIS Living Atlas. GeoVLM is also being released through ArcGIS Living Atlas as a Deep Learning Package, enabling organizations to incorporate multimodal AI into ArcGIS workflows while keeping their imagery and data within their own infrastructure.

How Can GIS Professionals Use These New Models?

The practical applications of geospatial foundation models span a wide range of real-world workflows that GIS professionals encounter regularly:

• Similarity Search and Clustering: Location embeddings enable professionals to find places with similar characteristics across large geographic areas, making it easier to identify comparable sites for analysis or decision-making without manually defining what makes locations similar.
• Predictive Modeling and Interpolation: By combining embeddings with traditional spatial information, professionals can build more accurate predictive models for tasks like site selection, market analysis, and spatial interpolation where demographic, socioeconomic, or environmental context plays an important role.
• Feature Extraction and Remote Sensing Tasks: GeoVLM enables natural language interaction with satellite imagery, allowing professionals to perform object detection, pixel classification, image captioning, and visual question answering without building separate models for each task.
• Change Detection and Environmental Monitoring: The Global Location Encoder can support change detection workflows, helping professionals monitor environmental changes over time by comparing embeddings from different time periods.

Evaluations across numerous predictive tasks have shown that geodemographic embeddings consistently improve predictive performance when demographic context is an important driver, particularly when combined with traditional explanatory variables.

Why Does This Represent a Major Shift in Geospatial AI?

The emergence of geospatial foundation models marks a departure from the traditional approach to Earth observation and location intelligence. Historically, organizations had to build and maintain separate models for each specific task, requiring extensive labeled training data and domain expertise. Foundation models change this equation by providing a pre-trained starting point that already understands geographic patterns and relationships. This reduces the barrier to entry for organizations wanting to leverage advanced AI in their geospatial workflows and accelerates the time from data to insights. The integration of multiple open-source remote sensing foundation models into ArcGIS also demonstrates how the geospatial AI community is moving toward standardized, interoperable tools that professionals can use without leaving familiar workflows.

The development of these models was supported by significant computational resources. Esri acknowledged AWS for providing the cloud infrastructure that helped accelerate the training of these models, underscoring the computational scale required to build foundation models that can understand geographic patterns across the entire planet.