U.S. NGA moves to replace manual geospatial analysis with AI

The United States National Geospatial-Intelligence Agency (NGA) issued a Request for Information on March 16, 2026, seeking industry solutions to develop an automated system for extracting geospatial features from imagery and maps as part of its Foundation Digital Twin program.

The agency is exploring Automated Feature Extraction (AFE) technologies that can detect, extract, and classify real-world objects such as buildings, roads, infrastructure, and terrain features directly from imagery and raster maps. The system is expected to operate as a service accessible through application programming interfaces and integrated into existing NGA systems.

The NGA said the capability is intended to support its long-term modernization goals under the GEOINT 2035 concept, which focuses on faster and more accurate intelligence delivery to decision-makers.

The RFI outlines key technical requirements for the system. Solutions must be capable of processing multiple data types, including electro-optical, synthetic aperture radar, hyperspectral, and multispectral imagery. They must also handle both single-image and large-scale batch processing across different geographic regions.

The system is required to automatically detect objects, generate accurate geometric representations, and assign attributes based on observable characteristics. These functions—object detection, geometry extraction, and attribution—form the core of the AFE capability.

According to the NGA, the system must preserve spatial accuracy, metadata, and relationships between features throughout processing. It must also provide confidence metrics to guide analysts during validation and correction tasks.

The agency is also requiring compatibility with standard geospatial formats such as GeoDatabase and Shapefile, ensuring integration with existing intelligence systems. Security requirements include compliance with U.S. intelligence community standards and support for controlled access based on user roles and clearances.

NGA is seeking solutions that can achieve high levels of accuracy. According to the accompanying technical questions, the agency is targeting objective accuracy thresholds of at least 90 percent, with a goal of reaching 99 percent for certain feature types.

The agency also outlined specific use cases to guide industry responses. These include identifying buildings and infrastructure, mapping transportation networks, detecting utilities such as power lines and pipelines, and extracting aeronautical features like runways and helipads.

For example, the system should be able to map entire urban areas, classify building types, and link structures into larger facilities such as military installations. It should also be capable of analyzing transportation routes, identifying bottlenecks, and assessing infrastructure constraints.

The AFE capability is expected to support both civilian and military intelligence functions, including mapping, navigation, infrastructure analysis, and operational planning.

From a technical standpoint, the system would rely heavily on artificial intelligence and machine learning models trained to recognize patterns in imagery. These models would process incoming data, identify relevant features, and convert them into structured geospatial datasets.

Automating feature extraction addresses the need to process large datasets quickly, enabling faster decision-making and improving situational awareness in both military and intelligence operations.