AI-powered High-Resolution Enhanced Land Cover Mapping with 20+ Classes and Contextual Change

Impact Observatory, a US Small Business, builds AI-powered geospatial monitoring to understand risks and anticipate change at unprecedented speed and scale.  In their previous NGA SBIR Phase II, “Seeing the World in Real-Time with Automated Land Use Mapping,” Impact Observatory demonstrated production of the world’s first 16-class on-demand land use land cover maps at 3m resolution using deep learning algorithms applied to Sentinel-2 and PlanetScope satellite imagery. We now propose to further enhance these capabilities to create maps with up to 20+ land cover classes, using deep learning AI applied to public and commercial sources of imagery and updated annually and quarterly for anywhere in the world.  This project will also advance the state of the art in wide-area monitoring with contextual change.  The fully automated prototype systems developed in this project will demonstrate the viability, performance metrics, process modifications, and engineering required to deliver the land cover maps and contextual change in raster format as well as in vector format consistent with the established VISNAV and TDS schemas.    Fusing imagery from public and commercial Earth observation constellations enables global coverage in near real-time. Automated analysis will be based on Impact Observatory’s deep learning image classification and analysis algorithms leveraging the cloud compute and imagery archives of Microsoft Azure Planetary Computer, and demonstrating easy, open distribution of results via cloud-based SpatioTemporal Asset Catalog (“STAC”) services, open data formats, and APIs. Leveraging a cloud-based capability, versus an on-premises solution, will enable further integration with new sources of data as they become available, for future resolution and temporal enhancements. The results of this work will be further developed and offered as a commercial service to the US Government and users in industry and finance.   We will develop a prototype system for mapping and monitoring land cover and contextual change in near real-time with several key outcomes:  Understand human activities and natural phenomena with 20+ land cover classes; Understand the sources and context of changes; Deliver and scale maps and change detection in both raster and vector formats; Support stakeholders across the US Government and Allies/Partner Nations; and Monitor climate/environmental risks and resource depletion driving future global conflict.