AWS Geospatial Agent Repo
An open‑source Geospatial Agent for Sentinel‑2 imagery was posted that ties Amazon Location with TiTiler and MapLibre to produce NDVI, NDWI and NBR analyses. (The public repo and demo were shared on social and highlight a practical Sentinel‑2 processing stack.) (x.com)
A new AWS sample shows how to turn plain-English questions into satellite-image maps for crops, water and fire scars using Sentinel‑2 data. (github.com) The public repository, `aws-samples/sample-geospatial-agent-on-aws`, describes a three-part stack: an agent, a tile server and a frontend. Its README says deployment takes about 15 minutes and lists example prompts for Central Park, Folsom Lake and Pacific Palisades. (github.com) The map layer is built from Sentinel‑2, a pair of European Space Agency satellites that provide optical land imagery about every five days. AWS’s open-data registry says Sentinel‑2 L1C data go back to June 2015 and L2A data have been available globally since January 2017. (registry.opendata.aws) The basic trick is band math: the software compares different wavelengths in the image to estimate conditions on the ground. The repo highlights three standard outputs — Normalized Difference Vegetation Index for plant vigor, Normalized Difference Water Index for water and moisture, and Normalized Burn Ratio for fire damage. (github.com) (docs.aws.amazon.com) The sample matters because it packages pieces that geospatial teams often wire together by hand. AWS says Amazon Location works with MapLibre for map rendering, and the repo uses that browser map with a separate tile server to display the analysis results. (docs.aws.amazon.com) (github.com) The README says the agent runs with Strands Agents and Claude Sonnet 4.6 on Amazon Bedrock AgentCore Runtime, while the user interface uses React, TypeScript and MapLibre GL. It also lists Cognito authentication and deployment on Amazon ECS Fargate behind CloudFront. (github.com) AWS already had another public geospatial-agent sample, but that one is broader and code-driven. A separate repo published about two weeks ago says users draw a polygon, ask a question, and the agent writes and executes Python to fetch imagery, run analysis and return overlays and statistics. (github.com) That older sample also reaches beyond Sentinel‑2. Its README lists Landsat Collection 2 Level 2 alongside Sentinel‑2, and says imagery is retrieved through SpatioTemporal Asset Catalog listings from Element 84’s Earth Search API as Cloud-Optimized GeoTIFF files. (github.com) The newer AWS sample is narrower and more productized: ask for vegetation health, water change or wildfire damage, and get a rendered map back in the browser. For developers building geospatial apps on AWS, the repo is less a research demo than a working reference architecture. (github.com)
