AWS Architecture Blog·October 22, 2025Scalable Geospatial Data Platform with STAC on AWS EKS
This article details BASF Digital Farming's architecture for a scalable geospatial data platform on AWS, leveraging Amazon EKS, S3, and RDS. It focuses on managing vast amounts of satellite imagery and other geospatial assets using the SpatioTemporal Asset Catalog (STAC) specification and the open-source eoAPI ecosystem. The solution emphasizes dynamic scaling, optimized storage, and efficient metadata management to process and deliver actionable insights to farmers globally.
Read original on AWS Architecture BlogBASF Digital Farming's xarvio FIELD MANAGER platform processes hundreds of millions of satellite images, generating billions of geospatial artifacts. Unlike traditional static data providers, the agricultural environment demands ingesting near-daily imagery from diverse sensors and providers. This necessitates a robust, cloud-based infrastructure capable of handling massive data velocity and volume, including advanced quality assurance and machine learning pipelines.
Leveraging STAC and eoAPI for Geospatial Data Management
The core of the solution is the SpatioTemporal Asset Catalog (STAC) specification, an open standard for describing and cataloging raster and vector datasets. STAC standardizes metadata across diverse sources (satellite imagery, UAV data, prescription maps), simplifying search, filtering, and retrieval. The platform is built on the eoAPI ecosystem, an integrated suite of open-source tools for cloud-based geospatial data lifecycle management.
- pgSTAC: A performant PostGIS-backed STAC API implementation used for indexing millions of STAC Items, supporting efficient spatial, temporal, and attribute-based filtering.
- Tiles in PostGIS (TiPG): Serves tiled vector data (field boundaries, management zones) directly from PostGIS as lightweight Mapbox Vector Tiles (MVT), eliminating the need for an external tile server.
- TiTiler: A dynamic tile server for Cloud Optimized GeoTIFFs (COGs), streaming imagery on-demand as WMTS or XYZ tiles, enabling dynamic rendering (e.g., NDVI) for web and mobile apps.
Architectural Overview on AWS
The solution is built on Amazon EKS for core computing, Amazon S3 for storage, and Amazon RDS with PostgreSQL for metadata management. It comprises four main layers: Core Services, Storage, Database, and Ingestion.
Core Services Layer
An EKS cluster hosts three key containerized services: stac-service (STAC API implementation), raster-service (TiTiler for COG rendering), and vector-service (TiPG for MVT serving). These are orchestrated for high availability and modularity, supporting CI/CD workflows. Kubernetes Event-Driven Autoscaling (KEDA) is used to dynamically scale pods based on metrics like STAC ingestion queue depth or visualization request load, ensuring responsive performance and cost efficiency.
- Geospatial Asset Storage Layer: All raw and processed geospatial assets are stored in S3 buckets, optimized for performance and durability. Cloud Optimized GeoTIFF (COG) for raster imagery and FlatGeobuf for vector data are chosen for streaming access, indexing, and cloud performance, reducing the need for heavy ETL pipelines.
- Database Layer: A PostgreSQL database on Amazon RDS, extended with pgSTAC, forms the metadata backbone. An Amazon RDS Proxy is crucial here, providing connection pooling and resiliency, preventing connection exhaustion during sudden scale-up events from KEDA-driven EKS pods.
- Ingestion Layer: An independent component handles batch or streaming geospatial data inputs, processing satellite imagery, drone data, or prescription maps. It pushes metadata to the STAC API and assets to S3. This decoupled engine supports high-throughput, reliable data integration at scale, capable of ingesting thousands of items per second.
Amazon API Gateway provides secure public access, acting as a unified entry point for browser-based and mobile clients, applying rate limiting, authorization, and routing policies.