Carbon Watch
Satellite, drone, and ground-truth data pipelines for carbon readiness, forest health analysis, and environmental reporting.
Problem
Environmental teams needed to combine satellite imagery, drone observations, and ground data into a trustworthy workflow for carbon and forest integrity analysis.
My role
AI Engineer working on geospatial data engineering, serverless processing, ML workflow patterns, and reporting architecture in a past Thaicom role.
System design
The system connected ingestion, validation, geospatial transformation, model inference, and reporting layers for raster and vector data.
Data / AI approach
Raster and vector sources were normalized, indexed, and prepared for spatial analytics and ML-assisted environmental assessment.
Architecture
AWS Lambda, S3, SageMaker, PostGIS, and observability tooling supported asynchronous geospatial processing and analysis workflows.
Impact
Carbon Watch helped translate complex environmental data into carbon readiness and forest health intelligence that stakeholders could review.
Tech stack
Key learnings
- Spatial systems need strong metadata and validation before model output is meaningful.
- Asynchronous processing fits heavy geospatial workloads better than request-time workflows.
- Environmental AI must make uncertainty and data provenance visible.
Expandable details
Geospatial Workflow+
Satellite, drone, and ground truth data move through validation, transformation, spatial indexing, inference, and reporting layers.
Operational Considerations+
Pipeline design prioritized lineage, repeatability, and monitoring so the system could support recurring environmental analysis.