How to Use This Book
Overview
This book presents an end-to-end account of best practices for producing large-scale geospatial products from Earth observation data using machine learning. It is organized around six interconnected themes that trace the pipeline from raw satellite data to a published, validated map product. Each chapter can be read independently or as part of the full pipeline.
Structure
The book is divided into the following parts:
Overview of the EO data infrastructure landscape covers the satellite missions and data access platforms available to practitioners, and the practical trade-offs involved in choosing between them.
Data selection and preprocessing addresses how to select which observations to include and how to process them into analysis-ready formats, covering both optical and SAR data.
Machine learning dataset construction and model training discusses data formats, spatial splitting strategies, gridding choices, and model design considerations specific to EO data.
Uncertainty quantification surveys sources of uncertainty in EO-derived maps and methods for quantifying, calibrating, and operationalizing uncertainty estimates.
Map production and distribution covers the engineering challenges of tiled inference at scale, post-processing, and dissemination in cloud-native formats.
Validation distinguishes model evaluation from map validation and presents design-based accuracy assessment and complementary validation approaches.
How to Cite
Contributing
This book is hosted on GitHub. Contributions, corrections, and suggestions are welcome via pull requests or issues.