From Machine Learning to Large-Scale EO Products
Good Practices and Guidelines for Making Maps
Welcome
The past decade has seen a rapid expansion in the production of large-scale geospatial products derived from Earth observation (EO) data using machine learning (ML). This website is the result of bringing together map-makers across the EO and ML communities. It presents an end-to-end account of the good practices, challenges, and common pitfalls that span the entire pipeline from raw satellite data to a published, validated map product, and treats that pipeline as an integrated workflow rather than a sequence of independent steps. The focus is on wall-to-wall raster maps at moderate resolution (10–30 m) and continental to global extent, produced with ML from open satellite archives, mainly Sentinel-1, Sentinel-2 and Landsat.
We organize the discussion around six interconnected themes which trace the pipeline from the EO data landscape and preprocessing, through ML-specific stages of dataset construction, model design, and training, to uncertainty quantification, map generation, dissemination, and validation.
Main Sections
Overview of the EO data infrastructure landscape — Satellite missions, data access platforms, and the practical trade-offs of working with petabyte-scale archives.
Data selection and preprocessing — Temporal strategies, cloud masking, SAR preprocessing, and the hidden choices that shape the training signal.
ML dataset construction and model training — Data formats, spatial splitting, gridding, model design, and training considerations specific to EO.
Uncertainty quantification — Sources of uncertainty, quantification methods, calibration diagnostics, and how to operationalize uncertainty in map products.
Map production and distribution — Tiled inference at scale, artifact mitigation, post-processing, and cloud-native dissemination.
Validation — Design-based accuracy assessment, the distinction between model evaluation and map validation, and reporting standards.