RapidAI4EO: A Corpus of Dense Time Series Satellite Imagery

The RapidAI4EO Corpus is a multimodal dataset of dense time series satellite imagery sampled at 500,000 locations across Europe. It combines monthly Sentinel-2 image mosaics with high cadence harmonized and cloud free Planet Fusion imagery. The corpus was designed to support the development of machine learning models for land use and land cover (LULC) classification and change detection in the ontology of the CORINE land cover (CLC) inventory. It can be used for other use cases especially those which benefit from high cadences time series of satellite imagery.

The corpus is available on Radiant Earth’s Source Cooperative https://rapidai4eo.source.coop

Key facts

• 500,000 data cube locations
• 600m x 600m patch size
• Sentinel-2 and Planet Fusion imagery (see table below)
• Multi-class annotations based on CORINE Land Cover 2018
• Locations were sampled to account for
  • CLC class distribution
  • Spatial distribution
  • Country representation
• Designed for LULC change detection, but generalisable to other research areas

Overview of corpus sampling locations (a) of Fusion tiles selected for subsampling and (b) of patch locations sampled from a single example Fusion tile.