Based on the EnMAP L2A data collection [1], a 3.3 terabytes dataset of over half a million hyperspectral image patches of size 128x128 pixels and 202 channels has been assembled: ″SpectralEarth″ [2] aims to share a valuable asset for the training of hyperspectral foundation models and self-supervised machine learning algorithms. To that end, SpectralEarth ships with a subset of annotations for various land cover classification tasks.

The data can be downloaded directly from the EOC Download Service. [3]

Further information on the data can be found in the publication. [4]