EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observations and Wikipedia

We propose a method to learn ecological properties of aerial images by learning an alignement with species habitat descriptions.

We release the WikiRS dataset, composed of triplets of images, species list and Wikipedia sentences :

91k high-resolution aerial images (50cm, RGB bands) from the swissIMAGE product 

crowd-sourced species observations from 2745 different species,  collected from GBIF, geolocated within the footprint of the aerial image.

103K different sentences describing the habitat of the observed species, extracted from the corresponding Wikipedia article.

For more details on the dataset construction, see the full article on  arxiv https://arxiv.org/abs/2504.19742 and the git repo