We are organising the Neural Earth Fields Benchmark & Hackathon, a 24-hour sprint on 7–8 December 2026 at the IMPULSE House of the University of Bonn. Teams will build neural networks that take longitude and latitude as input and reconstruct global geospatial layers, from water storage anomalies and ESA WorldCover land cover to iNaturalist biodiversity observations. The event is co-organised by Marc Rußwurm, Jürgen Gall, and Jürgen Kusche (University of Bonn) with Konstantin Klemmer (UCL), Alistair Francis (Asterisk Labs), Isaac Corley (Taylor Geospatial), and Kyle Gao (Aalto University).

Early-career researchers in geospatial science or machine learning can apply until 16 October 2026. More information →

The paper “Assessing the effectiveness of deep embeddings for tree species classification in the Dutch Forest Inventory” has been published in Science of Remote Sensing. The paper was led by Takayuki Ishikawa in collaboration with Carmelo Bonannella, Bas J.W. Lerink, and Marc Rußwurm.

The study evaluates the utility of deep earth observation embeddings for tree species classification using data from the Dutch National Forest Inventory (NBI), providing insights into the transferability of foundation model representations for forestry applications.

Read the paper

PatchMatch — our work on cross-sensor matching of marine litter patches — has been accepted as a poster at the ECCV 2026 Marine Vision Workshop. The paper is authored by Gabrielė Tijūnaitytė, Marc Rußwurm, and Emanuele Dalsasso. Read more →

Our perspective paper “Earth Embeddings: Toward AI-centric Representations of our Planet“ has been published in the IEEE Geoscience and Remote Sensing Magazine. We argue that Earth Embeddings represent the next abstraction layer for geospatial AI — shared representations of our planet learned from satellite imagery, maps, elevation, climate, and other geospatial data. Read more →

Marc Rußwurm will be on parental leave from (approximately) May 1st until September 2026. Responses will be delayed.