Papers, preprints, code repositories, and datasets from our research programme.
James GC Ball, Jana Annika Wicklein, Zhengpeng Feng, Jovana Knezevic, Clement Atzberger, Michele Dalponte, David Coomes
We evaluate two geospatial foundation-model embeddings, AlphaEarth and Tessera, for tree species classification in the Trentino region of northern Italy, using parcel-level forest inventories as reference data (18 species and species groups). Foundation-model embeddings substantially outperform composite-based multispectral satellite baselines, achieving higher accuracy with far fewer training labels and preserving ecologically meaningful structure aligned with functional and taxonomic groupings.
Complete species classification pipeline: soft-label training, model training, evaluation, and visualisation. Python, scikit-learn, PyTorch.
github.com/PatBall1/trentino-treesProcessed labels, evaluation results, and model outputs. Available upon publication.
DOI forthcomingFeng et al., 2025
The Tessera geospatial foundation model providing 128-dimensional embeddings from multi-sensor seasonal composites at 10 m resolution.
Brown et al., 2025
Google’s AlphaEarth Foundations geospatial foundation model providing multi-source Earth surface representations.
Ball et al., 2024
Prior work on the role of foundation models in reducing labelling bottlenecks for ecological remote sensing applications.
Jung, 2017
The first global, spatially explicit characterisation of terrestrial habitat types at 100 m resolution following the IUCN habitat classification scheme.