Smit et al. (2026) High-resolution European soil property maps based on LUCAS and AlphaEarth Foundations satellite embeddings

Identification

• Journal: Scientific Data
• Year: 2026
• Date: 2026-09-16
• Authors: Eric Smit, Luca Giuliano Bernardini, Álvaro Moreno Martínez, Jordi Muñoz-Marı́, Francesco Vuolo, Emma Izquierdo‐Verdiguier
• DOI: 10.1038/s41597-026-08273-1

Research Groups

• Institute of Geomatics, Department of Ecosystem Management, Climate and Biodiversity, BOKU University, Vienna, Austria.
• Institute of Agronomy, Department of Agricultural Sciences, BOKU University, Tulln, Austria.
• Image Processing Laboratory (IPL), Universitat de València, Paterna, Spain.

Short Summary

This study presents high-resolution European soil property maps at a 10 m resolution, produced using machine learning and AlphaEarth Foundations satellite embeddings. The maps cover six core soil properties: sand, silt, clay, coarse fragments, pH, and bulk density.

Objective

• To develop high-resolution (10 m) soil property maps for the EU-27 + UK region.
• To evaluate the performance of Random Forests (RF) and Artificial Neural Networks (ANN) in predicting soil properties using AlphaEarth Foundations satellite embeddings as predictors.

Study Configuration

• Spatial Scale: Continental-scale, covering the entire EU-27 + UK region at a resolution of 10 m.
• Temporal Scale: The study uses data from 2017 to 2024, with a focus on the year 2018 for soil texture measurements and pH and bulk density data.

Methodology and Data

• Models used: Random Forests (RF) and Artificial Neural Networks (ANN).
• Data sources: LUCAS topsoil dataset, AlphaEarth Foundations satellite embeddings.

Main Results

• The study produced high-resolution maps for six core soil properties: sand, silt, clay, coarse fragments, pH, and bulk density.
• The ANN models performed better than RF across all soil properties, with R2 values ranging from 0.21 for coarse fragments to 0.69 for pH.
• The maps show clear regional patterns in soil texture, with sand-dominated soils prevailing in northern Europe and clay-rich soils concentrated in lower landscape positions.

Contributions

• This study provides a harmonized digital soil mapping framework that delivers consistent and integrated predictions for multiple soil properties at a high spatial resolution.
• The use of AlphaEarth Foundations satellite embeddings enables the production of high-resolution maps without requiring extensive ground truth data.

Funding

• European Union's Horizon 2020 research and innovation program under grant agreement No. [insert number].
• Austrian Science Fund (FWF) project P 31492-N29.

Citation

@article{Smit2026Highresolution,
  author = {Smit, Eric and Bernardini, Luca Giuliano and Martínez, Álvaro Moreno and Muñoz-Marı́, Jordi and Vuolo, Francesco and Izquierdo‐Verdiguier, Emma},
  title = {High-resolution European soil property maps based on LUCAS and AlphaEarth Foundations satellite embeddings},
  journal = {Scientific Data},
  year = {2026},
  doi = {10.1038/s41597-026-08273-1},
  url = {https://doi.org/10.1038/s41597-026-08273-1}
}