SoilNet published in Geoderma

Our latest research got published in Geoderma (Elsevier): "SoilNet: A multimodal multitask model for hierarchical classification of soil horizons".

I am proud to see our work on SoilNet featured by our partners from the Federal Institute for Geosciences and Natural Resources (BGR). While general-purpose LLM's are impactful and impressive, our research in Geoderma demonstrates that for specialized fields like pedology they still hit a wall when faced with the structural logic and complexity.

Food security: Accurate soil data is the foundation of global food security, and by automating the classification process, SoilNet contributes towards scalable high-resolution mapping for sustainable land management. We are happy to share more details in the attached post.

Thanks: This work was a collaborative effort between geoscientists and ML researchers. A sincere thank you to my co-authors: Teodor Chiaburu, Einar Eberhardt, Stefan Broda, Joey Prüssing, Frank Haußer, and Felix Biessmann.

Read the full study here: https://doi.org/10.1016/j.geoderma.2026.117684