EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Modern Neural Networks for Small Tabular Datasets: The New Default for Field‐Scale Digital Soil Mapping?

Viacheslav Barkov, Jonas Schmidinger, Robin Gebbers, Martin Atzmueller

European Journal of Soil Science · 2026-03 · 77 (2)

DOI: 10.1111/ejss.70299 ↗

Kurz gefasst

Fachartikel in der Zeitschrift European Journal of Soil Science, online erschienen am 24.02.2026. Verfasst von Viacheslav Barkov und 3 weiteren Autor:innen. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 76 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deutsche Forschungsgemeinschaft.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In the field of pedometrics, tabular machine learning is the predominant method for soil property prediction from remote and proximal soil sensing data, forming a central component of Digital Soil Mapping (DSM). At the field‐scale, this pedometric modeling task is typically constrained by small training sample sizes and high feature‐to‐sample ratios, particularly when using spectroscopic data. Traditionally, these conditions have proven challenging for conventional deep learning methods. Classical machine learning algorithms, particularly tree‐based models like Random Forest and linear models such as Partial Least Squares Regression, have long been the default choice for pedometric modeling within DSM. Recent advances in artificial neural networks (ANN) for tabular data challenge this view, yet their suitability for field‐scale DSM has not been proven. We introduce a comprehensive benchmark that evaluates state‐of‐the‐art ANN architectures, including the latest multilayer perceptron (MLP)‐based models (TabM, RealMLP), attention‐based transformer variants (FT‐Transformer, ExcelFormer, T2G‐Former, AMFormer), retrieval‐augmented approaches (TabR, ModernNCA), and an in‐context learning foundation model (TabPFN). Our evaluation encompasses 31 field‐ and farm‐scale datasets containing 30–460 soil samples and three critical soil properties: soil organic matter or soil organic carbon, pH, and clay content. Our results reveal that modern ANNs consistently outperform classical methods on the majority of tasks, demonstrating that deep learning has matured sufficiently to overcome the long‐standing dominance of classical machine learning for soil property prediction. Notably, TabPFN delivers the strongest overall performance, showing robustness across varying conditions. We therefore recommend the adoption of modern ANNs for field‐scale DSM and propose TabPFN as the new default choice in the toolkit of every pedometrician.

Förderung

  • Deutsche Forschungsgemeinschaft

Zitieren

APA

Barkov, V., Schmidinger, J., Gebbers, R., & Atzmueller, M. (2026). Modern Neural Networks for Small Tabular Datasets: The New Default for Field‐Scale Digital Soil Mapping?. European Journal of Soil Science, 77(2). https://doi.org/10.1111/ejss.70299

Vancouver

Barkov V, Schmidinger J, Gebbers R, Atzmueller M. Modern Neural Networks for Small Tabular Datasets: The New Default for Field‐Scale Digital Soil Mapping?. European Journal of Soil Science. 2026;77(2). doi:10.1111/ejss.70299

BibTeX
@article{Barkov2026,
  author = {Barkov, Viacheslav and Schmidinger, Jonas and Gebbers, Robin and Atzmueller, Martin},
  title = {Modern Neural Networks for Small Tabular Datasets: The New Default for Field‐Scale Digital Soil Mapping?},
  journal = {European Journal of Soil Science},
  year = {2026},
  volume = {77},
  number = {2},
  doi = {10.1111/ejss.70299}
}
Kostenlos starten →