A flowchart.
The workflow used in the paper is illustrated here, showing how synthetic geological models were constructed and how their geophysical properties were derived via forward simulation. Then, different statistical techniques were applied to the data, including the training of a neural network to recognize lithology (the “ground truth”), an assessment of collinearity of features, and a projection of the data using principal components (effectively removing collinearity). Self-Organizing maps are then tested on different subsets of this data, and the paper demonstrates that the algorithm is relatively stable even when colinear features are included. Credit: Xu et al. [2026], Figure 3
Editors’ Highlights are summaries of recent papers by AGU’s journal editors.
Source: Journal of Geophysical Research: Machine Learning and Computation

Is collinearity, the tendency of variables to be related to one another, always a problem in machine learning? This has been a long-held view in statistical machine learning because linear models like linear regression become instable if you use variables that are collinear as input parameters.

In a new study, Xu et al. [2026] demonstrate that although this thinking applies to a lot of algorithms used to group similar facies in seismic data, this is not true for self-organizing maps (a form of grouping algorithm). The authors did this by looking at both synthetic and real data, both using seismic data and geological maps. This is truly exciting as there might be other algorithms out there that are robust to this problem of collinearity, and this has implications beyond geophysics.

Citation:Xu, L., Feltrin, L., & Green, E. C. R. (2026). Rethinking collinearity in self-organizing maps: Evidence from geophysical data classification. Journal of Geophysical Research: Machine Learning and Computation, 3, e2025JH001107.  https://doi.org/10.1029/2025JH001107

—Cedric John, Editor, JGR: Machine Learning and Computation

This research is included in AGU’s Special Collection “Advancing Interpretable AI/ML Methods for Deeper Insights and Mechanistic Understanding in Earth Sciences: Beyond Predictive Capabilities.“

Text © 2026. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.