Learning to Guide Ontology-Driven Geological Interpretation with Graph Neural Networks
Abstract
Building 3D geological models from sparse observations relies on expert interpretation whose reasoning is rarely documented, making models difficult to validate and reproduce. An ontology-based system, built around the Minimal Ontology for Geological Interpretation (MOGI) and a Geological Interpretation Process (GIP), was recently proposed to make this reasoning explicit. It formalizes interpretation as an iterative cycle and records every reasoning step as a trace. However, two decisions of this cycle, which situations to interpret next and which geological explanation to assign to them, are currently made randomly or with simple heuristics. We propose to learn these decisions from past reasoning traces, each trace is encoded as a graph labelled with a quality score. A Relational Graph Convolutional Network (R-GCN)-based evaluator is trained to predict this score and guide decisions during the interpretation cycle. In a proof-of-concept on a synthetic fold dataset, the evaluator ranks interpretation states with a Pearson correlation of 0.98.
Related
- An AI-assisted conditioning and geological interpretation workflow for usage in implicit geological modeling
- Emirates Exploration Imager (EXI) Reveals Crater and Boulder Distribution on Deimos: Insights Into Its Dynamical and Geological Evolution
- Escaping Geological Generative Prior Distributions in Bayesian Travel Time Tomography
- Two-stage assessment of regional geological hazard susceptibility under mining disturbance
- Rock carving depth reflects carvers’ geological insight
BibTeX
@misc{attia2026learning,
title = {{Learning to Guide Ontology-Driven Geological Interpretation with Graph Neural Networks}},
author = {Aya Attia and Gautier Laurent and Imadeddine Laouici},
howpublished = {HAL},
year = {2026},
month = nov,
url = {https://hal.science/hal-05754951v2},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Learning to Guide Ontology-Driven Geological Interpretation with Graph Neural Networks
AU - Aya Attia
AU - Gautier Laurent
AU - Imadeddine Laouici
PY - 2026
DA - 2026/11/04
PB - HAL
UR - https://hal.science/hal-05754951v2
AB - Building 3D geological models from sparse observations relies on expert interpretation whose reasoning is rarely documented, making models difficult to validate and reproduce. An ontology-based system, built around the Minimal Ontology for Geological Interpretation (MOGI) and a Geological Interpretation Process (GIP), was recently proposed to make this reasoning explicit. It formalizes interpretation as an iterative cycle and records every reasoning step as a trace. However, two decisions of this cycle, which situations to interpret next and which geological explanation to assign to them, are currently made randomly or with simple heuristics. We propose to learn these decisions from past reasoning traces, each trace is encoded as a graph labelled with a quality score. A Relational Graph Convolutional Network (R-GCN)-based evaluator is trained to predict this score and guide decisions during the interpretation cycle. In a proof-of-concept on a synthetic fold dataset, the evaluator ranks interpretation states with a Pearson correlation of 0.98.
ER -