Locating the Coppershale in Dutch well logs: label reliability and machine learning across regions

Locating thin stratigraphic horizons in legacy well logs requires reliable reference depths and transfer between regions. We investigated the Dutch Coppershale using public NLOG stratigraphy and gamma-ray and sonic logs from 902 well systems in seven spatial blocks. We compared gamma-ray rules, template matching, boosted trees and matched hard-label and noise-aware convolutional networks, searching the full usable log interval. A local contrast analysis of 965 wellbores gave a median absolute label-to-log offset of 0.41 m, with 14.9% above 2 m. Blinded AI review of 300 panels yielded only 27 usable reference tops; a repeat pass found no agreement within 2 m in the four jointly picked cases, and no geologist reviewed the tops. Across held-out blocks, hard-label and noise-aware networks located 33.06% and 34.83% of reference tops within 2 m. Their paired difference was 1.77 percentage points (95% interval -0.95 to 3.67); database-top rates were 55.99% and 55.54%. Noise-aware training showed no detectable gain on either endpoint. Channel removal and coarser sampling reduced database agreement, while selective prediction traded acceptance for lower error. Documentary checks of 40 unrecorded penetrations found six explicit Coppershale labels and 34 undecidable cases, without establishing copper content or new occurrences. The low agreement between repeat picks limits what these comparisons can establish about geological accuracy.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24369113
Primary Topic
Geological Modeling and Analysis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Locating the Coppershale in Dutch well logs: label reliability and machine learning across regions

S. M. Abdulla Al Mamun
Zenodo (CERN European Organization for Nuclear Research)
Geological Modeling and Analysis
preprint

Locating the Coppershale in Dutch well logs: label reliability and machine learning across regions

S. M. Abdulla Al Mamun
preprint en

Abstract

Locating thin stratigraphic horizons in legacy well logs requires reliable reference depths and transfer between regions. We investigated the Dutch Coppershale using public NLOG stratigraphy and gamma-ray and sonic logs from 902 well systems in seven spatial blocks. We compared gamma-ray rules, template matching, boosted trees and matched hard-label and noise-aware convolutional networks, searching the full usable log interval. A local contrast analysis of 965 wellbores gave a median absolute label-to-log offset of 0.41 m, with 14.9% above 2 m. Blinded AI review of 300 panels yielded only 27 usable reference tops; a repeat pass found no agreement within 2 m in the four jointly picked cases, and no geologist reviewed the tops. Across held-out blocks, hard-label and noise-aware networks located 33.06% and 34.83% of reference tops within 2 m. Their paired difference was 1.77 percentage points (95% interval -0.95 to 3.67); database-top rates were 55.99% and 55.54%. Noise-aware training showed no detectable gain on either endpoint. Channel removal and coarser sampling reduced database agreement, while selective prediction traded acceptance for lower error. Documentary checks of 40 unrecorded penetrations found six explicit Coppershale labels and 34 undecidable cases, without establishing copper content or new occurrences. The low agreement between repeat picks limits what these comparisons can establish about geological accuracy.

Zenodo (CERN European Organization for Nuclear Research)
University of Manchester (GB)
Geological Modeling and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon