A paper from the Schools of Emergent Priorities and Underlying Conditions compares an expert system built from field geologists' stated rules with one learned from their archived section logs, in the long-running problem of telling glacial till from debris-flow deposits.
A paper from two schools builds the same classifier twice, once from interviews and once from decades of section logs, and finds the logs the better teacher
Telling a glacial till from the debris flow that has reworked it is one of the oldest judgement calls in field geology, and a new paper asks where that judgement is kept: in what geologists say about it, or in what they have written down at the outcrop.
What the Section Logs Knew is by Dr Fenna Okoro, Senior Lecturer and Convenor of the Horizon Register in the School of Emergent Priorities, and Associate Professor Ragnhild Uzoma, Lead of the Basement Relations Group in the School of Underlying Conditions.
We owe this paper to nine colleagues who were generous with their afternoons and more generous still with their filing cabinets. The School is grateful to everyone who has ever taken the trouble to fill in the remarks line.
— Dr Fenna Okoro, Senior Lecturer and Convenor, Horizon Register
The study situates knowledge engineering within the everyday practice of Quaternary mapping, setting an interview-led rule base beside one learned from the field record and testing both against exposures neither had seen. Across 86 blind sections the log-based system came within two calls of the geologists themselves, while the interview-based system found the debris flows hardest of all.
The work sits comfortably within the University’s growing interest in what its records know before its people are ready to say it. Associate Professor Uzoma said the paper had left the School of Underlying Conditions “with some very good questions about its own log sheets, which is exactly where a field season ought to leave a school”.
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.jgky9c.