Are geological description practices and classification schemes fit for future use? Breccias as an example

Is peer-reviewed geoscience research literature, with its extensive quantitative, semi-quantitative, and qualitative information, fit for use for artificial intelligence (AI) applications – both as potential training datasets for machine learning, and as a tool to help researchers keep up to date with the latest research? We address this question by examining data collection and reporting philosophies and practices in the literature for carbonate breccias – rocks that yield a spectrum of qualitative and quantitative data. These breccias can form by a wide range of processes in many different environments. Accurate interpretation of their formation mechanism can be important for many different geoscience applications, from environmental reconstructions through to understanding subsurface fluid flow.

We explore 7 different types of carbonate breccia summarising their formation mechanisms and characteristics and use this to isolate the breccia characteristics most valuable for their description and interpretation. We then examine 59 published case studies, and 8 breccia classification schemes and find that reporting of breccia characteristics is inconsistent between case studies. The characteristics most often reported in research and used in classification schemes are common to all breccia types and are of low diagnostic value, while some of the most valuable characteristics for interpretation (e.g. nature of clast boundaries) are the least reported. We propose a suite of observations that should be made for all carbonate breccia studies and recommend that negative observations should be explicitly recorded. Without this, using published literature in AI applications is likely to yield unreliable results.