Reproducible Geochemical Data Workflows with pyrolite

Abstract:

With growing volumes of open data, geochemists are increasingly able to use data-driven approaches to investigate geological processes. The adoption of high-level programming languages and common frameworks (e.g. the Jupyter ecosystem) provides opportunities to address reproducibility challenges, including through the sharing of workflows, analyses and tools. However, a relative scarcity of geochemistry-focused tools adds to the need for developing appropriate skills. In this presentation we describe and demonstrate the use of ‘pyrolite’, an open source Python package which provides tools for geochemical data processing, transformation and visualization.

pyrolite provides reusable components and functions for geochemical data analysis workflows. It encourages researchers to conduct more robust statistical analysis of geochemical datasets (e.g. via log-transforms for compositional data), make use of novel algorithms, and get started with machine learning. The packaging and documentation of these tools and functions allows them to be more easily tested and verified, shared, versioned and referenced; all of these aspects contribute to the repeatability and reproducibility of associated workflows. Through pyrolite and related packages, we hope to make data-driven approaches to geochemistry readily accessible, and easier to get started with. Here we provide a series of examples of using pyrolite in data processing and exploratory data analysis workflows, illustrating how it can readily be integrated with other scientific Python packages for flexibility and customization.