PyRotate: An automated framework for the analysis, visualisation, and interpretation of tectonic rotations using palaeomagnetic data

Palaeomagnetic measurements provide the most robust, quantitative constraints on vertical axis rotations from outcrop scale blocks to entire tectonic plates. Vertical axis rotations are typically shown on maps as vectors representing local palaeodeclinations or individual rotations to a reference at the site level. When dealing with large and noisy datasets, however, this approach struggles to pick out underlying spatial patterns or finer scale regional displacements, especially when the data has variable density across regions.

These limitations are compounded in areas where several rotations events have been recorded across multiple scales or are poorly recorded in the palaeomagnetic record. As a result, both the visualisation and interpretation of complex rotation fields are hindered.

Here, we introduce PyRotate: a robust and scalable, end-to-end workflow for integrating and analyzing locality-level palaeomagnetic data from multiple sources. This framework allows users to curate and clean data through custom parameters, and to apply bootstrap techniques to determine the rotation of selected blocks and plates. Novel tools include kriging of rotations by age to simplify areas of highly variable rotation amounts and estimate interpolations between sites, giving an overlain interpolated rotation field as a visualisation tool, as well as automated outlier detection and labelling through cluster analysis. In addition, outputs are compatible with both PmagPy (Tauxe et al., 2016) and PyGPlates (Mather et al., 2024) as the basis of further analysis.

To illustrate the capabilities of PyRotate, we analyse a curated paleomagnetic dataset from Japan, and multiple-source datasets directly obtained from the Magic database. This analysis tests the ability of the workflow to deal with noisy and unevenly distributed datapoints. The tools introduced in PyRotate enable easier identification of areas needing additional sampling, synthesis of disparate datasets, and interpretation of the rotation of blocks through time.