Paleomagnetic variations along the Southern Coast of Brazil from 46.06-5.36 ka BP using a Gaussian Process Technique

Understanding geomagnetic variations observed at the Earth’s surface requires knowledge of processes occurring in the Earth’s interior that are associated with the geodynamo. Because these processes occur at inaccessible depths, numerical models are used to simulate the expected conditions within these regions. On the other hand, data-based reconstructions provide constraints in terms of the field morphology at the Earth’s surface and at the core-mantle boundary. Such models are constrained by direct measurements for recent periods and by indirect data for older intervals. For the time span covering the last few centuries to approximately 50 ka, the available paleomagnetic data are unevenly distributed in both space and time, with a strong bias toward the Northern Hemisphere and more recent periods. These data limitations influence the reliability of geomagnetic model predictions, particularly in poorly sampled regions such as South America. In recent years, new paleomagnetic data have been obtained from ocean-floor sediments off the southern coast of Brazil. However, integrating paleomagnetic data from geographically close locations, together with reliable age and sedimentation rate models, remains a major challenge and limits the reproducibility of records from nearby sites. To address this issue, we applied a machine-learning approach based on Gaussian process regression to construct inclination and relative paleointensity curves using data from three sediment cores collected in this region (29°S, 47°W). This method allows for the estimation of a modeled curve that integrates data from the three cores, which have irregular time intervals between samples, and provides associated uncertainties. The resulting inclination and relative paleointensity curves were compared with predictions from the GGF100k geomagnetic model. Overall, the results show good agreement with model predictions, although some discrepancies are observed in both inclination and intensity over the studied interval. These findings demonstrate that Gaussian process regression is a robust and effective tool for integrating paleomagnetic data from oceanic sediment cores.