From XRF and hyperspectral analyses to an automatic way to detect flood events on sediment cores

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

This Preprint has no visible version.

Download Preprint

Download Preprint


Rapuc William, Kévin Jacq , Anne-Lise Develle, Pierre Sabatier, Fanget Bernard, Yves Perrette, Coquin Didier, Debret Maxime, Wilhelm Bruno, Fabien Arnaud


Long-term changes in flood activity have often been reconstructed to understand their relationships to climate changes. This requires to identify flood layers according to certain characteristics (e.g. texture, geochemical composition, grain-size) and then to count them using naked-eye observation. This method is, however, time-consuming, and intrinsically characterized by a low resolution that may lead to bias and misidentifications. To overcome this limitation, high-resolution analytical approaches can be used, such as X-ray fluorescence spectroscopy (XRF), X-ray computed tomography, or hyperspectral imaging (HSI). When coupled with discriminant algorithms, HSI allows to automatically identify event layers but HSI alone cannot be used for instance to interpret the triggering process of an event layers. Here, we propose a new method of flood layers identification and counting based on the combination of both HSI and XRF core scanner analyses, applied on a Lake Bourget (French Alps) sediment sequence. We use a hyperspectral sensor from the short wave-infrared (SWIR) spectral range to create a discrimination model between event layers and continuous sedimentation. This first step allows the estimation of a classification map, with a prediction accuracy of 0.96, and then the automatic reconstruction of a reliable chronicle of event layers (including occurrence and deposit thicknesses). XRF signals are then used to discriminate flood layers among all identified event layers based on site-specific geochemical elements (in the case of Lake Bourget: Mn and Ti). This results in an automatically generated floods chronicle. Changes in flood occurrence and event thickness through time reconstructed from the automatically generated floods chronicle are in good agreement with the naked-eye-generated chronicle. In detail, differences rely on a larger number of detected flood events (i.e. increase of 9% of the number of layers detected) and a more precise layer thickness estimation, thanks to a higher resolution. Therefore, the developed methodology opens a promising avenue to increase both the efficiency (timesaving) and robustness (higher accuracy) of paleoflood reconstructions from lake sediments. Also, this methodology can be applied to identify any specific layers (e.g. varve, tephra, mass-movement turbidite, tsunami) and, thereby, it has a direct implication in e.g. paleolimnology, paleoflood hydrology, paleoseismology from sediment archives.



Earth Sciences, Physical Sciences and Mathematics, Sedimentology


Automatic flood chronicle, hyperspectral analyses, Lake sediment, XRF geochemical analysis


Published: 2020-07-29 17:31


Academic Free License (AFL) 3.0

Additional Metadata

Data Availability (Reason not available):
Data will be made available on request

Add a Comment

You must log in to post a comment.


There are no comments or no comments have been made public for this article.