Machine-learning techniques for geochemical discrimination of 2011 Tohoku tsunami deposits

Sediment core analysis using artificial intelligence

Subsurface stratigraphic modeling is crucial for a variety of environmental, societal, and economic challenges. However, the need for specific sedimentological skills in sediment core analysis may constitute a limitation. Methods based on Machine Learning and Deep Learning can play a central role in automatizing this time-consuming procedure. In this work, using a robust dataset of high-resolution digital images from continuous sediment cores of Holocene age that reflect a wide spectrum of continental to shallow-marine depositional environments, we outline a novel deep-learning-based approach to perform automatic semantic segmentation directly on core images, leveraging the power of convolutional neural networks. To optimize the interpretation process and maximize scientific value, we use six sedimentary facies associations as target classes in lieu of ineffective classification methods based uniquely on lithology. We propose an automated model that can rapidly characterize sediment cores, allowing immediate guidance for stratigraphic correlation and subsurface reconstructions.

Machine-learning techniques for quantifying the protolith composition and mass transfer history of metabasalt

The mass transfer history of rocks provides direct evidence for fluid-rock interaction within the lithosphere and is recorded by compositional changes, especially in trace elements. The general method adopted for mass transfer analysis is to compare the composition of the protolith/precursor with that of metamorphosed/altered rocks; however, in many cases the protolith cannot be sampled. With the aim of reconstructing the mass transfer history of metabasalt, this study developed protolith reconstruction models (PRMs) for metabasalt using machine-learning algorithms. We designed models to estimate basalt trace-element concentrations from the concentrations of a few (1-9) trace elements, trained with a compositional dataset for fresh basalts, including mid-ocean ridge, ocean-island, and volcanic arc basalts. The developed PRMs were able to estimate basalt trace-element compositions (e.g., Rb, Ba, U, K, Pb, Sr, and rare-earth elements) from only four input elements with a reproducibility of ~ 0.1 log10 units (i.e., ± 25%). As a representative example, we present PRMs where the input elements are Th, Nb, Zr, and Ti, which are typically immobile during metamorphism. Case studies demonstrate the applicability of PRMs to seafloor altered basalt and metabasalt. This method enables us to analyze quantitative mass transfer in regional metamorphic rocks or alteration zones where the protolith is heterogeneous or unknown.

A Possible Tsunami Deposit Associated to the CE 1755 Lisbon Earthquake on the Western Coast of Portugal

The CE 1755 Lisbon tsunami was the largest historical tsunami to affect the Atlantic coasts of Europe and North Africa. This study presents the results obtained from the application of different sedimentological techniques (e.g., grain size, morphoscopy, microtextural analysis, geochemistry, radiocarbon dating) on sediments retrieved from the Alcabrichel River alluvial plain (of about 500 m far away from its mouth and approximatively 50 km northwest of Lisbon, Portugal). The results allowed the identification of a sandy layer that was associated with the CE 1755 tsunami. Furthermore, a new microtextural semi-quantitative classification was applied to enhance the identification of extreme marine inundation deposits. Based on sedimentological data, three different tsunami inundation phases were identified, including two inundations and a likely backwash. This innovative work offers physical evidence of the spatial presence of the CE 1755 tsunami event on the western coast of Europe. It also enables a reconstruction of tsunami inundation dynamics, with two flooding waves and an interspersed backwash.

Geochemical Discrimination of Monazite Source Rock Based on Machine Learning Techniques and Multinomial Logistic Regression Analysis

Detrital monazite geochronology has been used in provenance studies. However, there are complexities in the interpretation of age spectra due to their wide occurrence in both igneous and metamorphic rocks. We use the multinomial logistic regression (MLR) and cross-validation (CV) techniques to establish a geochemical discrimination of monazite source rocks. The elemental abundance-based geochemical discrimination was tested by selecting 16 elements from granitic and metamorphic rocks. The MLR technique revealed that light rare earth elements (REEs), Eu, and some heavy REEs are important discriminators that reflect elemental fractionation during magmatism and/or metamorphism. The best model yielded a discrimination rate of ~97%, and the CV method validated this approach. We applied the discrimination model to detrital monazites from African rivers. The detrital monazites were mostly classified as granitic and of garnet-bearing metamorphic origins; however, their proportion of metamorphic origin was smaller than the proportion that was obtained by using the elemental-ratio-based discrimination proposed by Itano et al. in Chemical Geology (2018). Considering the occurrence of metamorphic rocks in the hinterlands and the different age spectra between monazite and zircon in the same rivers, a ratio-based discrimination would be more reliable. Nevertheless, our study demonstrates the advantages of machine-learning-based approaches for the quantitative discrimination of monazite.

Development of a predictive model for lead, cadmium and fluorine soil-water partition coefficients using sparse multiple linear regression analysis

In this study, we applied sparse multiple linear regression (SMLR) analysis to clarify the relationships between soil properties and adsorption characteristics for a range of soils across Japan and identify easily-obtained physical and chemical soil properties that could be used to predict K and n values of cadmium, lead and fluorine. A model was first constructed that can easily predict the K and n values from nine soil parameters (pH, cation exchange capacity, specific surface area, total carbon, soil organic matter from loss on ignition and water holding capacity, the ratio of sand, silt and clay). The K and n values of cadmium, lead and fluorine of 17 soil samples were used to verify the SMLR models by the root mean square error values obtained from 512 combinations of soil parameters. The SMLR analysis indicated that fluorine adsorption to soil may be associated with organic matter, whereas cadmium or lead adsorption to soil is more likely to be influenced by soil pH, IL. We found that an accurate K value can be predicted from more than three soil parameters for most soils. Approximately 65% of the predicted values were between 33 and 300% of their measured values for the K value; 76% of the predicted values were within ±30% of their measured values for the n value. Our findings suggest that adsorption properties of lead, cadmium and fluorine to soil can be predicted from the soil physical and chemical properties using the presented models.