11:00 AM - 11:15 AM
[MIS23-08] Compositional data analysis of sediment geochemistry: Comparison of paleoclimate and tectonic setting discrimination by old and new methods
Keywords:Compositional Data Analysis, Multivariate statistical analysis, Machine learning, Sedimentary rock
As a basic premise, statistical analysis, multivariate analysis, and artificial intelligence (AI) cannot be performed to geochemical data of sediments. This is because geochemical data are compositional data, which are not real numbers, and the sample space is not real space. Therefore, four arithmetic operations, digit “zero”, distance function and probability density functions are undefined in compositional data. Furthermore, compositional data does not meet the prerequisites for statistical analysis, multivariate analysis, and AI because of the constant-sum constraint, which is inherent in compositional data.
Fortunately, with the development of logratio analysis, simplex analysis and absolute variation method, now multivariate analysis and AI can be applied to the geochemical composition of sediments. In this presentation, the impact of multivariate analysis and AI to sediment geochemistry will be discussed with specific examples and comparison with the classical methods will be reviewed.
As an example of multivariate analysis, the weathering index W value will be introduced, which was extracted from the geochemical composition of soils. First, geochemical data of pristine igneous rocks and their weathered products (soils) were collected globally. Principal component analysis was performed on this database through the simplex analysis. As a result, “chemical variation induced by source rock type” and “chemical variation due to weathering (W value)” were mathematically isolated. Because these two types of chemical variations are overlapped in geochemical data, the conventional weathering indices could not accurately quantify the degree of weathering. On the other hand, the W value is an index that can quantify the weathering effect individually. Furthermore, when the W value was applied to modern soils around the world, it was found that the climate in which the soils were formed could be estimated. Therefore, the W index can be used to predict the paleoclimate if applied to geological records (sedimentary rocks and paleosols).
In the next example, application of AI to discriminate the tectonic setting from the geochemical composition of sediments will be demonstrated. In this study, sediment geochemical data from recent island arc (58), continental arc (89), craton (99) and collision zone (59) were collected globally, and the random forest was conducted using logratio analysis, which is a type of machine learning technique. The correct response rate of the tectonic setting was 47-67% for the conventional discrimination diagrams. In contrast, the result of random forest using logratio analysis showed a 93.4% correct response rate. This scheme can reconstruct the tectonic setting in the past from the composition of sedimentary rocks.
These examples shown in this study suggest that the application of multivariate analysis and AI to sediment geochemistry through the logratio and simplex analysis will be potentially promising method for solving various sedimentological problems.
Fortunately, with the development of logratio analysis, simplex analysis and absolute variation method, now multivariate analysis and AI can be applied to the geochemical composition of sediments. In this presentation, the impact of multivariate analysis and AI to sediment geochemistry will be discussed with specific examples and comparison with the classical methods will be reviewed.
As an example of multivariate analysis, the weathering index W value will be introduced, which was extracted from the geochemical composition of soils. First, geochemical data of pristine igneous rocks and their weathered products (soils) were collected globally. Principal component analysis was performed on this database through the simplex analysis. As a result, “chemical variation induced by source rock type” and “chemical variation due to weathering (W value)” were mathematically isolated. Because these two types of chemical variations are overlapped in geochemical data, the conventional weathering indices could not accurately quantify the degree of weathering. On the other hand, the W value is an index that can quantify the weathering effect individually. Furthermore, when the W value was applied to modern soils around the world, it was found that the climate in which the soils were formed could be estimated. Therefore, the W index can be used to predict the paleoclimate if applied to geological records (sedimentary rocks and paleosols).
In the next example, application of AI to discriminate the tectonic setting from the geochemical composition of sediments will be demonstrated. In this study, sediment geochemical data from recent island arc (58), continental arc (89), craton (99) and collision zone (59) were collected globally, and the random forest was conducted using logratio analysis, which is a type of machine learning technique. The correct response rate of the tectonic setting was 47-67% for the conventional discrimination diagrams. In contrast, the result of random forest using logratio analysis showed a 93.4% correct response rate. This scheme can reconstruct the tectonic setting in the past from the composition of sedimentary rocks.
These examples shown in this study suggest that the application of multivariate analysis and AI to sediment geochemistry through the logratio and simplex analysis will be potentially promising method for solving various sedimentological problems.