CoDa-session at IAMG2023 (August 5-12)
on behalf of J. Mckinley and K. Hron (CoDa-session Chairs) here an announcement:
Dear Members of CoDa Association,
we would like to invite you to participate in compositional session at the IAMG 2023 conference which takes place in Trondheim, Norway, on August 5-12, 2023 (https://www.iamgconferences.org/iamg2023/). CoDa Association has a long-term cooperation with International Association for Mathematical Geosciences (formalized also by Memorandum of Understanding) and compositional data analysis was supported by IAMG from its very early stages. Abstract of the session is below, we would like to encourage you to participate!
Jennifer McKinley, Karel Hron
- Session name: Practical Aspects and Applied Studies in Geochemical Exploration and Mapping with Compositional Logratio Techniques
- Abstract: This session will offer a practical forum of discussion for people concerned with the statistical treatment, modelling and interpolation of compositional data in geochemical applications, particularly focused on geochemical exploration and mapping. The session will invite papers related to the problems of geochemical mapping, and the compositional topics. Geochemical survey data have both compositional and spatial character. Thus, there are two spaces for these data; the geochemical and the geospatial domains. The multivariate relationships of the elements typically reveal patterns associated with processes determined by mineralogy, weathering, comminution and mass transport. Typically, these processes display a continuous spatial character in both 2 and 3 dimensions and time. Many methods are available for process discovery such as principal components or minimum/maximum autocorrelation factor analysis. The evaluation of geochemical data can be carried out using the approach of process discovery and process validation. In the context of Process Discovery, different methods use different measures of association. Some multi-element associations are common amongst all of the metrics, whilst other measures are unique to the metric. These metrics use both linear and non-linear mappings to define the associations. Typically, the approach is unsupervised with the intent to define multi-element associations that reflect mineralogy associated with rock types. When the procedure of process discovery identifies patterns that represent geologic phenomena, these patterns can be tested using classification methods and is termed process validation. Methods such as discriminant analysis operate in linear and non-linear modes to carry out prediction. Neural networks, logistic regression, random forests, support vector machines use various measures of association for classification using heuristics such as decision trees and multivariate measures of distance. This approach is usually described as a supervised way of testing the classes defined through process discovery or by pre-defined categorical information assigned with each geochemical composition. This session invites contributions where multivariate geoscience data (geochemistry, geophysics, geodetic, geologic) can be processed within a compositional framework that includes an evaluation of the data, or results of processing the data, in a geospatial context.