CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Mapping soil texture in the basque country using compositional data analysis: A Bayesian geo-additive approach Authors:  Joaquin Martinez-Minaya - Basque Center For Applied Mathematics (BCAM) (Spain) [presenting]
Dae-Jin Lee - BCAM - Basque Center for Applied Mathematics (Spain)
Lore Zumeta-Olaskoaga - BCAM (Spain)
Abstract: Compositional data (CoDa), consisting of proportions or percentages of disjoint categories adding to one, play an important role in many fields such as ecology, geology, etc. The two most popular families to deal with them are the Dirichlet and the logistic-normal using Aitchison geometry. Recent developments on the simplex geometry allow us to express the regression model in terms of coordinates and estimate its coefficients. Once the model is projected in the real space, we can employ a multivariate Gaussian regression to deal with it. In order to allow for more flexibility to relate coordinates with covariates or spatial components, penalized splines smoothing can be employed. One of the main goals is to show how to fit a Geo-additive compositional regression from the Bayesian perspective using the software brms. Another key question when we deal with CoDa is how to perform the model validation process. We propose two Bayesian CoDa regression measures to assess the goodness of fit of the model. This method was applied in mapping soil texture distribution at a finer scale, based on 2279 soil samples surveyed in the Basque Country between 2010-2018. We considered different covariates such as elevation or slope, as well as geological information. The proposed methodology showed a good overall performance in different scenarios providing high-resolution maps at a fine-scale for the agricultural sector in the Basque Country