Mapping depth to Pleistocene sand with Bayesian generalized linear geostatistical models

Spatial soil applications frequently involve binomial variables. If relevant environmental covariates are available, using a Bayesian generalized linear model (BGLM) might be solution for mapping such discrete properties. The geostatistical extension, (BGLGM), adds spatial dependence and is thus potentially better equipped. objective of this work was to evaluate whether it pays off extend from BGLM BGLGM binary properties, evaluated in terms prediction accuracy modelling complexity. As motivating example, we mapped the presence/absence Pleistocene sand layer within 120 cm land surface Dutch province Flevoland, implementation R-package geoRglm. We found that yields considerably statistical validation metrics compared BGLM, especially with – as our case large (n = 1,000) observation sample few available. Also, inferred posterior parameters enable quantification relationships. However, calibrating applying quite demanding respect minimal required size, tuning algorithm, computational costs. replaced manual by an automated algorithm (which eases implementing applications) composition delivers meaningful results 50 h calculation time. With gained insights shared scripts practitioners researchers can their specific cases if feasible extra gain worth effort. Highlights Does adding correlation GLM variable pay off? aim make hierarchical accessible pedometricians. Most models well when enough observations provided, even without covariates. Including sometimes effort