Unified geostatistical modeling for data fusion and spatial heteroskedasticity with R package ramps

Abstract

This article illustrates usage of the ramps R package, which implements the reparameterized and marginalized posterior sampling (RAMPS) algorithm for complex Bayesian geostatistical models. The RAMPS methodology allows joint modeling of areal and pointsource data arising from the same underlying spatial process. A reparametrization of variance parameters facilitates slice sampling based on simplexes, which can be useful in general when multiple variances are present. Prediction at arbitrary points can be made, which is critical in applications where maps are needed. Our implementation takes advantage of sparse matrix operations in the Matrix package and can provide substantial savings in computing time for large datasets. A user-friendly interface, similar to the nlme mixed effects models package, enables users to analyze datasets with little programming effort. Support is provided for numerous spatial and spatiotemporal correlation structures, userdefined correlation structures, and non-spatial random effects. The package features are illustrated via a synthetic dataset of spatially correlated observation distributed across the state of Iowa, USA.