geoGAM: Select Sparse Geoadditive Models for Spatial Prediction

A model building procedure to build parsimonious geoadditive model from a large number of covariates. Continuous, binary and ordered categorical responses are supported. The model building is based on component wise gradient boosting with linear effects, smoothing splines and a smooth spatial surface to model spatial autocorrelation. The resulting covariate set after gradient boosting is further reduced through backward elimination and aggregation of factor levels. The package provides a model based bootstrap method to simulate prediction intervals for point predictions. A test data set of a soil mapping case study in Berne (Switzerland) is provided. Nussbaum, M., Walthert, L., Fraefel, M., Greiner, L., and Papritz, A. (2017) <doi:10.5194/soil-3-191-2017>.

install.packages("geoGAM")

Try the geoGAM package in your browser