spatialRF Easy Spatial Modeling with Random Forest

Introduction

The package spatialRF trains explanatory spatial regression models by combining Random Forest with spatial predictors that help the model reduce the spatial autocorrelation of the residuals and return honest variable importance scores.

The package is designed to minimize the code required to fit a spatial model from a training dataset, the names of the response and the predictors, and a distance matrix, as shown in the mock-up call below.

m <- spatialRF::rf_spatial(
  data = df,
  dependent.variable.name = "response",
  predictor.variable.names = c("pred1", "pred2", ..., "predN"),
  distance.matrix = distance_matrix
  )

spatialRF uses the fast and efficient ranger package under the hood (Wright and Ziegler 2017), so please, cite the ranger package when using spatialRF!

This package also provides tools to identify potentially interesting variable interactions, tune random forest hyperparameters, assess model performance on spatially independent data folds, and examine the resulting models via importance plots, response curves, and response surfaces.

However, there are several things this package cannot do:

• Predict model results over raster data.

• Predict a model over a different place with a different spatial structure.

• Work with “big data”, whatever that means.

• Imputation or extrapolation (it can be done, but models based on spatial predictors are hardly transferable).

• Take temporal autocorrelation into account.

Data requirements

spatialRF is designed to work with:

• Data frames: tibbles and sf dataframes are not fully supported.
• Continuous responses: binomial responses are partially supported (however, spatial models hardly work with them), but categorical and factor responses are not.
• Small data: spatial modelling operations are RAM-hungry, the maximum number of rows must be somewhere between 5000 and 10000, but in the end it depends on the available RAM.
• Must be free of NA.
• Columns cannot have near-zero variance. You don’t want near-zero variance columns in your data anyway.

Citation

If you find spatialRF useful, please cite the ranger package as well.

Marvin N. Wright, Andreas Ziegler (2017). ranger: A Fast Implementation of Random Forests for High Dimensional Data in C++ and R. Journal of Statistical Software, 77(1), 1-17. doi:10.18637/jss.v077.i01

Blas M. Benito (2025). spatialRF: Easy Spatial Regression with Random Forest. R package version 1.1.5. doi: 10.5281/zenodo.4745207. url: https://blasbenito.github.io/spatialRF/

Install

The version 1.1.5 can be installed from CRAN:

install.packages("spatialRF")

The package can also be installed from GitHub as follows. There are several branches in the repository:

• main: latest stable version (1.1.5 currently).
• development: development version, usually very broken.
• v.1.0.9 to v.1.1.4: archived versions.

remotes::install_github(
  repo = "blasbenito/spatialRF",
  ref = "main",
  force = TRUE,
  quiet = TRUE
  )

Getting Started

This README provides a quick introduction to spatialRF. For detailed tutorials, see:

Tutorials

• Non-Spatial Random Forest Models: Learn how to fit and interpret standard random forest models using rf(). Covers data exploration, variable interactions, model evaluation, importance scores, response curves, spatial cross-validation, and prediction. Start here if you’re new to spatialRF.

• Spatial Random Forest Models: Discover how to address spatial autocorrelation using rf_spatial(). Demonstrates spatial predictor generation with Moran’s Eigenvector Maps, optimization, hyperparameter tuning, and model comparison.