sdmTMB provides an R package for fitting spatial and spatiotemporal generalized linear mixed-effects models (GLMMs), particularly for geostatistical data and species distribution modeling. It offers a fast, flexible, and user-friendly interface leveraging efficient computational methods.

How It Works

The package utilizes Template Model Builder (TMB) for fitting and fmesher for constructing Gaussian Markov Random Fields via the Stochastic Partial Differential Equation (SPDE) approach. This enables computationally efficient modeling of spatial/spatiotemporal correlation, extending GLMMs with features like random fields, time-varying, and spatially varying coefficients.

Quick Start & Requirements

Installation is via CRAN (install.packages("sdmTMB")) or development version (pak::pak("sdmTMB/sdmTMB")). A C++ compiler is required; an optimized BLAS library is recommended for performance. Extensive documentation and vignettes cover basic/advanced usage, index standardization, and Bayesian MCMC.

Highlighted Details

sdmTMB supports diverse response families (Tweedie, nbinom, delta/hurdle). Advanced features include time-varying coefficients (random walks), spatially varying coefficients (SVCs) for heterogeneity, breakpoint/threshold effects, and random intercepts. It integrates with Stan via tmbstan for Bayesian inference, allows custom meshes (including barrier meshes), and provides utilities for data simulation, prediction uncertainty, cross-validation, and ecological index generation.

Maintenance & Community

User support is via a discussion board; bugs/features via an issue tracker. Past workshops have provided materials