Skip to contents

Allows a full Bayesian analysis of linear and additive models using Integrated Nested Laplace approximation. Engine has been largely superceded by the engine_inlabru package and users are advised to us this one, unless specific options are required.


  optional_mesh = NULL,
  optional_projstk = NULL,
  max.edge = NULL,
  offset = NULL,
  cutoff = NULL,
  proj_stepsize = NULL,
  timeout = NULL,
  strategy = "auto",
  int.strategy = "eb",
  barrier = FALSE,
  type = "response",
  area = "gpc2",
  nonconvex.bdry = FALSE,
  nonconvex.convex = -0.15,
  nonconvex.concave = -0.05,
  nonconvex.res = 40,



distribution() (i.e. BiodiversityDistribution) object.


A directly supplied "INLA" mesh (Default: NULL)


A directly supplied projection stack. Useful if projection stack is identical for multiple species (Default: NULL)


The largest allowed triangle edge length, must be in the same scale units as the coordinates. Default is an educated guess (Default: NULL).


interpreted as a numeric factor relative to the approximate data diameter. Default is an educated guess (Default: NULL).


The minimum allowed distance between points on the mesh. Default is an educated guess (Default: NULL).


The stepsize in coordinate units between cells of the projection grid (Default: NULL).


Specify a timeout for INLA models in sec. Afterwards it passed.


Which approximation to use for the joint posterior. Options are "auto" ("default"), "adaptative", "gaussian", "simplified.laplace" & "laplace".


Integration strategy. Options are "auto","grid", "eb" ("default") & "ccd". See also


Should a barrier model be added to the model?


The mode used for creating posterior predictions. Either summarizing the linear "predictor" or "response" (Default: "response").


Accepts a character denoting the type of area calculation to be done on the mesh (Default: 'gpc2').


Create a non-convex boundary hulls instead (Default: FALSE) Not yet implemented


Non-convex minimal extension radius for convex curvature Not yet implemented


Non-convex minimal extension radius for concave curvature Not yet implemented


Computation resolution for nonconvex.hulls Not yet implemented


Other options.


An engine.


All INLA engines require the specification of a mesh that needs to be provided to the "optional_mesh" parameter. Otherwise the mesh will be created based on best guesses of the data spread. A good mesh needs to have triangles as regular as possible in size and shape: equilateral.

* "max.edge": The largest allowed triangle edge length, must be in the same scale units as the coordinates Lower bounds affect the density of triangles * "offset": The automatic extension distance of the mesh If positive: same scale units. If negative, interpreted as a factor relative to the approximate data diameter i.e., a value of -0.10 will add a 10% of the data diameter as outer extension. * "cutoff": The minimum allowed distance between points, it means that points at a closer distance than the supplied value are replaced by a single vertex. it is critical when there are some points very close to each other, either for point locations or in the domain boundary. * "proj_stepsize": The stepsize for spatial predictions, which affects the spatial grain of any outputs created.

Priors can be set via INLAPrior.


How INLA Meshes are generated, substantially influences prediction outcomes. See Dambly et al. (2023).


  • Havard Rue, Sara Martino, and Nicholas Chopin (2009), Approximate Bayesian Inference for Latent Gaussian Models Using Integrated Nested Laplace Approximations (with discussion), Journal of the Royal Statistical Society B, 71, 319-392.

  • Finn Lindgren, Havard Rue, and Johan Lindstrom (2011). An Explicit Link Between Gaussian Fields and Gaussian Markov Random Fields: The Stochastic Partial Differential Equation Approach (with discussion), Journal of the Royal Statistical Society B, 73(4), 423-498.

  • Simpson, Daniel, Janine B. Illian, S. H. Sørbye, and Håvard Rue. 2016. “Going Off Grid: Computationally Efficient Inference for Log-Gaussian Cox Processes.” Biometrika 1 (103): 49–70.

  • Dambly, L. I., Isaac, N. J., Jones, K. E., Boughey, K. L., & O'Hara, R. B. (2023). Integrated species distribution models fitted in INLA are sensitive to mesh parameterisation. Ecography, e06391.


if (FALSE) {
# Add INLA as an engine (with a custom mesh)
x <- distribution(background) |> engine_inla(mesh = my_mesh)