/commsim

simulate communities

Primary LanguageJupyter Notebook

commsim: spatial community clustering simulation

This program simulates spatial community assembly for use in downstream community clustering and SDM-type algorithms (e.g., Maxent or Bayesian logistic regression)

  • Model assumptions:

    • assemblages are composed of species with overlapping geographic ranges.
    • a species potential niche represents its abiotic constraints (n-dimensional).
    • a species realized niche is a subset of its potential niche due to competition.
    • a species geographic range is bounded by its realized niche, but with dispersal limits.

  • Model features:

    • a geographic area matrix (n x m).
    • a complete float matrix of environmental data (nvariables x n x m)
    • a sparse binary matrix of observed species occurrences (nspecies x n x m).

  • Model goals:

    • jointly infer:
      • species occurrence covariance matrix (n**2)
      • species niche thresholds (nvar * n)
      • species geographic center of origin (n)
      • species decay parameters (n)
      • total: 2n + n * nvar + n**2
    • Use inferred parameters for posterior predictive fit to sparse occurrence data.
    • Use inferred parameters to interpolate species across unsampled grid cells.

  • Inference methods:

    • Gaussian Processes to infer covariance matrix from occurrences + env variables.
    • Logistic Regression to infer decay parameters and origin.
    • Performed together in Pymc3 Bayesian model.

  • Generative data:

    • Generate environment.
    • Get species potential niche.
    • Get species realized niche (geographic range) by comp. exclusion