/bloomcpf

Primary LanguageR

BLOOM: Bayesian caLibration Of pOllination Model

blOOm is a set of R functions for the simulation and calibration of a pollination model based on central-place foraging (CPF) theory.

Introduction

Here we present the code to run and calibrate the CPF model using Approximate Baeysian Computation. The statistical model assumes that bee abundances are Poisson-distributed, with the logarithm of the Poisson intensity being a linear function of the logarithm of the CPF predictions.

In the right-hand side equations, \nu_i(\theta,\mathcal{M}_{jk}) represents the output from the CPF model, \theta is the set of parameters from the CPF model and \mathcal{M}_{jk} is the rasterized map for year j and period k.

blOOm proposes a Bayesian approach to perform parameter estimation, using approximate bayesian computation (ABC) since the likelihood is intractable.

Contents

The repository has the following structure:

.
├── analyses
│   └── results
├── data
├── man
├── R

Analyses

In the data folder, several data are provided: example files needed as inputs for the CPF model, as well as prior distributions for the parameters. Summary statistics are provided both on simulated and real data. An object generated from the blOOm framework, of class simulationObj and containing all the required information about the data is also available. Note that only true summary statistics are provided, but that true raw data are not provided.

In the analysis folder, you can find R scripts for data analysis and calibration. This is also where the result of the analyses are saved (in a subfolder named results).

File createABCtable.R allows to generate several samples from the model, according to priors which must be specified. This part of the analysis is time-consuming. Therefore, example files are provided in the results folder to illustrate the analyses.

Once the ABC table is computed, summary statistics can be computed. The corresponding scripts are computeSumStats.R for ABC tables, and computeSumStatsObsData.R for observed data.

Once the summary statistics are computed for both the observed data and the ABC table, the analyses can be performed.

  • runABC_simuData.R runs different ABC approaches on simulated data, using a subset of the simulations as references.
  • runABC_realData.R runs different ABC approaches on real data.
  • analyzeResults.R and analyzeResults_realData.R are used to produce graphs and compute relative absolute errors
  • posteriorPredictiveCheck.R generate samples from the posterior predictive distribution