hmsc_pipeline

R utility functions to define, fit and evaluate HMSC models. These functions build on top of HMSC framework to provide a full Species Distribution Modelling pipeline. See demo.pdf for an illustration of these tools.

Requirements

• Hmsc
• ggplot2
• raster
• vioplot
• docstring

Available features

WORK IN PROGRESS

1. Dataset sanity checks --> read_data.R
2. Create HMSC model instance. Different model are available, including spatial models or Hurdle modelling approach. Possibility to perform predictor selection (spike_slab_jointly, spike_slab_separately, pca, rrr). --> define_model.R
3. Fit the model by sampling the posterior with block-conditional Gibbs MCMC sampler. --> fit_model.R
4. Evaluage MCMC convergence. By looking at the effective size of the posterior sample: beta-parameters (species niches) and V-parameters (variation in species niches). By plotting the Gelman diagnostics, i.e. the Potential scale reduction factors. --> evaluate_convergence.R
5. Compute model fit, both explanatory and predictive powers. R2, TjurR2, RMSE, AUC are computed. A 5-fold cross validation is used for the predictive power. --> compute_model_fit.R
6. XX make_predictions.R
7. Create a grid of environmental predictors by extracting data from a larger raster --> create_grid.R
8. XX make_spatial_predictions.R
9. XX show_env_niche.R
10. XX show_parameter_estimates.R
11. XX show_models_fit.R

Help

To get functions' description, you can use the following:

?make_spatial_predictions

which provides the function description:

Description
This function load a model and make some predictions given some predictors and spatial coordinates. The predictions are then saved under fname_out.

Usage
make_spatial_predictions(S, X, model_path, fname_out)

Arguments
S	: Study design dataframe containing "Longitude" and "Latitude"

X: Predictors dataframe

model_path: Path towards the fitted models

fname_out: Predictions of each model will be saved under this filename

Note:
TODO: Adapt to non spatial model

References

See HMSC Github documentation section.

Acknowledgements

• N. Hill for hmsc_CVpred_parallel and env_niches functions.
• HMSC team for the fantastic 2020 HMSC online workshop.