Plug-and-Play Species Distribution Modelling

What is Plug-and-Play?

In general usage, the term plug-and-play (PNP) refers to software or hardware that can be connected without any additional setup or configuration. In the context of species distribution models, plug-and-play is a framework developed by Drake and Richards (https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecs2.2373) that recognizes that species distribution models can be constructed by "plugging in" any methods that can estimate density functions or the ratio of two density functions.

The PBSDM R package

This repo contains a draft R package to implement PNP SDMs, with an emphasis on presence-only and presence-background models (hence the current working name, PBSDM for presence-background species distribution models), as well as density-ratio models. Most of the important functions in this package are wrappers around existing functions that handle density estimation or density-ratio estimation. Much of this code was created by modifying existing code at https://github.com/DrakeLab/PlugNPlay in order to make functions more modular and extensible.

How it works

The package is build on a hierarchy of modular functions, each of which calls on lower-level functions:

1. The highest-level functions are make_range_map and evaluate_range_map, which are wrappers for...
2. The next-highest-level functions, fit_plug_and_play, fit_density_ratio, project_plug_and_play, and project_density_ratio, which are wrappers for ...
3. Internal modules such as pnp_kde or dr_ulsif. These modules both model the environmental covariates and predict values at environmental covariates from fitted models. These modules are largely wrappers around existing functions for fitting density functions or density-ratios. Modules beginning with "pnp_" pertain to density functions while models beginning with "dr_" pertain to density ratio functions.

This hierarchical structure built on low-level internal modules is designed to allow for the easy addition of new methods by adding small, self-contained modules.