Movatterモバイル変換

Small SampleSize Species Distribution Modeling

The S4DM R package

This repository contains an R package that implements SpeciesDistribution Modeling methods which work even when there are relativelyfew occurrence records (as is the case for poorly-sample orrange-restricted species). These methods were primarily developed by theDrake lab, and include three types of methods: 1) Plug-and-play models,2) environmental-range models, and 3) density-ratio models. Most of theimportant functions in this package are wrappers around existingfunctions that handle density estimation or density-ratio estimation.Much of this code was created by modifying existing code athttps://github.com/DrakeLab/PlugNPlay in order to make functions moremodular and extensible.

How it works

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

1. The highest-level functions aremake_range_map andevaluate_range_map, which are wrappers for…
2. The next-highest-level functions,fit_plug_and_play,fit_density_ratio,project_plug_and_play, andproject_density_ratio, which are wrappers for …
3. Internal modules such aspnp_kde ordr_ulsif. These modules both model the environmentalcovariates and predict values at environmental covariates from fittedmodels. These modules are largely wrappers around existing functions forfitting 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 isdesigned to allow for the easy addition of new methods by adding small,self-contained modules. The highest-level functions are intended onlyfor quick-and-dirty analyses or quick visualizations. We recommend thatusers focus on the “fit” and “project” functions for work intended forpublication.

What is Plug-and-Play?

In general usage, the term plug-and-play (PNP) refers to software orhardware that can be connected without any additional setup orconfiguration. 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.