README.md

In barnabywalker/tidyassessments: Model Wrappers for Automated Conservation Assessments

tidyassessments

This is a package to make some models used for occurrence-based automated extinction risk assessments work with the tidymodels workflow.

This package was developed to help the analysis for a paper developing guidelines for automated conservation assessments. To add new models to tidymodels, it’s best to package them up so they work in parallel across platforms. This is a package to make some models used for occurrence-based automated extinction risk assessments work with the tidymodels workflow.

So far, the models implemented produce binary classifications of species as “threatened” or “not threatened”.

Installation

You can install the development version of tidyassessments like so:

remotes::install_github("barnabywalker/tidyassessments")

Example

You can set up a tidymodels model specification like this:

library(tidyassessments)

spec <- rule_based(engine="eoo")

How is extinction risk assessed?

Species can be put at risk of extinction from all manner of things based on intrinsic properties of the species, pressure from human activity, threats from the natural world, or any number of interactions between these things. Having a way to assess and compare which species are most at risk helps to focus policy and funding to help slow or prevent biodiversity loss.

The most widely recognised system for assessing extinction risk is the IUCN Red List of Threatened Species. Adding species to the Red List involves gathering all available data about a species and the threats it faces. An assessor will then apply one or more quantitative criteria to this data and determine which Red List category the species falls under. The Red List categories are Least Concern (LC), Near Threatened (NT), Vulnerable (V), Endangered (EN), Critically Endangered (CR), Extinct in the Wild (EW), and Extinct (EX). When there is not enough data to sort a species into one of these categories, it is designated as Data Deficient (DD).

Why automate assessments?

Gathering all this data for a single species takes a lot of time. On top of this, each assessment needs to be reviewed by experts before it is published. Ideally, all species should be assessed for the Red List, and these assessments should be periodically updated. However, only a fraction of some groups has been assessed. For example, \~15% of the c. 400,000 plant species are documented on the Red List.

Researchers have proposed many different approaches and tools to automate the assessment process. These focus on a few different goals and use a broad range of data as inputs, but all are intended to speed up or reduce the burden of carrying out assessments. Cavazali et al. (2021) have a good review of automated assessment methods and how the gap between researching and using these methods can be closed.

That review splits automated assessment methods into two distinct categories: criteria-explicit and category-predictive. Criteria-explicit methods focus on automating the calculation of parameters that can be used to directly apply one or more of the IUCN Red List criteria. Category-predictive methods try to predict the Red List category of a species from correlates of extinction risk. While criteria-explicit methods apply the Red List criteria as classification rules, category-predictive methods often use statistical or machine learning to learn classification rules.

How can tidymodels help?

Despite these differences, developing or trying out an AA method generally follows a similar set of steps:

1. Get a list of accepted species together for a particular taxonomic group or area of interest (e.g. all species in the plant genus Myrcia).
2. Find all existing assessments for your species list.
3. Gather together all relevant data (e.g. all occurrence records for your species from GBIF).
4. Carry out any pre-processing/cleaning on your input data (e.g. remove occurrence records with dubious coordinates).
5. Calculate the parameters/predictors necessary for your chosen AA method.
6. Evaluate the performance of your chosen method on species that have already been assessed. This step also involves hyperparameter tuning (if needed) and training the model for machine-learning approaches.
7. Use your chosen method to predict the status of unassessed species.

tidymodels can help with steps 6 and 7: tuning, training, evaluating, and applying AA methods. It does this by providing a consistent interface to different machine learning and statistical models. We think that reduces the barrier to using these models while ensuring good practice. We’ve made this package to provide interfaces to some criteria-explicit and category-predictive automated assessment methods.

Our paper “Evidence-based guidelines for automated conservation assessments of plant species” might be interesting if you’d like more information about automated assessment methods.

We’ve added a handful of tutorials if you’d like more information about using tidymodels for automated assessments.

barnabywalker/tidyassessments documentation built on April 8, 2022, 12:32 p.m.