physioSDM

Code and experimentation for the inclusion of physiological data in species distribution models.

Experiment

We're trying to determine whether including physiological information as a prior in species distribution models can improve or change future projections. To do this we're building a physiologically-informed SDM for a list of species for which we have physiological data, and comparing the results to a non-physiological gaussian-process based SDM and MaxEnt. We compare the models in both their ability to recover current distributions and in how they project into the future.

Architecture.

This is an experimentation suite built in R upon the following packages, all of which are dependencies:

• dismo
• GrAF
• raster
• dplyr, pryr and purrr (from the tidy verse)
• ROCR
• uuid
• futile.logger
• jsonlite
• rgbif
• caret

The code is architected such that it takes as input a .csv file containing various physiological parameters for a list of species, where each row is a species and each column represents some physiological quantity. The code also accepts a JSON file of parameters which determine many runtime behaviors. The same experimentation is performed

All experimentation code is located in the /r directory. The following is an explanation of the files therein:

• main.r: contains the main experimentation loop and code to process input parameters.
• occurrences.R: responsible for fetching occurrence information from GBIF.
• climate.R: contains code for extracting climatological information from already-downloaded climate data which is stored locally.
• sdm.r: contains code for constructing all of the models used in experimentation.
• priors.R: contains code for constructing physiological priors for the gaussian process model.

Installation.

1. Clone this repository to your local computer.
2. Acquire the datasets for your experimentation. More information on how to get our specific data will come with publication.
3. Edit parameters.json file to reflect these data and your chosen parameters.
4. Install dependencies.
   1. apt-get install libgeos-dev libgdal-dev default-jdk. Note that depending on your system you may need to install additional libraries. You'll know this after the next step; R will tell you what's missing.
   2. Run install.packages(c('dismo', 'GRaF', 'raster', 'tidyverse', 'ROCR', 'uuid', 'futile.logger', 'jsonlite', 'rgbif', 'caret', 'rgdal', 'rJava')) in R.
5. Attempt dry run by changing into r/ directory and running Rscript main.r. You will likely need to download maxent.jar and install it. The program will instruct you on how to do this.

Running Experiments

more later.