oceanhackweek/tutorials_marine_sdm

Tutorial for marine SDMs

Marine Species Distribution Model (SDM) Tutorial

Overview

This tutorial was developed during OceanHackWeek2023 to provide a simple workflow to developing a marine Species Distribution Model (SDM) using R programming. To see the OHW23 project at the end of OHW23, go to the ohw23_proj branch or see the ohw23_proj release.

Background

Species Distribution Modelling (SDM) also known as niche/environmental/ecological modelling uses an algorithm to predict the distribution of a species across space and time using environmental data. An understanding of the relationship between the species of interest and the physical environment they occupy will inform the selection of relevant environmental factors that will be included in the model.

Biotic information is also needed by SDMs and at the very least locations of individuals are needed. Abundance or densities can also be used as inputs, but are not compulsory. It is worth noting that absences, that is, the locations where individuals of a species are NOT present is just as important because it provides information about the environmental conditions where individuals are not usually sighted. Often absences are not recorded in biological data, but we can use background points (also known as pseudo-absences), which provide information about the full range of environmental conditions available for the species interest in our study area.

For a review of the performance of different SDM algorithms, see the following publications:

• Valavi, Guillera-Arroita, Lahoz-Monfort, Elith (2021). Predictive performance of presence-only species distribution models: a benchmark study with reproducible code. DOI: 10.1002/ecm.1486

• Elith et al (2006). Novel methods improve prediction of species’ distributions from occurrence data. DOI: 10.1111/j.2006.0906-7590.04596.x

For a discussion on the impact of background data on SDMs see: Phillips et al (2009). Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. DOI: 10.1890/07-2153.1. For a background sample generation refer to work by Valavi.

Datasets used in the tutorial

Biological Data

Our area of interest is the Indian Ocean, where four species of sea turtles have been reported to occupy this area:

• Loggerhead, Caretta caretta
• Green, Chelonia mydas
• Olive Ridley, Lepidochelys olivacea
• Hawksbill, Eretmochelys imbricata

For this tutorial, we will focus on predicting the areas occupied by loggerhead sea turtles. To do this, we will use presence-only data from 2000 until present, which have been sourced from the Ocean Biodiversity Information System (OBIS) via the robis package.

Environmental Data

This tutorial focuses on regions in the northern Indian Sea, specifically the western Arabian Sea, Persian Gulf, Gulf of Oman, Gulf of Aden and Red Sea. Environmental predictor variables were sourced via the sdmpredictors R package. The package give access to the https://bio-oracle.org/ and http://marspec.org/ high-resolution layers of various marine variables. Note these variables are location specific but not time specific: they are average values over time periods.

Workflow/Roadmap

This tutorial is based on the notes by Ben Tupper (Bigelow Lab, Maine), and highlights modeling presence-only data via maxnet R package.

Tutorial roadmap

1. Presence Data -- obtain Loggerhead sea turtle (C. caretta) presence data from OBIS via robis
2. Background Points -- shows two methods to create random background points within our area of interest
3. Environmental Data -- obtain environmental predictors of interest using SDMpredictors
4. Model -- run species distribution model and predict using maxnet
5. Data Visualizations

References

• Bosch S, Fernandez S (2022). sdmpredictors: Species Distribution Modelling Predictor Datasets. R package version 0.2.14, http://lifewatch.github.io/sdmpredictors/.

• OBIS (2023) Ocean Biodiversity Information System. Intergovernmental Oceanographic Commission of UNESCO. www.obis.org. Accessed: 2023-08-08.

• Steven J. Phillips, Miroslav Dudík, Robert E. Schapire. [Internet] Maxent software for modeling species niches and distributions (Version 3.4.1). Available from url: http://biodiversityinformatics.amnh.org/open_source/maxent/. Accessed on 2023-08-10.

Tutorial developers

• Eli Holmes: Research Fisheries Biologist, Northwest Fisheries Science Center, NOAA Fisheries.
• Catherine Courtier:
• Mackenzie Fiss: Fifth-year PhD student at Northeastern University studying carbon cycling and microbial interactions in salt marshes.
• Denisse Fierro Arcos: PhD candidate at the Institute for Marine and Antarctic Studies (IMAS) and Data Officer at the Integrated Marine Observing System (IMOS)
• Paulo Freire: PhD candidate at the University of North Carolina at Charlotte (UNCC) studying marine microbial ecology.
• Jade Hong: Recently finished ungraduate studies majoring Biology and Marine Science at Boston University.
• Tylar Murray: USF IMaRS Software Engineer - code whisperer, data viz enthusiast, scientific generalist, compulsive overengineerer, & UX PhD
• Caitlin O'Brien: Research Scientist, Columbia Basin Research, School of Aquatic Fishery and Sciences, University of Washington
• Mary Solokas: John A. Knauss Marine Policy Fellow, National Oceanic and Atmospheric Administration
• Laura Tsang: Recent Master's graduate from Northeastern University
• Ben Tupper: Senior Research Associate at Bigelow Laboratory for Ocean Science

---

Who is this tutorial intended for?

Some experience programming in R is needed to make the most of this tutorial. To run this tutorial make sure you clone this repository into your local machine by creating a new project that uses version control (git).

The tutorial content was developed in a R version 4.2.2 for Linux.

Additional resources

If you need additional support with R programming, you can check the following resources:

• Ben Best's labs on SDMs
• R for Data Science - 2nd edition by Wickham, Çetinkaya-Rundel and Grolemund.
• Data analysis and visualisation in R for ecologists
• For information on how to use git and GitHub with R, Happy Git and GitHub for the useR by Jenny Bryan is a great resource.

tutorials_marine_sdm's Issues

Created a little map with the dataset that Tylar posted

Is there a way to incorporate the environmental data from the dataset? It has bathymetry, sst, and sss. Or is it easier to use the method that Dr. Eli used in https://eeholmes.github.io/Species-Dist-Modeling---Trillium/variables.html#climatic-data?

Try using sdmpredictors and Ben Tupper's sample code

https://github.com/lifewatch/sdmpredictors
Ben Tupper gave a link to some helpful scripts that we might use. Does not use rasters for the env data like dismo requires for its model fitting
https://github.com/BigelowLab/maxnet/wiki/stars

Has example code for downloading predictors and occurance data

Task: Create dataframe of species locations in Arabian Sea and/or Bay of Bengal

https://seamap.env.duke.edu/

Here are some sea turtle species data available:

Looks like more data is available on the east coast of Africa.

Getting data from a zarr into R

identfy "data stories" to look for

Looking for ideas on what data stories we might be able to look for using SDMs. The ideal data stories will have lots of data and be easy to see. Seasonal migrations might be a good starting point.

Within sea turtle data in the Arabian/Indian seas:

• seasonal upwellings bring up cold water + chlorophyll blooms. How might this affect sea turtles?

Add your info to the README.md and INDEX.md files

hey @cacourtier @Pfreire29 @lauratsang @btupper

Please add any info you want to the tutorial developer's section within these two files:
README.md and INDEX.md

If anyone else has Collins or Jade's info, feel free to add it as well!

Is there metadata for green_turtle.csv?

predict function issue

Hi, I am trying to run the predict function following the same steps as in the exercise but I get the following error:

predicted <- predict(sdm.model,
envs.tars %>% sf::st_crop(bb),
clamp = clamp, type = type)

Error in [.stars(newdata, , v) :
selecting using invalid value label(s)?

Everything else I managed to replicate without problem. But I have not been able to solve this error.

Day 2: Learn about SDMs

dismo package
library(dismo)

identify gridded input layers

Input layers for generating an SDM need to be identified. Popular ones are Sea Surface Temperature, Cholorophyll A concentration, and bathymetry.

I have experience pulling these through ERDDAP, but there are other sources too. If using ERDDAP we can try using the (experimental) extractr library. A list of datasets in ERDDAP I like can be found here.

identify taxa occurrence data source(s)

I recommend OBIS (and|or) GBIF for this. The robis library works well for this. The data can be messy but it is still the best I have seen for large spatiotemporal regions of interest.

The holders of SEAMAP data listed in #1 are working to convert their data into Darwin Core so it can also be added to OBIS.

Adjusted get_seaturtle_data.Rmd

@lidefi87 Just a heads up I adjusted a few lines on the intro/end of your get_seaturtles_data.RMD to match better into the quarto.yml flow. Feel free to let me know if you have any questions.