Species distribution modeling

Overview

question Questions
  • How to build & evaluate a niche model, visualize and project it ?

objectives Objectives
  • Find and download occurrences data from GBIF

  • Find and download environmental data

  • Process both occurrences and environmental

  • Partition occurrence data

  • Model a theoretical ecological niche and predict species distribution in a future climate scenario by using SDM

requirements Requirements

time Time estimation: 1 hour

Introduction

Species Distribution Modeling (SDM) can help understand the distribution of a species depending on its environment. It can also attempt to quantify the impact of climate change on the species habitat, direct conservation efforts and predict invasive species distributions. This is done by associating data of species occurrences (observations) with a set of environmental data (such as temperature and precipitation).

The goal of this tutorial is to model a theoretical ecological niche and predict species distribution in a future climate scenario by using SDM with the Wallace interactive environment on Galaxy. We’ll use the data occurrences of US Chrysemys picta (Painted turtle) from the North America region.

Agenda

In this tutorial, we will deal with:

  1. Step 1: Import data from public databank
  2. Step 2: Development and evaluation of SDM using Wallace
    1. Obtain occurrence data
    2. Process occurrence data
    3. Obtain Environmental Data
    4. Process environmental Data
    5. Partition Occurrence Data
    6. Build and Evaluate Niche Model
    7. Visualize Model Results
    8. Project Model
  3. Conclusion

Step 1: Import data from public databank

In this study the datasets are all imported from the GBIF databanks. It is also possible to import from bison, iNaturalist and others.

hands_on Hands-on: Import the data

  1. Create a new history for this tutorial and give it a proper name

    tip Tip: Creating a new history

    Click the new-history icon at the top of the history panel

    If the new-history is missing:

    1. Click on the galaxy-gear icon (History options) on the top of the history panel
    2. Select the option Create New from the menu

    tip Tip: Renaming a history

    1. Click on Unnamed history (or the current name of the history) (Click to rename history) at the top of your history panel
    2. Type the new name
    3. Press Enter
  2. Get species occurrences data tool with the following parameters
    • “Scientific name”: Chrysemys picta
    • “Data source”: gbif
    • “Number of records to return”: 10000
  3. Inspect the generated file with the Scratchbook

We have now a tabular file with about 10,001 lines and many columns.

question Questions

  1. What do the lines in the files represent?
  2. What are the columns 1, 2, 3 and 50 of the generated file?

solution Solution

  1. Each line represent an observation of Chrysemys picta, with its location, and some other metadata
  2. Column 1 is the name, column 2 the longitude of the observation of Chrysemys picta, column 3 the latitude and column 50 the country code of the location

We would to extraction the 4 columns previously cited and keep only occurrence records from US.

hands_on Hands-on: Import the data

  1. Cut columns tool with the following parameters
    • “Cut columns”: c1,c2,c3,c50
    • “Delimited by”: Tab
    • param-file “From”: output of Get species occurrences data
  2. Inspect the generated file and check the 4 columns
  3. Filter data tool with the following parameters
    • param-file “Filter”: output of Cut columns
    • “With following condition”: c4=='US'
    • “Number of header lines to skip”: 1
  4. Inspect the generated files

    question Question

    How many occurrences have been conserved?

    solution Solution

    71.18% of the occurrences (7,118) are conserved

  5. Tabular to CSV tool
    • param-file “tabular file”: output of Filter
    • “output csv Separator”: ,
    • “Header in file”: Yes

Step 2: Development and evaluation of SDM using Wallace

Wallace is a R Shiny app integrated into Galaxy. It provides an interactive environment for the rapid and effective development and evaluation of SDM, including data download, cleaning, partitioning, modeling, visualisation and predictions.

Obtain occurrence data

With this you can either upload file you’ve loaded earlier from Galaxy data or you can download data directly from Wallace. Let’s use the data from your Galaxy history:

hands_on Hands-on: Launch Wallace

  1. Click on Visualization (top panel) and then Interactive Environments
  2. Select
    • “GIE”: interactiveShiny
    • “Image”: valentinchdock/wallace-galaxy-ie
    • “Datasets”: output of Tabular to CSV
  3. In Wallace
    1. Go to 1 Occ Data
    2. In Obtain Occurrence Data
      • “Modules Available”: Galaxy History User
      • “Select from your Galaxy History User csv file”: output of Tabular to CSV
      • Click on Load Occurrences

In the main panel of Wallace, the data will load. They can now be inspected using the map, the table, etc

You now have your occurrence records on Wallace!

Process occurrence data

For the modelling, we need to the choose the occurrences we want to use. There is 4 ways to do it in Wallace in the 2 Process Occ panel:

  1. Selecting occurrences on the map by delimiting an interesting geographic area
  2. Removing occurrences by ID
  3. Delimiting a spacial thin

    This allow to select occurrences by setting a minimum distance (in km) between the different occurrences. For example, by typing 30km, we will get all the occurrences on the map which are at minimum 30km from each other.

  4. Considering all the occurrences

We want to work on the data from the US, we will select the occurrences with the 1st option

hands_on Hands-on: Select interesting occurrences

  1. Go to 2 Process Occ
  2. In Process Occurrence Data
    • “Modules Available”: ` Select Occurrences On Map`
  3. In the middle panel,
    1. Click on the polygon icon on the map
    2. Draw a polygon to delimit the area
  4. On the left panel, click on Select Occurrences

Obtain Environmental Data

We would like now to provide environmental variables, that are considered relevant for biological purposes. The Bioclimatic variables consist of such original and derived variables. We will associate them with the occurrence data, using the WorldClim Bioclims module.

This module provides a raster with environmental variables from online sources. The raster is composed of environmental information. Each layer of the raster contains a climatic variable: starting from BIO1 = Annual mean temperature, to BIO19 = Precipitation of Coldest Quarter.

hands_on Hands-on: Obtain Environmental Data

  1. Go to 3 Env Data
  2. In Obtain Environmental Data
    • “Modules Available”: WorldClim Bioclims

      comment Using your own raster

      If you prefer to load your own raster:

      1. Import it into Galaxy in the GeoTIFF format
      2. In Obtain Environmental Data
        • “Modules Available”: Galaxy History User
        • “Select from your Galaxy History User csv file”: imported GeoTIFF file
    • “Select WorldClim bioclimatic variable resolution”: 10 arcmin

      This is the lowest resolution.

  3. Click on the Load Env Data

Process environmental Data

Wallace will now associate environmental data and occurrences data to train a model.

  1. It first creates a buffer zone around the occurrences

    We can choose the size of the buffer zone to control the area we will be working with and on which a map of suitability will be made.

    Several background extent can be used:

    • Bounding box will define an area where the occurrences centered
    • Minimum convex polygon will make an area considering the repartition of the occurrences
    • Point buffers will use occurrences localities to build a buffer zone around each occurrence
  2. It samples Background Points

hands_on Hands-on: Obtain Environmental Data

  1. Go to 4 Process Envs
  2. In Process Environmental Data
    • “Modules Available”: Select Study Region
  3. In Step 1: Choose Background Extent
    • “Background Extents”: Minimum convex polygon
    • “Study region buffer distance (degree)”: 1
  4. Click on Select
  5. In Step 2: Sample Background Points
    • “No. of background points”: 100000
  6. Click on Sample

Partition Occurrence Data

By partitioning data, a dataset is divided into subsets (i.e. bins). A model is then built on each of subsets but one and test it on the last one (assuming that all the groups are independent).

There is 2 possible way to partition data:

  1. Non-spatial Partition: a partition used when there is no bias due to space, time or sampling method
    1. Jakknife (k=n): each occurrence in the dataset is considered as equal to a bin. This is usually used when you have a small dataset with no known bias
    2. Random k-fold: partition the data randomly in a number of bins set by the user with the option Number of Folds
  2. Spatial Partition: a partition used when there could be bias due to time, space or sampling method
    1. Block (k=4): the area is divided in four and the different occurrences are put equally into four bins
    2. Checkerboard 1 (k=2): two bins are used according to the position of the occurrence on the grid
    3. Checkerboard 2 (k=4): four bins are used according to the position of the occurrence on the grid. This require an aggregation factor, which is the size of a second grid put on a first one. For example, with a factor 4, the grids size will be 4x4

      Checkerboard 2
      Figure 1: Example of checkerboard (k=4) with a factor 4. Source: doi:10.0.4.87/2041-210X.12261

For both of these techniques the number of occurrences into each bin may vary.

hands_on Hands-on: Partition Occurrence Data

  1. Go to 5 Partition Occs
  2. In Partition Occurrence Data
    • “Modules Available”: Non-spatial Partition
    • “Options Available:”: Checkerboard 2 (k=4)
    • “Aggregation Factor”: 6
  3. Click on Partition

Build and Evaluate Niche Model

Wallace can build different models using either:

  1. The presence-only approach BIOCLIM (Module BIOCLIM) (Booth 2018)
  2. The presence-background (presence-pseudo absence) algorithm Maxent (Module Maxent)

To evaluate these models, Wallace computes the performance on a hold-out dataset (data not used for training) and provide evaluation metrics as the AUC (Area Under the Curve) mean. As a rule of thumb, an AUC of 0.75 and above is considered good, and closer to 1 is better.

hands_on Hands-on: Build and evaluate the niche model

  1. Go to 6 Model
  2. In Build and Evaluate Niche Model
    • “Modules Available”: BIOCLIM
  3. Click on Run

Visualize Model Results

We would like now to visualize the model of the theoretical niche. First we will make a chart to simulate an ecological niche:

  • X-axis: temperature
  • Y-axis: the annual precipitation

hands_on Hands-on: Visualize Model Results

  1. Go to 7 Visualize
  2. In Visualize Model Results
    • “Modules Available”: BIOCLIM Envelope Plots
    • “Axis 1”: 1
    • “Axis 1”: 12
    • “Set threshold”: 0.75

temperature vs annual precipitation

The optimum environmental parameters for this species (represented in green) is between 5°C and 15°C (on the graph, values are x10) with an annual precipitation between approximately 700mm and 1250mm.

To visualize the suitability on the map, we can use the map prediction module.

hands_on Hands-on: Visualize Model Results

  1. In Visualize Model Results
    • “Modules Available”: Map Prediction
    • “Set threshold”: No threshold

      It will display a gradient of predicted presence or use. If you prefer a map with the predicted presence and predicted absence, select minimum training presence.

  2. Click on Plot

map prediction

Project Model

Wallace can use the trained model to predict possible species distributions in a different area, outside of the sampled one. Here, we will try it for the Canada.

hands_on Hands-on: Visualize Model Results

  1. Go to 8 Project
  2. In Project Model
    • “Modules Available”: Project to New Extent
      1. In the middle panel,
      2. Click on the polygon icon on the map
      3. Draw a polygon around a part of Canada
  3. In Project Model
    • “Set threshold”: Minimum Training Presence
  4. Click on Project

We can also predict not only for a different area, but also different timing and climate. The Global Circulation Model (GCM) are used here to predict atmospheric fluctuation and then study climate change. Each model is different and use parameters like ocean atmosphere and others. The prediction need also a RCP scenario, a scenarios about the amount of greenhouse gases emitted in the near future. This scenario may have different predicted presence models.

We would like to predict the presence of Chrysemys Picta in Canada in 2050, given the model CCSM4, a US model based on earth circulation, and a RCP value of 2.6

hands_on Hands-on: Visualize Model Results with a 2.6 RCP

  1. In Project Model
    • “Modules Available”: Project to New Time
    • “Select time period”: 2050
    • “Select global circulation model”: CCSM4
    • “Select RCP”: 2.6
    • “Select threshold”: Minimum Training Presence
  2. Click on Project

CCM4 with a 2.6 RCP

We would like now to compare the prediction for RCP value of 8.5 .

hands_on Hands-on: Visualize Model Results with a 8.5 RCP

  1. In Project Model
    • “Modules Available”: Project to New Time
    • “Select time period”: 2050
    • “Select global circulation model”: CCSM4
    • “Select RCP”: 8.5
    • “Select threshold”: Minimum Training Presence
  2. Click on Project

CCM4 with a 8.5 RCP

Conclusion

Following this tutorial, we have been able here to load a dataset of occurrences used in the shiny app Wallace and model the repartition of Chrysemys picta (Painted turtle) with the Species Distribution Modeling (SDM) method. It allowed us to visualize it’s ecological niche and how climate change can influence it’s future repartition on North America. The project saved can help for future similar studies.

details Some useful references

  • Predicting species distributions for conservation decisions (Guisan et al. 2013).
  • BIOCLIM: the first species distribution modelling package, its early applications and relevance to most current MAXENT studies. Diversity and Distributions (Booth 2018).
  • ENMeval: An R package for conducting spatially independent evaluations and estimating optimal model complexity for MAXENT ecological niche models. (Muscarella et al. 2014).

Here for informations on Global Circulation Model(GCM), how it’s done what is taken in consideration and more.

keypoints Key points

Useful literature

Further information, including links to documentation and original publications, regarding the tools, analysis techniques and the interpretation of results described in this tutorial can be found here.

References

  1. Guisan, A., R. Tingley, J. B. Baumgartner, I. Naujokaitis-Lewis, P. R. Sutcliffe et al., 2013 Predicting species distributions for conservation decisions. Ecology letters 16: 1424–1435.
  2. Muscarella, R., P. J. Galante, M. Soley-Guardia, R. A. Boria, J. M. Kass et al., 2014 ENM eval: An R package for conducting spatially independent evaluations and estimating optimal model complexity for Maxent ecological niche models. Methods in Ecology and Evolution 5: 1198–1205.
  3. Booth, T., 2018 BIOCLIM–the first species distribution modelling package.

congratulations Congratulations on successfully completing this tutorial!