Range edges of North American marine species are tracking temperature over decades

A. Fredston, M. Pinsky, B. Selden, C. Szuwalski, J. T. Thorson, S. D. Gaines, B. S. Halpern

To reference the paper, data, or methods, please cite Fredston et al. 2021. Contact A. Fredston with questions or for a PDF of the paper at [email protected]. This repository also has a .

Overview

This repository contains code to:

• Estimate the annual position of range edges from NOAA trawl survey data using VAST
• Use hindcast SST datasets to reconstruct the extreme temperatures found at species' range edges each year
• Conduct statistical analyses of range edge shifts and edge thermal niche shifts over time

To fully reproduce the analysis, users will need to install VAST. However, this repository contains outputs from VAST and subsequent statistical analyses for users who just want to examine the results or reproduce figures.

The repository is organized as follows:

• raw-data contains raw datasets obtained from other sources that have not been changed in any way. Where possible, data are downloaded within scripts, so future users do not need to have data files. Where that was not possible, scripts using the data files contain instructions to download them.
• scripts contains all code to analyze or transform data.
• processed-data contains outputs of scripts that filter, clean, summarize, or analyze data, e.g., a dataframe with the higher taxonomy of species used in the final analysis.
• functions contains homemade functions called in scripts.
• figure-scripts contains code to produce figures in the manuscript (the outputs are in results).
• results contains figures, tables, and other outputs that are used in the manuscript.

There are some additional directories for model outputs that are not version controlled. Large files (e.g., raw temperature datasets), images, and PDFs are not version controlled.

Instructions

Scripts should be run in the following order. The bulk of the analysis can be found in run_VAST.R, where range edges are estimated, and calculate_edge_thermal_niches.R, where Bayesian linear regressions are fitted to change in range edge and thermal niche position over time.

1. get_neus_data.R imports Northeast data (the only dataset requiring manual download) and reformats it to match those downloaded with FishData.
2. get_axes_of_measurement.R creates a custom axis for use with VAST in each region.
3. run_VAST.R uses VAST to calculate range edges for all three regions. It runs in parallel (don't forget to update the number of cores based on your machine!) but even so may take several days for all species and regions. Note that the output data frames are also in the repository.
4. get_taxonomy.R fetches higher taxonomy of study species from WORMS using taxize. This needs to be re-run every time VAST is re-run in run_VAST.R
5. validate_range_edges.R filters the VAST output for only range edges that actually fall in the study region, based on passing certain filters. It then matches the remaining edges with traits from Beukhof et al. 2019. This needs to be re-run every time VAST is re-run in run_VAST.R.
6. prep_sst.R fetches historical SST data from the NOAA ERDDAP server for each region, crops rasters to the extent of a bathymetric mask for each region that is also created in this script, performs a mean bias correction to combine different SST datasets, and writes the SST data out as dataframes. This does not need to be re-run unless spatial or temporal extent change.
7. match_sst_to_axis.R matches SST values to points along the axis of range limit measurement for each study region, and combines SST datasets where necessary. This doesn't need to be re-run if new edges are generated, because it takes as input the VAST coordinates, not the range edge positions.
8. calculate_edge_thermal_niches.R fits Bayesian models to range edges and edge thermal niches. It also generates a number of plots that require the full output from the sampling algorithm, including the example plot in the main text, and the traits analysis in the Supplement. DANGER: this script was developed on a remote server and may overwhelm personal computers' processing capacity.
9. analyze_range_edges.R conducts the main analyses in the paper.
10. paper_stats.R calculates miscellaneous statistical results reported in the manuscript.

Code to generate figures in the manuscript and supplementary materials can typically be found in the figure-scripts folder. A few exceptions:

• Figures plotting posterior distributions require the full output from the Bayesian models and are generated in calculate_edge_thermal_niches.R.
• The figures in Appendix 2 showing how the axes of measurement were developed are generated in get_axes_of_measurement.R.

Computational requirements

Analyses were conducted in R 3.6.0 on a machine with the following specifications: Windows Server 2019 2-Intel 6154 Xeon 3.0Ghz 18cores(36 threads)each 512 GB memory 6TB- SSD storage 18TB- HDD storage NVIDIA P4000 8Gb 10Gb ethernet

On this machine, the VAST models (run_VAST.R) for all regions were run in parallel on 18 cores overnight. Bayesian models in calculate_edge_thermal_niches.R are not currently scripted to run in parallel; on this machine, each set of models takes several hours to run. The Bayesian models, especially the second set fitting edge thermal niche change over time, are highly memory-intensive and may crash systems with low memory capacity.

Use, problems, and feedback

If you encounter any errors or problems, please create an Issue here. Likewise, please consider starting an Issue for any questions so that others can view conversations about the analysis and code. Again, don't hesitate to contact the lead author, A. Fredston, at [email protected].