Marine fish communities cycle between homogenized and differentiated states through time

Currently in review at PNAS.

Zoë J. Kitchel, Aurore A. Maureaud, Alexa Fredston, Nancy Shackell, Bastien Mérigot, James T. Thorson, Laurène Pécuchet, Juliano Palacios-Abrantes, Maria L.D. Palomares, Antonio Esteban Acón, Mark Belchier, Gioacchino Bono, Pierluigi Carbonara, Martin A. Collins, Luis A. Cubillos, Tracey P. Fairweather, Maria Cristina Follesa, Cristina Garciá Ruiz, Maria Teresa Farriols Garau, Germana Garofalo, Igor Isajlović, Johannes N. Kathena, Mariano Koen-Alonso, Porzia Maiorano, Chiara Manfredi, Jurgen Mifsud, Richard L. O’Driscoll, Mario Sbrana, Jón Sólmundsson, Maria Teresa Spedicato, Fabrice Stephenson, Karl-Michael Werner, Daniela V. Yepsen, Walter Zupa, Malin L. Pinsky

Please contact Z.J. Kitchel with questions about this project. The results are partially reproducible from the scripts and data files that can be shared are either in this GitHub repository or (if they are too large) will be hosted on OSF (as described below).

Where do data come from?

We used of a number of datasets that are already publicly available and/or published with this project, in addition to some regional datasets that may be available upon request from data providers. These are fully described and cited in the manuscript, but we also list them here for ease of download, access, and attribution.

Publicly available trawl data from FISHGLOB, a project to harmonize publicly available trawl survey records from federal agencies around the globe. The raw data files from FISHGLOB are too big to host on GitHub but they will be available on OSF. FISHGLOB has observations for 55 surveys, and 42 of these were appropriate for use in our biodiversity x time analyses.
Some regions are unable to share trawl data publically. Data may be available on request from data provider. See Table S1 for contact information.
Sea bottom temperature data from SODA 3.3.2, an ocean reanalysis data product by the University of Maryland available beginning in 1980.
Sea surface temperature data from OISST, a historical satellite temperature record from the U.S. National Oceanic and Atmospheric Administration beginning in 1982.
Historical fishing pressure estimates from Sea Around Us.

What's in this repository?

The repository is organized as follows:

data contains raw datasets obtained from other sources (described above).
data_prep_code contains Rmd scripts to access (when possible) and prepare trawl, temperature, and fishing datasets.
analysis_code contains Rmd scripts to run all dissimilarity analyses.
figure_code contains Rmd scripts to make figures not made in other scripts.
output contains outputs of scripts that filter, clean, summarize, or analyze data. Also contains csv, rds, and xlsx files generated by scripts for main and supplementary tables.
figures contains image files generated by scripts for the main and supplemental figures.
archive includes a mismash of scripts,output, figures, and tables that were a part of project development, but do not directly connect to currently used scripts, outputs, figures, or tables. Viewer beware, these are very possibly wrong, broken and non-functional but mean something significant to the authors, and are therefore included because ZJK is a hoarder. NB: some of the folders above have their own internal archives

What's not in the repository?

The following files are too big to host on GitHub and are available on OSF only:

The raw publicly available FISHGLOB data (a slightly earlier version of the dataset described in Maureaud et al. 2023), FISHGLOB_public_v1.5_clean.csv. This should be downloaded from OSF and moved to raw-data to reproduce analyses for public surveys only.
Norway (Barents Sea; Nor-BTS) required a data update in September 2023. Therefore, Norway is excluded from FISHGLOB_public_v1.5_clean.csv, and Norway data can be found in data > Norway_Sep2023 > NOR-BTS_clean.RData. This file can be merged (with care) with FISHGLOB_public_v1.5_clean.csv.
Queen Charlotte Sound (DFO-QCS) required a data update in September 2023. Therefore, QCS is excluded from FISHGLOB_public_v1.5_clean.csv, and QCS data can be found in data > DFO-QCS_Sep2023 > QCS_clean.RData
Greenland (GRL) required a data update in September 2023. Therefore, QCS is excluded from FISHGLOB_public_v1.5_clean.csv. Greenland is only available by request, and therefore raw data is not included in this repository.

We did not host the raw SODA and OISST data on OSF because the source files are very large. The SODA sea bottom temperature data were downloaded from http://www.soda.umd.edu/ (accessed May 28, 2023). The NOAA OISST data were downloaded from https://www.ncei.noaa.gov/data/sea-surface-temperature-optimum-interpolation/ (accessed May 19, 2023).

We respect that trawl survey data for some regions are not currently publically available. Therefore, we cannot share survey data for South Georgia,the Mediterranean, South Africa, Chile, Namibia, Newfoundland, Greenland, Iceland, or New Zealand. This repository includes scripts that will prepare and process these data, but does not include raw data files for these regions. Table S1 lists data providers/corresponding authors for each region.

In what order should things be run?

To reproduce the analysis, run these scripts in order:

Trawl data preparation:

data_prep_code > fishglob_species > prepare_fishglob_dataset_1.Rmd: This script prepares bottom trawl survey data observations for analysis using survey by survey specific edits. It also integrates data that needed to be updated from FISHGLOB v1.5 -- Norway and Greenland specifically.
data_prep_code > fishglob_species > standardize_temp_space_2.Rmd: This script standardizes bottom trawl survey data across space and time. Specifically, it assigns all trawl locations to hexagonal cells, removes any years for a survey that sample less than 70% of cells ever sampled for that survey and then removes any cells that are sampled in less than 70% of years.

Dissimilarity time series

analysis_code > dissim_metric_space_time_3.Rmd: This script calculates Bray Curtis and Jaccard total and partitioned dissimilarities for every pairwise tow combination in each survey x year combination. Then, it averages across pairs for annual average for each survey x year.
analysis_code > dissim_metric_space_time_3b_15perc_exclude.Rmd: Same as above but excluding 15% of species from each survey with lowest overall biomass through time series.
analysis_code > dissim_metric_space_time_3c_onethird_exclude.Rmd: Same as above but excluding species that are present in less than 1/3 of years for each survey.
analysis_code > year_dissimilarity_BC_total_models_4a.Rmd: This script performs main dissimilarity time series analysis.
analysis_code > year_dissimilarity_BC_total_models_4b_nullmodelreshuffle.Rmd: Same as above but reshuffling years 1000x times (null analysis for # of surveys we'd expect to be homogenizing and differentiating by chance if there were no directional trends in dissimilarity through time).
analysis_code > year_dissimilarity_BC_balanced_models_4b.Rmd: Same as above but for Bray Curtis balanced changes component of dissimilarity. (Yes two b's, my apologies)
analysis_code > year_dissimilarity_BC_gradient_models_4c.Rmd: Same as above but for Bray Curtis abundance gradients component of dissimilarity.
analysis_code > year_dissimilarity_jaccard_total_models_4d.Rmd: Same as above but for total Jaccard dissimilarity.
analysis_code > year_dissimilarity_BC_total_models_15perc_excl_4e.Rmd: Same as above but excluding 15% of species from each survey with lowest overall biomass through time series.
analysis_code > year_dissimilarity_BC_total_models_onethird_excl_4f.Rmd: Same as above but excluding species that are present in less than 1/3 of years for each survey.

Temperature data preparation

data_prep_code > temperature > Import_SODA_web_convert_raster_bottomtemp_5a_1.Rmd
data_prep_code > temperature > pulling_averaging_SODA_temp_data_5a_2.Rmd
data_prep_code > temperature > pulling_averaging_OISST_surface_temp_data_5b_2.Rmd

Fishing data preparation

data_prep_code > fishing > Pull_SeaAroundUs_6.Rmd

Temperature and Fishing as predictors of annual dissimilarity

analysis_code > Regional_statistics_7.Rmd: merge fishing, and dissimilarity data sets, and then calculate regional statistics, and look at patterns of survey characteristics versus trends in dissimilarity.
analysis_code > annual_dissim_fishing_temp_model_8.Rmd: Runs and ranks models using temperature, fishing, and survey characteristics to predict annual dissimilarity within surveys.
analysis_code > annual_dissim_fishing_temp_model_8b_oisst.Rmd: Same as above, but for surface temperature (OISST).

Misc scripts Scripts without an associated number are accessory scripts. They are not needed for the main or supplemental analyses, but are informative. See text within specific scripts for info on what they do.

One that is particularly helpful:

1. data_prep_code > fishglob_species >data_clean_visualization_cell_year.Rmd: This script visually shows the years and cells excluded from the analyses for each survey (these plots created in data_prep_code > fishglob_species > standardize_temp_space_2.Rmd)

Species Names

For list of species names, see here: output > spp_list.txt

Source scripts for figures and tables

Main text:

Figure 1: figure_code > visualization_homogenization_differentiation.Rmd then figure_code > Map_surveys_coefvalues.Rmd
Figure 2: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure 3: analysis_code > annual_dissim_fishing_temp_model_8.Rmd

Supplemental Figures:

Figure S1: analysis_code > year_dissimilarity_BC_total_models_4b_nullmodelreshuffle.Rmd
Figure S2: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure S3: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure S4: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure S5: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure S6: analysis_code > year_dissimilarity_BC_total_models_15perc_excl_4e.Rmd
Figure S7: analysis_code > year_dissimilarity_BC_total_models_onethird_excl_4f.Rmd
Figure S8: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Figure S9: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Figure S10: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Figure S11: analysis_code > annual_dissim_fishing_temp_model_8b_oisst.Rmd
Figure S12: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Figure S13: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Figure S14: analysis_code > Regional_statistics_7.Rmd
Figure S15: analysis_code > Regional_statistics_7.Rmd

Supplemental Tables

Table S1: data_providers_allsurveys.xlsx, and see Maureaud et al. 2021 Global Change Biology
Table S2: analysis_code > Regional_statistics_7.Rmd
Table S3: analysis_code > year_dissimilarity_BC_total_models_4a.Rmd
Table S4: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Table S5: analysis_code > annual_dissim_fishing_temp_model_8.Rmd
Table S6: analysis_code > annual_dissim_fishing_temp_model_8b_oisst.Rmd
Table S7: analysis_code > annual_dissim_fishing_temp_model_8b_oisst.Rmd
Table S8: data_prep_code > fishglob_species > standardize_temp_space_2.Rmd
Table S9: data_prep_code > fishing > Pull_SeaAroundUs_6.Rmd

Notes

Upon publication, package versions used in this analysis will be captured by renv and listed in the lockfile. Here is a summary of how renv works.
Most of these analyses were run on a personal computer using R 4.2.1, and should be easy to reproduce for public data. The exception is the temperature extraction and averaging, as these require averaging of extremely large source files. Therefore, we recommend users to start with averaged temperature files, SODA_data_temp_avgs_full.rds & OISST_data_temp_avgs_full.rds.