This repository goes through the analysis for the manuscript titled: Overfishing and habitat loss drives range contraction of iconic marine fishes to near extinction by HF Yan, PM Kyne, RW Jabado, RH Leeney, LNK Davidson, DH Derrick, B Finucci, RP Freckleton, SV Fordham, and NK Dulvy. The code was written by HF Yan. The paper can be found here: https://advances.sciencemag.org/content/7/7/eabb6026.

We conducted the analyses using R v.3.5.2 on macOS 10.15.7 with the following sessionInfo()

tidyverse 1.3.0
gbm 2.1.4
dismo 1.1-4
ROCR 1.0-7
ModelMetrics 1.2.2
parallel 
doSNOW 1.0.16
foreach 1.4.4
brms 2.10.0


1. Data folder
The file called OccurrenceData.csv is the data used to run the Boosted Regression Tree analyses and the dynamic geography analyses. The headers are as follows:

logtotalGearTonnes: ln gear-specific marine fisheries landings
logPprodMean: ln marine primary productivity
logCoastPop: ln coastal population
logEstDis: ln estuarine discharge rate
logProteinDiet: ln marine protein consumption
logGDP: ln Gross Domestic Produc
HDI: Human Development Index
logMang: ln mangrove area
WGI: World Governance Index
logChondCatch: ln chondrichthyan catch
logShelfAreaShallow: ln continental shelf area (shallow)
SstMean: mean sea surface temperature (ºC)
logFishEffort: ln fishing effort
speciesdwarf: dummy-coded species Pristis clavata
speciesgreen: dummy-coded species Pristis zijsron
specieslarge: dummy-coded species Pristis pristis
speciesnarrow: dummy-coded species Anoxypristis cuspidata
speciessmall: dummy-coded species Pristis pectinata

Occurrence is coded as follows: 
1 - present
0 - extinct
2 - Presence Uncertain

Countries are coded using their ISO 3-letter country code as denoted by the United Nations. 

2. Model Outputs folder
DGmod.rds is the dynamic geography model (i.e., logistic regression of occurrence as a function of shelf area, mangrove area, gear-specific landings, and country ID as a grouping factor). 

The BRT daughter folder has the 1000 bootstrapped dataframes for: (1) the relative influences of each variable, (2) the predictions on the test set, (3) the predictions for the Presence Uncertain nations, and (4) the model performance results. Note, these are all dated for 24 August 2020 as that was the last day that these models were run. 

Within the BRT folder, there is a PDP daughter folder for the results of the predicted partial dependences of each variable. Again, these are also dated for 24 August 2020. 

3. Analysis R script
The analysis is run in parallel and thus requires multiple cores. The following model metrics are calulated within the parallel loop:

cvauc - cross-validated AUC
cvcorr - cross-validated correlation
dev_exp - deviance explained
evauc - evaluation AUC
evdev - evaluation deviance