We used the Communities-at-Sea framework developed by St. Martin and colleagues (St Martin and Hall-Arber, 2008\u00a0 (Human Ecology Review); St Martin and Hall-Arber, 2008 (Marine Policy); St Martin and Olson, 2017) to define fishing communities. This approach aggregates peer groups of vessels into community-based fleets, each defined as a unique combination of port, gear, and vessel size. We used two categories of vessel size: vessels that are longer than or equal to 65\u2019, and those that are shorter. Each trip in the vessel trip report (VTR) trawl data was grouped into a community fleet by the port of landing and vessel size, as long as at least one of the following two conditions were met: (1) the vessel landed at least 50% of its trips that year at that port, or (2) the vessel reported that landing port as its home port or principal port on its permit. Trips that landed at a port that did not meet either category were not included in this analysis. To ensure confidentiality of harvester information, we only analyzed community fleets that included at least three vessels in a given year.
\nFor each community fleet in a given year (a \"fleet-year\"), we estimated an annual geographic effort-weighted centroid of fishing activity using a bootstrapping approach. We first used DePiper's (2014) model to calculate a 90% confidence interval for each individual fishing trip. Then, using that confidence interval as a radius around the reported trip location and assuming a uniform distribution, we chose a random point within that area to represent the trip location. For each fleet-year, we used all estimated trip locations to calculate a weighted geographic centroid. Each centroid was weighted by crew size multiplied by trip length to represent a measure of labor time and investment (St Martin and Olson, 2017). We repeated this process 1000 times to generate a distribution of centroids for each community-year.\u00a0See script: bootstrap_centroid.R\u00a0(input: Communities-at-Sea table, output: centroids table).
\nWe used an inverse weighted regression analysis to assess if and to what degree the annual fishing center for each community fleet shifted significantly over time. For each community fleet, we fit a linear regression of latitude against year for each set of bootstrap-replicated centroids described above. We weighted each centroid by the inverse variance of the trip latitudes used to calculate that centroid. This approach weights a centroid with tightly clustered trips more heavily than one with more dispersed trips. We used the mean effect strength from those 1000 regressions as the rate of change in latitude for each fleet. To ensure sufficient data for analysis over time, we restricted this analysis to only community fleets with at least seven years of trip data. See script: build_delta_gf.R (input: Communites-at-Sea table, output: delta_gf table).
\nWe summarized the delta_gf table in a results table. See script: build_results_table.R (input: delta_gf table, output: results_table.csv (this dataset; use \"Get Data\" button). Also see\u00a0Table S1 of Young et al. (2018).
\nIn order to assess the effect of factors correlated with changes in fishing latitude, we fit a series of multiple linear regressions between the rate of latitudinal change and five explanatory variables:
\n(1) Vessel size,
\n(2) Species diversity of catch,
\n(3) Change in composition of catch species,\u00a0
\n(4) Change in depth of fishing location. In order to estimate change in depth of fishing location for each community, we first found the nearest depth recording for each trip using a U.S. coastal relief model (NOAA National Centers for Environmental Information, U.S. Coastal Relief Model, n.d.), and calculated an effort-weighted average depth for all the trips in a community year. As above, we then regressed depth against year and used the resulting slope as the covariate.
\n(5) Port latitude. Species diversity of catch was correlated with port latitude (fleets from more northern ports had greater catch diversity), so we also included latitude of port as a covariate so that we could assess the separate effects of catch diversity and port latitude.
We evaluated models with all possible combinations of main effects as well as three interactions: vessel size and port latitude, vessel size and catch species diversity, and catch species diversity and change in catch species composition. We calculated the corrected Akaike Information Criterion (AICc) for each model and the Relative Variable Importance (RVI) for each variable and interaction included in the model. See script: lat_shift_model.R (input: delta_gf table).
\nIn order to assess factors mediating changes in community size, we fit a series of regressions to assess the effect on rate of change in community size (change in number of unique permits over time; linear) and community disappearance (fewer than 3 permits by 2014; logistic) of three predictor variables: (1) vessel size, (2) species diversity of catch, and (3) port latitude. We evaluated models with all possible combinations of main effects as well as interactions, and calculated AICc and RVI as described above. We used the number of unique fishing permits in a community as a proxy for community fleet size. See script: community_decline_model.R (input: delta_gf table).
\nAdditional details and references can be found in Young et al. (2018).
This dataset describes changes in groundfish fishing communities in the northeast US from 1997-2014 as captured in the vessel trip report (VTR) data collected by the National Oceanic and Atmospheric Administration National Marine Fisheries Service Northeast Fisheries Science Center (NOAA-NMFS-NEFSC), including:
\n- annual latitudinal shift in fishing location;
\n- whether the community is still extant in the dataset (and if not, the year it disappeared from the dataset);
\n- mean shannon-wiener diversity metric of species composition of catch;
\n- percentage of species composing 90% of catch.
Data were processed in R 3.4.4.