Introducingselfisher: open source software for statistical analyses of fishing gear selectivity

Abstract

Fishing gear is constantly being improved to select certain sizes and species while excluding others. Experiments are conducted to quantify the selectivity and the resulting data needs to be analyzed using specialized statistical methods in many cases. Here, we present a new estimation tool for analyzing this type of data: an R package named selfisher. It can be used for both active and passive gears, and can handle different trial designs. It allows fitting models containing multiple fixed effects (e.g. length, total catch weight, mesh size, water turbidity) and random effects (e.g. haul). A bootstrapping procedure is provided to account for between and within haul variability and overdispersion. We demonstrate its use via four case studies including (1) covered codend analyses of four gears, (2) a paired gear study with numerous potential covariates, (3) a catch comparison study of unpaired hauls of gillnets and (4) a catch comparison study of paired hauls using polynomials and splines. This free and open source software will make it easier to model fishing gear selectivity, teach the statistical methods, and make analyses more repeatable.

Evaluation of a novel research trap for surveys of blue swimmer crab populations

“…We applied a doublebootstrapping method (1000 replications, Herrmann et al 2017;Savina et al 2017) to calculate 95% confidence intervals and incorporate the uncertainty in the estimation resulting from between-deployment (trawl or trap) variation in catch efficiency and availability of crabs, as well as uncertainty associated with the size structure of the catch across individual deployments. This procedure was implemented with the 'selfisher' R package (see https://github.com/ mebrooks/selfisher; Brooks et al 2020) and because of the double bootstrapping procedure inherently controlling for variation which would traditionally be simulated with a random effect of deployment or trap, we did not include any random effects. During this analysis, we initially included estuary as a factor in the models but this was removed and the estuaries pooled, because likelihood ratio tests showed the models including estuary gave no improvements over the simpler model for any of the pairwise comparisons (P > 0.6).…”

Detection and quantification of differences in catch rates among research vessel gears and commercial vessels

“…, where 𝑛1 and 𝑛2 are the CPUE from a single size class (𝑙) at each haul (ℎ) from two vessels (Krag et al 2014;Kotwicki et al 2017;Brooks et al 2020). This catch comparison rate is useful because it allows for hauls to remain paired during the analyses and is a binomial variable, the modelling of which is supported in many statistical computer packages.…”

“…Catch comparison rate (∅ ℎ,𝑙 ) was modelled by a logit link (Eq 1), based on the ratio of swept area of one vessel relative to the other (𝑝 ℎ ), size of scallop (𝑠 𝑖,ℎ,𝑙 ) and size retention model of each vessel's gear (𝑟 𝑖 (𝑙)) (Brooks et al 2020).…”

“…The retention models (𝑟 𝑖 (𝑙)) are a measure of a gear's ability to retain scallops of a given size compared to another gear and are constrained to the range of zero and one by a link function (Brooks et al 2020).…”

“…, where 𝑞 1 and 𝑞 2 are the raising factors at each haul from each of the vessels in a comparison (Brooks et al 2020). Inspection of plots of model fits were used to select the best number of degrees of freedom permitted in the shape of the model-estimated curves, and AIC values were used to select between polynomial and basis spline models after determining the best degrees of freedom (Akaike 1974;Brooks et al 2020).…”