Assisting Ecosystem‐Based Fisheries Management Efforts Using a Comprehensive Survey Database, a Large Environmental Database, and Generalized Additive Models

Abstract

Statistical habitat models, such as generalized additive models (GAMs), are key tools for assisting ecosystembased fisheries management (EBFM) efforts. Predictions from GAMs can be used, for example, to produce preference functions for the ecosystem-modeling platform Ecospace; preference functions permit a flexible representation of spatial distribution patterns in Ecospace by defining the preferences of marine organisms for certain environmental parameter values. Generalized additive model predictions can also be used to map species distributions for assisting marine protected area (MPA) planning. In this study, we applied a recently proposed methodology to produce preference functions for the fish and invertebrates represented in an Ecospace model of the West Florida Shelf (WFS) and to map the hotspots of juveniles and adults of three economically important species for informing future MPA planning in the WFS region. This proposed methodology consists of (1) compiling a comprehensive survey database blending all of the encounter and nonencounter data of the study ecosystem collected by the fisheries-independent and fisheries-dependent surveys that employ random sampling schemes, (2) developing a large environmental database to store all of the environmental parameters influencing the spatial distribution patterns of the marine organisms of the study ecosystem, (3) using the comprehensive survey database and the large environmental database to fit binomial GAMs that integrate the confounding effects of survey and year, and (4) making predictions with fitted GAMs to Subject editor: Donald Noakes,

Protecting juveniles, spawners or both: A practical statistical modelling approach for the design of marine protected areas

“…Based on previous research (e.g. Austin, ; Grüss, Chagaris, et al, ; Leathwick, Elith, & Hastie, ), for each life stage of each species of interest, one should ideally consider: (a) only monitoring programs providing at least 20–50 encounter estimates; and (b) only years associated with at least four encounter estimates.…”

Seasonal Distribution and Habitat Preferences of Crimson Seabream Parargyrops edita: Implications for a Marine Protected Area in Beibu Gulf, Northern South China Sea

“…Eastings and northings were measured based on the distance from a reference location to each sampling site. If a pair of variables exhibited a correlation coefficient value higher than 0.7, the variable that displayed a lower correlation with the other variables was retained for the following analyses (Guisan et al 2002;Leathwick et al 2006;Grüss et al 2018). Among the 13 sampling sites, sites 3, 4, 6, 7, 9, 10, 11, and 12 were inside of the experimental zone of the marine protected area and sites 1, 2, 5, 8, and 13 were outside of the marine protected area.…”

Producing Distribution Maps for a Spatially-Explicit Ecosystem Model Using Large Monitoring and Environmental Databases and a Combination of Interpolation and Extrapolation

“…We did not choose this option, because GAMMs are computationally intensive and are likely to face convergence issues when fitted to very large monitoring datasets like ours. However, Grüss et al (2017a) showed that the spatial patterns of probability of encounter predicted by GAMs treating monitoring program as a fixed effect factor are unaffected by the monitoring program factor, and that the gross magnitude of the probabilities of encounter predicted by these GAMs are only slightly affected by the monitoring program factor. We employed thin-plate regression splines with shrinkage, and each thin-plate regression spline was limited to four degrees of freedom (Roberts et al, 2016;Mannocci et al, 2017).…”

“…. , x n are the environmental covariates selected after the collinearity analysis; and year and monitoring program are "nuisance variables" (i.e., variables that are not of immediate interest but that must be accounted for in the analysis) treated as fixed effect factors (Farmer and Karnauskas, 2013;Grüss et al, 2016aGrüss et al, , 2017a. The fact that year and monitoring program are fixed effect factors entails that it will be necessary to choose a given year and a given monitoring program to make predictions with fitted GAMs (in this case, the average year effect and the monitoring program effect with the highest selectivity; Punt et al, 2000;Maunder and Punt, 2004;Farmer and Karnauskas, 2013;Grüss et al, 2016a; see subsection Production of Distribution Maps for the GOM LME from the Predictions Made by Fitted Binomial GAMs).…”

“…In addition to the shrinkage approach, we applied an extra penalty to each environmental covariate as the smoothing parameter approached zero, which allowed its removal when the smoothing parameter was equal to zero (Marra and Wood, 2011;Drexler and Ainsworth, 2013;Grüss et al, 2014). Once a GAM was fitted, if an environmental covariate p-value had a p > 0.05, it was removed and the model was refitted (Koubbi et al, 2006;Weber and McClatchie, 2010;Grüss et al, 2014Grüss et al, , 2016dGrüss et al, , 2017aChagaris et al, 2015). The restricted maximum likelihood (REML) optimization method was employed (Wood, 2011).…”