Prioritizing monitoring and conservation efforts for fish spawning aggregations in the U.S. Gulf of Mexico

In the U.S. Gulf of Mexico (U.S. GOM), the identification and characterization of transient fish spawning aggregation (FSA) sites is recognized as a regional priority for conservation, but progress is hindered by a lack of understanding of FSA distributions for most exploited species. We employed information compiled in regional databases on FSAs and monitoring for the U.S. GOM to fit species distribution models and produce maps showing the areas likely to host single- and multi-species transient FSA sites. Our results revealed two distinct regions of the U.S. GOM for prioritizing monitoring and conservation efforts for transient FSAs: the coastal waters surrounding major bay systems, particularly those of Texas and Louisiana, and portions of the continental shelf edge (the Flower Garden Banks area and the West Florida shelf edge). The next step would be to locate and characterize actual transient FSA sites in the U.S. GOM by surveying within the areas we identified.

SciEntific REPoRtS | (2018) 8:8473 | DOI: 10.1038/s41598-018-26898-0 poses a substantial issue. The U.S. GOM alone has a surface area of around 697,000 km². Therefore, there is a need for identifying the areas likely to host transient FSA sites (henceforth "potential FSA areas") in the GOM to prioritize monitoring and conservation efforts for transient FSAs in the region.
In this study, we use the information on transient FSAs (hereafter usually called "FSAs") compiled within the RESTORE project, and a large monitoring database compiled in previous studies [14][15][16] , to fit species distribution models (SDMs) and then map the distribution of potential FSA areas in the U.S. GOM. First, we employ the information on lengths at sexual maturity and spawning months compiled within the RESTORE project to extract relevant data from the large monitoring database. Then, we fit SDMs to the monitoring data and make predictions with the fitted SDMs to map the encounter probability of adult fish during the spawning season. Finally, we use the encounter probability maps for adult fish during the spawning season and a simple algorithm to produce maps showing the distribution of potential single-and multi-species FSA areas in the U.S. GOM. Here, we consider the 17 species forming transient FSAs or a mix of resident and transient FSAs identified in the RESTORE project (Table 1).

Results
We were able to fit geostatistical generalized linear mixed models (GLMMs) for four species and generalized additive models (GAMs) for eight species. The twelve SDMs passed the validation tests (Supplementary Data) and were therefore used to produce maps (Fig. 2, Supplementary Fig. S1). Due to an absence or lack of monitoring data, we were unable to fit SDMs for the following five species: yellowfin grouper (Mycteroperca venenosa), yellowmouth grouper (M. interstitialis), Nassau grouper (Epinephelus striatus), goliath grouper (E. itajara) and cubera snapper (Lutjanus cyanopterus) (Supplementary Table S1).
The location of potential FSA areas tends to vary from one grouper-snapper-jack species to another (Fig. 3A-H). The potential FSA areas of greater amberjack (Seriola dumerili) occur along the entire GOM shelf edge (Fig. 3A), while those of almaco jack (S. rivoliana) are concentrated in the Flower Garden Banks area, in Alabama and Florida Panhandle shelf waters and in the Pulley Ridge and Dry Tortugas areas (Fig. 3B). Potential FSA areas are found all over the GOM shelf for yellowedge grouper (Hyporthodus flavolimbatus) and scamp (M. phenax) (Fig. 3C,E). Those of Warsaw grouper (H. nigritus) are concentrated on the Louisiana shelf and in the Pulley Ridge and Dry Tortugas areas (Fig. 3D), while those of gag (M. microlepis) are concentrated on the West Florida shelf south of Apalachicola, Florida (Fig. 3F). Finally, the potential FSA areas of black grouper (M. bonaci) and mutton snapper (Lutjanus analis) occur on the southwestern Florida shelf (Fig. 3G-H).
Regarding the coastal species considered in this study, potential FSA areas for black drum (Pogonias cromis) are found in the coastal waters of Texas near the Mexican border, eastern Texas, Louisiana and Mississippi, and in the Cedar Key region (Fig. 3I). Those of red drum (Sciaenops ocellatus) occur in the coastal waters of all U.S. GOM states; in Florida, they are predicted to be located in the Turkey Point, Cedar Key, Tampa Bay and If we consider all the species for which we fitted SDMs, potential multi-species FSA areas are primarily located in coastal waters surrounding major bay systems, particularly those of Texas and Louisiana, and in the Flower Garden Banks area and along the West Florida shelf edge (Fig. 4A). The potential multi-species FSA areas of groupers-snappers-jacks occur primarily in the Flower Garden Banks area and along the West Florida shelf edge south of 26°N (Fig. 4B), while those of coastal species are primarily located in Texas and Louisiana coastal waters (Fig. 4C). Our FSA index maps suggest that existing FSA-based marine protected areas (MPAs), which are all located in West Florida waters, offer protection to multi-species FSAs (Fig. 5). However, those are not the regions of West Florida that include the potential FSA areas of the largest possible number of species. Based on our predictions, off West Florida, it would be advantageous to implement FSA-based MPAs in the Pulley Ridge region, particularly for protecting the FSAs of multiple grouper, snapper and jack species (Fig. 5). Similarly, in the western U.S. GOM, if the FSA sites of grouper, snapper and jack species could be documented in the Flower Garden Banks area, it would be beneficial to set them aside as MPAs (Fig. 4B).

Discussion
We were able to construct maps showing areas likely to host the FSA sites (the potential FSA areas) of 12 of the 17 study species. Our maps revealed the existence of two groups of species forming transient FSAs in the U.S. GOM: (1) a group made up of the coastal (Sciaenidae, Sparidae and Paralichthyidae) species, which spawns in coastal waters, particularly those surrounding major bay systems (Figs. 4C); and (2) a group consisting of the groupers, snappers and jacks, which reproduces offshore (Fig. 4B). Consequently, potential multi-species FSA areas in the U.S. GOM are located both in coastal waters surrounding major bay systems, particularly those of Texas and Louisiana, and in offshore regions associated with the continental shelf edge (the Flower Garden Banks area and the West Florida shelf edge). Our predictions concur with the findings of a paper in preparation, where  Table 1. Characteristics of the species considered in this study. The information provided, which was compiled in Biggs et al. 12  FSA locations in the U.S. GOM are being mapped from the existing literature, data derived from historical histology collections, reliable accounts and personal observations from fishers' logbooks and data collected by some of the authors of the present study 17 . Moreover, the spatial distribution patterns predicted for the U.S. GOM in the present study are congruent with those reported in adjacent marine regions. For example, the major channel passes along the coasts of Georgia, North Carolina and South Carolina are known to harbor FSAs of red drum, black drum, southern flounder and sheepshead at different times of the year [18][19][20][21][22][23] . Similarly, numerous species of groupers (e.g. Nassau grouper, black grouper, yellowfin grouper), snappers (e.g. cubera snapper, mutton snapper) and jacks (e.g. greater amberjack, almaco jack) co-occur at multi-species FSA sites in the Caribbean along the continental shelf edge [24][25][26] .
We were unable to produce maps showing the potential FSA areas of five study species: yellowfin, Nassau, yellowmouth and goliath groupers, and cubera snapper. We were not surprised to be unable to generate maps for Nassau grouper, which is absent from the U.S. waters of the GOM 27 , and for yellowfin grouper, which occurs only in a few locations of the U.S. GOM 28 .
We did not have enough data to fit SDMs and produce maps for cubera snapper and goliath grouper. Yet, there are anecdotal reports of cubera snapper forming FSAs off Texas, Louisiana and Florida 13 . Regarding goliath grouper, the species is relatively frequently encountered by the Reef Environmental Education Foundation (REEF) Fish Survey, a monitoring program based on the visual observations of volunteer divers 29 ; however, the REEF Fish Survey does not measure fish lengths and was therefore not considered in this study. Recent reports suggest that goliath grouper is present throughout the U.S. GOM 13,30 . Monitoring efforts targeting cubera snapper and goliath grouper throughout the U.S. GOM should be initiated. This is particularly critical in the case of goliath grouper to improve our knowledge of the spatial distribution and ecology of the species, because the species has long been listed as "critically endangered" (by the International Union for Conservation of Nature and the U.S. Endangered Species Act), but there have been discussions about reopening its fishery 31 .
It was surprising not to have enough data to fit a SDM and produce maps for yellowmouth grouper, given that the species may be common in the western U.S. GOM 32 . We suspect that this lack of data may be because yellowmouth groupers are often misidentified as scamps 13 . Consequently, Fig. 2D actually shows the potential FSA areas of scamp and/or yellowmouth in the U.S. GOM. In the future, efforts should be made to clearly distinguish between the yellowmouth groupers and scamps that are sampled and landed to improve the assessment and management of the two species.
It is important to emphasize that, because transient FSAs are generally short-lived events, some of the areas identified in our maps may not host FSA sites, but may rather be areas where spawners feed, rest or occupy cleaning stations before or between spawning events 3 , or adult home ranges because evidence suggests that, in many fish, some adults skip spawning 33,34 . Yet, even if our study determined only areas likely to host transient FSA sites, it provides valuable information for prioritizing validation and monitoring efforts for transient FSAs; we identified a relatively small number of 20 km × 20 km areas of the vast U.S. GOM region that are likely to host FSA sites for 12 species of primary concern or multi-species FSA sites.
Conservation efforts, especially in vast regions such as the U.S. GOM, require projects synthesizing large amounts of information and projects developing management support tools, but also interactions between these two types of projects. In this study, we employed the information compiled by empiricists for a synthesis project   interest. This endeavor, which should involve fishers, would consist of locating candidate FSA sites within the 20 km × 20 km areas based on fishers' interviews, satellite images, aerial photographs and bathymetric charts; and then collecting evidence of spawning at the sites, such as videos of gamete release or females with hydrated oocytes 35 . Ultimately, it will be possible to establish well-informed spatial management plans to better protect transient FSAs from overfishing in the U.S. GOM. Our results suggest that a priority is to survey the Pulley Ridge and Flower Garden Banks areas to locate and verify the actual FSA sites of groupers, snappers and jacks in these areas and set these FSA sites aside as MPAs (Figs 4 and 5). Achieving ambitious conservation objectives requires that empiricists, modelers, fishers and resource managers all work closely together 2,35 .

Methods
Study species. The 17 species considered in this study include two jacks (Carangidae), nine groupers (Epinephelidae), two snappers (Lutjanidae), two Sciaenidae, one Sparidae, and one Paralichthyidae ( Table 1). The study groupers are all protogynous, i.e., mature first as females and then change into males, while all the other species considered in this study are gonochoristic. Among the grouper species, gag (Mycteroperca microlepis), is relatively unique in that the adult females and males of the species are spatially segregated during most of the year; adult male gags stay at FSA sites year-round and are joined by adult females during the spawning season 7,36 . Thus, to identify gag potential FSA areas, it is more relevant to determine the annual spatial distribution of adult male gags than the spatial distribution of the entire adult gag population during the spawning season 37 .
Monitoring data. Previous studies 14-16 compiled a large monitoring database gathering all the monitoring data collected in the U.S. GOM over the period 2000-2016 using random sampling methods. For this study, we employed the 26 fisheries-independent and eight fisheries-dependent datasets from the large monitoring database for the U.S. GOM that collated length information ( Table 2, Supplementary Table S2). From each dataset, we extracted the following information: (1) the longitudes and latitudes at which monitoring took place; (2) the years and months during which monitoring took place; and (3) whether the adults (all species but gag) or adult males (gag) of the study species were encountered or not during sampling events. Encounters/non-encounters for adult fish were obtained using the lengths collated during monitoring events and the lengths at sexual maturity compiled in the RESTORE project 12 (Table 1). Encounters/non-encounters for adult male gag were extracted using the lengths collated during monitoring events and gag length at sex change (102 cm TL 33 ).
For each species, we established which monitoring datasets, years and months should be considered to fit SDMs. To select monitoring datasets and years for a given species, we applied, to the extent possible, the following rules: (1) monitoring datasets with fewer than 20-50 encounters were excluded, following the recommendations of Leathwick et al. 38 and Austin 39 ; and (2) years with fewer than five encounters were excluded. The latter rule was established in recent studies that fitted SDMs to the large monitoring database for the U.S. GOM for generating products for ecosystem models [14][15][16] . For all study species except gag, only monitoring data collected during the spawning months of the species (Table 1) were considered to fit SDMs; in the case of gag, monitoring data collected during any month of the year were considered. The monitoring data employed in this study and the products derived from them are not available to readers, because they include fisheries-dependent data that are confidential.
Statistical modeling. We proceeded in two steps. First, for each species, we tried to fit a geostatistical binomial GLMM 15,37 to monitoring data. Geostatistical binomial models are based on the tenet that encounter probability at a given location resembles more encounter probability at neighboring locations than encounter probability at remote sites, i.e., these models account for spatial structure at a fine spatial scale. Thus, geostatistical binomial GLMMs estimate a smoothed surface that accurately describes how encounter probability varies over space 40 . Second, for those species for which we were unable to fit a geostatistical GLMM (because the model did not converge due to a lack of encounter estimates), we fitted a binomial GAM accounting for spatial structure at a broad spatial scale (through the integration of an interaction term between eastings and northings) 14 to monitoring data.
Briefly, our geostatistical binomial GLMMs integrate Gaussian Markov random fields to model spatial residuals in encounter probability, as well as the fixed effect of year and the random effect of monitoring program 15,37 . Template Model Builder called within the R environment 41 was used to estimate geostatistical GLMM parameters (Supplementary Methods).
In addition to the interaction term between eastings and northings, our binomial GAMs integrate the fixed effects of monitoring program and year 14 . GAMs were fit in the R environment, using the "mgcv" package 42 (Supplementary Methods).
Mapping. We mapped the encounter probability and the hotspots of adults of the study species during the spawning season. First, to be able to generate encounter probability maps, we produced prediction grids for the study species from a 20 km × 20 km spatial grid covering the whole U.S. GOM. The prediction grids were constructed based on the ranges of latitude, longitude and depth at which adults of the study species are encountered by monitoring programs during the spawning season (or year-round in the case of gag). To determine depth ranges, we relied on a 20 km × 20 km depth raster generated from the SRTM30 PLUS global bathymetry grid from the GOM Coastal Ocean Observing System 43 .
Second, for the species for which it was possible to fit a geostatistical GLMM, we employed the fitted GLMM and the prediction grid for the species to produce an encounter probability map for each of the sampling years. Then, the encounter probability maps for each sampling year were averaged to generate one long-term encounter probability map 15,37 . For the species for which it was only possible to fit a GAM, we produced long-term encounter probability maps using the fitted GAM and the prediction grid for the species, and the average year effect and the monitoring program effect with the highest selectivity 14 Third, we constructed hotspot maps from the long-term encounter probability maps. The hotspots of a given species during the spawning season (year-round in the case of gag) are the cells of its prediction grid where, during the spawning season (year-round in the case of gag), the encounter probability the species is equal to or greater than the mean encounter probability of the species over the entire prediction grid 15,46,47 . The hotspot map generated for a given species indicate the location of its potential FSA areas.
We also produced "FSA index maps" by adding together hotspot maps for individual species 14 ; the higher the FSA index in a given area, the higher the chances that multi-species FSAs occur in that area. The first FSA index map was created by adding together hotspot maps for all the study species. The second FSA index map was constructed from the hotspot maps generated for groupers, snappers and jacks. Groupers and snappers are the FSA-forming species that usually receive the most attention 1 . While the timing of spawning for jacks may differ from that for groupers and snappers, both greater amberjack (Seriola dumerili) and almaco jack (S. rivoliana) have been documented to aggregate and spawn at the same sites as groupers in numerous locations in the Caribbean 25,26 . Finally, the third hotspot index map was constructed from the hotspot maps generated for coastal species, namely the Sciaenidae, Sparidae and Paralichthyidae species considered in this study.
To gauge the degree of protection offered by existing FSA-based marine protected areas (MPAs), we overlaid FSA index maps with the boundaries of these MPAs. At present, four FSA-based MPAs are implemented in West Florida waters: the Madison-Swanson MPA, Steamboat Lumps, the Edges, and the Dry Tortugas Marine Reserve (Fig. 1). The Madison-Swanson MPA, Steamboat Lumps and the Edges are partial-take year-round MPAs, where only recreational pelagic trolling activities are allowed during a fraction of the year 7,9 ; these MPAs were established to protect the FSAs of reef fish species, particularly gag and scamp. The Dry Tortugas Marine Reserve is a no-take year-round MPA, which was created in 2001 in part to protect the FSAs of mutton snapper 10,11,48 .  Table 2. Fisheries-independent and fisheries-dependent datasets from the large monitoring database for the U.S. Gulf of Mexico considered in this study. Details about the datasets can be found in Supplementary Table S2.