Assessment of over Four Decades the Status of White Grouper Epinephelus aeneus (Geoffroy Saint-Hilaire, 1817) Population in the Eastern Central Atlantic

Abstract

Senegalese and Mauritanian fisheries exploit the same species of white grouper but have different exploitation histories. In Senegal, white grouper has been fished for a very long time (since the 1970s), whereas it is relatively recent in Mauritania. In addition, Senegalese small-scale fishermen exploit this species in the Gambia. Nevertheless, mainly for practical reasons, all attempts to assess the status of the stock have, until recently, been conducted at a national level except by the FAO CECAF North working group who assumed it to be a single stock for the three countries. However, their analysis gives very little attention to length frequency data, even though the fisheries have different selectivity that are likely to affect exploitation rates. In addition, management is mainly focused on length, with no TACs or quotas established at national or sub-regional levels. This work is based on a large compilation of available data from the databases of three countries complemented by the collection of length frequencies data within the framework of the European Union Demerstem/PESCAO project. Two approaches were combined (catch and length-based model) to establish a reliable diagnosis of the state of this resource, along with a spatial analysis to identify the areas most affected by fishing. The results obtained for the sub-region show a severe overexploitation of biomass revealed by the Bayesian biomass production surplus model (JABBA). Stock status indicators show overfishing with low biomass (B/Bmsy = 0.34) and high fishing mortality (F/Fmsy = 5.79). Overall, the trajectory of the state of the stock illustrated by the Kobe figure indicates that the white grouper stock has been overexploited since the 1990s. Fishing pressure reached its highest levels during the recent period of 2016–2018 and these results are consistent when considering stock assessment at the national level. However, the length-based model (LBB) indicates a deterioration in average length, particularly in Senegal and the Gambia, especially since 2014. Adults, who migrate, are more abundant in Mauritania. Therefore, due to its high market value, white grouper is increasingly targeted, resulting in an unprecedented rise in fishing mortality over the past decade, particularly among larger individuals, with the most significant pressure observed in Senegal and the Gambia. Given that the stock spans three countries, a coordinated management approach at the stock level is essential. However, management measures must also be adapted to the specific status of the population within each country. Without such a concerted effort, this trend is likely to persist, further endangering the resource.

1. Introduction

The sustainability of marine fisheries has become a global concern due to increasing fishing pressure and unregulated exploitation, which can lead to negative consequences for ecosystems and societies [1]. This issue is particularly evident in developing countries where appropriate management tools are lacking as seen in the Mauritania–Senegalese zone. This region, part of West Africa, is a hotspot with high productivity of fishery resources due to intense upwelling. Fisheries management in this region urgently needs significant improvement since 60% of demersal stocks assessed by FAO CECAF are still overexploited [2]. The ideal stock assessment would be able to estimate all of the key parameters related to population processes within a framework that assigns appropriate weight to the data, fits the data adequately, and captures all sources of uncertainty related to estimation, including model uncertainty, process uncertainty, and observation uncertainty [3].

The white grouper (Epinephelus aeneus) is an iconic coastal demersal species of West Africa belonging to the Serranidae family and one of the most sought-after demersal fish. In Senegal, the Gambia, and Mauritania, the white grouper (or thiof in Senegalese local language) is a voracious predator that feeds on fish, cephalopods, and crustaceans. Adults’ distribution extends across depths ranging from 20 to 200 m mostly on rocky seabeds on the continental shelf while juveniles (less than 30 cm) are concentrated along the coastline, particularly in estuaries [4]. The two main spawning areas are off the coast of Senegal and the southern part of the Baie du Lévrier in Mauritania [5]. Furthermore, a recent study shows the importance of the coastal fringe as an essential habitat for juvenile white grouper [6].

White grouper is intensively exploited by small-scale fisheries, coastal and offshore to a lesser extent [7]. The Senegalese and Mauritanian fisheries exploit the same white grouper species but have relatively different exploitation histories. In Senegal, the white grouper fishery is very old (operating since the 1970s), while it is relatively recent in Mauritania (in action since the 1980s). However, in terms of catches, small-scale fishery represents more than 70% of the catches of white grouper in the sub-region. Moreover, this fishery uses more selective gears such as lines and longlines which capture larger individuals. On the other hand, the nets used by the artisanal fishery and industrial fishery trawls tend to capture smaller individuals.

Champagnat and Domain [8] assume that adults of this population migrate from Cap Blanc on the northern border of Mauritania to Cape Roxo in southern Senegal, and as such, consider it as a single stock. Nevertheless, mainly for practical reasons (i.e., the difficulty of merging data from these two countries), all attempts to assess the status of the stock have, until recently, been conducted at the national level, i.e., in Mauritania by [9] and in Senegal by [7,10]. On the other hand, the Fishery Committee for the Eastern Central Atlantic (CECAF-FAO) working group regularly assesses white grouper as a single stock between Mauritania, Senegal, and the Gambia while giving very little attention to length frequencies data and formulating advices and recommendations on exploitation only at a regional level and by considering management by focusing on fishing pressure through catch levels. However, no TACs or quotas are established at sub-regional and national levels and the only measures edited in the national fishing code focus on length-at-catch.

Hence, the challenge of this paper was to carry out an assessment of the scale of the North-West African sub-region by combining different stock assessment methods (i.e., catch and length-based model) in order to show its relevance for fisheries management in a data-poor context. Indeed, the quality of data on small-scale fisheries requires appropriate approaches capable of producing reliable management advice. This work is thus based on a large compilation of available data from three countries’ (Mauritania, Senegal, and the Gambia) databases and complemented by the collection of length frequencies data within the framework of the European Union Demerstem/PESCAO project. Thus, two stock-assessment models are used: (i) the surplus production model fitted in a Bayesian statistical framework called JABBA (Just Another Bayesian Biomass Assessment; [11]) and (ii) the Length-based Bayesian Biomass (LBB) were used. The JABBA model used updated catches and CPUE (Catch per unit effort) time series standardized from 1980 to 2019 and series of abundance indices from scientific campaigns standardized by the Delta-GLM model over the period 1980–2019. The Length-based Bayesian Biomass (LBB) approach is used to analyze length frequency data from exploited fish or invertebrate populations in which all relevant parameters are estimated synchronously using the Bayesian Monte Carlo Markov Chain (MCMC) approach [12,13]. The LBB model is used to enrich the results and makes it possible to address the appropriate management measures based on the size limits and the impact of the gear used in each country. These models were fitted to data in order to (1) quantify fishing pressure on this resource by applying recent and appropriate stock assessment tools, (2) establish the diagnosis of the status of white grouper stock, (3) illustrate the relevance of combining different approaches by indicating countries’ exploitation rate specificities and then (4) propose the most appropriate measures that should be prioritized for an ecosystem-based approach and efficient fisheries management at national and sub-regional levels.

2. Data and Methods

2.1. Data Used

Several data sources were used for this study from both countries: Catch and length frequency data from scientific surveys and commercial fishery.

Table 1. Data summary.

Country	Datatype	Dataset	Period	Remarks on Missing Years
Senegal	Catch	Scientific Survey	1971–2016	Regularly missing (50%)
		Commercial	1974–2018	∅
	Length Frequencies	Scientific Survey	1987–2016	Regularly missing (50%)
		Commercial	2004–2020	∅
Gambia	Catch	Commercial	1990–2018	∅
Mauritania	Catch	Scientific Survey	1982–2019	Seldom missing (5%)
		Commercial	2006–2018	∅
	Length Frequencies	Scientific Survey	1987–2018	Occasionally missing (10%)
		Commercial	2019–2020	∅

presents a summary of all data and their availability over the different periods. This species is mainly targeted by artisanal fisheries; however, it is fished incidentally by offshore fisheries, which makes it difficult to estimate its catch in these offshore fisheries, because the catches are reported in aggregated species categories unlike in the artisanal fisheries databases. However, even the artisanal statistical monitoring and sampling systems often suffer from inadequate spatial and temporal coverage, posing significant challenges to data reliability and comprehensiveness. To fill these gaps, among the alternatives adopted was the collection of size-frequency data using a harmonized data collection framework. These data were used for alternative assessment methods that are more suited to data-poor contexts. The data from scientific campaigns available in Mauritania and Senegal are also highlighted and used to describe abundance trends and model calibration. Furthermore, it should be noted that the Gambia is taken into account in the scientific sampling conducted by scientists from Senegal, including scientific campaigns at sea and the collection of size frequency data at landing. The Gambia does not have specific scientific campaigns, but its geographical position surrounded to the north and south by Senegal creates a kind of continuity in the areas. Senegalese scientists usually carry out sampling in both countries. In general, the scientific data from Senegal cover the Gambian area

2.1.1. Catch Data

• Scientific surveys

In Mauritania, scientific surveys include all demersal surveys conducted along the Mauritanian continental shelf from 1982 to 2019, totaling 89 surveys and 7496 trawls. Two vessels were used during this period, N’Diago (1982–1996; 478 kW) and Al-Awam (1997–2019; 735 kW) and recent years were not sampled for vessel availability issues. Each observation corresponds to one trawl with an average horizontal opening of 17 m dragged at a trawl speed of 3.2 knots, for a standard duration of 30 min. The fishing gear was modified in 1989 and hence the correction coefficients estimated by [14], using data from inter-calibration surveys reported by [15], were applied to data gathered prior to that date.

Since 1970, scientific surveys have been conducted on the Senegalese continental shelf. While the oldest ones mainly focused on the study of fish biology, such as L. Amaro (1970–1974), since the 1980s surveys carried out by L. Sauger (1986–1999; 175 kW) and I. Deme (2001–2016; 190 kW) aimed to provide a fishery-independent measure of fish distribution and abundance for all species that can be sampled using bottom trawl [10]. A total of 24 surveys covering the whole continental shelf, from 10 to 200 m deep, and using the same sampling protocol, were considered in the current analyses. They cover a 23-year period of time, from 1986 to 2016 (with missing years, Table 1), and gather 2675 hauls. Catches per trawl haul (30 min) were used to estimate survey abundance indices. The Gambian zone is covered by Senegal’s scientific surveys.

• Commercial data

The Mauritanian, Senegalese, and Gambian artisanal fisheries data are respectively collected by IMROP (2006–2018), CRODT (1974–2018), and the Gambian Fisheries Department (1990–2018). They are aggregated by year, month, landing site, and fishing gear over the period.

Given the multitude and diversity of artisanal fishing gears, only those targeting white grouper (high landings of white grouper) have been retained and grouped into homogeneous and coherent sets to estimate the CPUE:

In Senegal:

-

Lines: The pirogue glacier line (LPG); the octopus’s line (LPO); the single line of motorized pirogues (LSM); the single line of non-motorized pirogues (LSNM) and the longline (PAL);

-

Nets: The bottom set net (FD), the bottom drift gillnet (FMDF), and the trammel net (TM).

In Mauritania:

-

Handline targeting demersal fish

-

Jig to target the Octopus vulgaris and demersal fish

2.1.2. Length Frequencies Data

• Scientific survey

In both countries, several scientific survey sampling length frequencies were conducted between 1987 and 2019. However, there is a lack of data for the earlier years.

• Commercial data

In Senegal, the data provided are landings from small-scale fisheries targeting white grouper for the period 1974–2018. The gear with the most consistency and availability were the lines. Samples were selected upon the regions with the highest sampling coverage (Cayar, Saint Louis, and Saloum) and extrapolated with the annual catch. Data were aggregated by 4-years to produce homogeneous groups with the same data size. In addition, a biological sampling was conducted during the Demerstem project (2019–2020), and data collected for the three countries were added to the analysis. Mauritania data were collected at the artisanal port of Nouadhibou.

2.2. Abundance Indices

In this study, the spatio-temporal dynamics of the stock abundance are analyzed using linear modeling of the Delta-GLM type. The Delta-GLM model is used to calculate annual abundance indices and determine a spatio-seasonal distribution pattern for thiof. These abundance indices are then incorporated into the JABBA surplus production model [11] to conduct a stock assessment.

Abundance indices are generally estimated from observation data from scientific surveys or from fishing statistics from commercial fleets. In both cases, estimating indices of abundance remains a relatively complex task, particularly in view of the often highly unbalanced sampling design. It is therefore clear that the methods used are often not statistically optimal, which can lead to biased estimates and/or very high uncertainties [16]. In addition, data with a high prevalence of zeroes and skewed distributions can be problematic for fitting typical distributions and modeling the effects of factors. One solution to this problem is to apply a Delta-GLM model [17,18]. This model is obtained by combining two sub-models: (i) one to model the absence and presence of the species in the trawl hauls; (ii) the other to model positive values of CPUE [9,10]. Hence, the delta model is the product of a binomial GLM sub-model, which estimates the presence/absence probability of each taxon/species during a sea trip or trawl haul and a Gaussian GLM sub-model on log-transformed non-null values of raw CPUE or densities per trawls corrected with [19] formula in order to obtain unbiased values of the expected abundance index (for more details, see [9].

For standardization issues among countries, abundance indices were analyzed and produced separately. Hence, for each country, two series of annual abundance indices were estimated using a Delta-GLM; one is based on CPUE from commercial artisanal fisheries data, the other from scientific surveys.

Variables and their modality used in the calculation of abundance indices using the Delta-GLM model are presented in

Table 2. Variables and modality inputs for Delta-GLM models.

Country	Dataset	Variables and Modality
Senegal	Scientific survey	year, month, season, area (Sud, Small Coast, North), depth (5–10 m, 10–20 m, 20–40 m, 40–60 m, 60–100 m, 100–200 m)
	Artisanal fishery	year, month, season, area (Dakar, Small Coast, Great Coast), fishing gear (handline, fixed bottom net)
Mauritania	Scientific survey	year, month, season, area (Sud, Center, North), depth (5–20 m, 20–40 m, 40–60 m, 60–100 m)
	Artisanal fishery	year, month, season, area (Dakar, Small Coast, Great Coast), fishing gear (handline, hooks)

. For each model, the variables with significant deviances were retained in the estimation of abundance indices and the model with the lowest AIC was considered.

In addition to the abundance series, maps were produced by extracting abundance indices by strata and region from the scientific models to examine spatio-temporal distribution patterns over a 10- and a 4-year period.

3. Stock Assessment Methods

JABBA presents a unifying, flexible framework for biomass dynamic modeling, runs quickly, and generates reproducible stock status estimates and diagnostic tools [11]. Its flexibility is used here to estimate some model parameters before and after some changes in the research vessels. The LBB is used to complement the JABBA diagnostic with the output of a size-based model to obtain information on the distribution of fishing mortality across different size classes and to measure changes in the average size of the species. Indeed, fisheries exhibit different selectivity which might affect exploitation rates.

3.1. Just Another Bayesian Biomass Assessment (JABBA)

JABBA is a documented open-source R package (https://github.com/jabbamodel/JABBA-Select accessed on 14 June 2024) that has been applied recently on several ICCAT stocks like swordfish, Atlantic shortfin mako swordfish albacore, blue marlin, Atlantic white marlin, yellowfin tuna, and Atlantic bigeye tuna [11].

• A state space production model in a Bayesian framework.

The stock assessment is performed in a Bayesian framework using the JABBA package [11] which allows an analysis based on the Pella Tomlinson model with the following equation:

$$g\left(B_t\right)={\displaystyle \frac{r}{m-1}}·B_t(1-{\left({\displaystyle \frac{B_t}{K}}\right)}^{m-1})$$

(1)

With r the intrinsic growth rate, K the biotic capacity of the environment and m a shape parameter determining for which ratio B/K, the maximum surplus production is reached.

Based on the work of [20], the process equation is rewritten in a stochastic model with variables expressed as a biomass depletion formulated as the proportion of the stock biomass at time t to its biomass in the pristine state ($P_t$= ${\displaystyle \frac{{ B}_t}{K}}$). The initial biomass depletion is estimated by introducing the parameter. The latent space equation is then:

$$P_t=\left\{\begin{array}{c}\varphi ·e^{{\eta }_t}; t=1 \\ P_{t-1}+{\displaystyle \frac{r}{m-1}} · P_{t-1} \left(1-{P_{t-1}}^{m-1}\right)-{\displaystyle \frac{C_t}{K}}·e^{{\eta }_t}; t>1\end{array}\right.$$

(2)

where ${\eta }_t$ is the process error, such as ${\eta }_t ~ Normal(0,{\sigma }_{\eta }^2)$, $C_t$ is the catch of the year $t$. The observational model connects the latent space to the abundance index $I_{t,i}$ assuming it is proportional to the biomass. The observation equation is then:

$$I_{t,i}=q·B_t·e^{{ϵ}_{t,i}}$$

(3)

With $q$ the catchability coefficient and ${ϵ}_{t, i}$ the observation error such as ${ϵ}_{t, i} ~ Normal(0, {\sigma }_{{ϵ}_{t, i}}^2)$ with ${ϵ}_{t, i}$ the variance of the observation error in year $t$ for the AI serie $i$. The full latent space Bayesian model requires a joint probability distribution over all unobservable hyper-parameters Θ$(K,r,q,\varphi ,{\sigma }_{\eta }^2,{\sigma }_{{ϵ}_{t, i}}^2)$ and the $n$ process errors related to the vector of latent states $\eta =\left\{{\eta }_1, {\eta }_2, \dots , {\eta }_t\right\}$ related to all observational data in the form of relative abundance indices $i$,$I_i=\left\{I_{i,1}, I_{i,2}, \dots , I_{i,t}\right\}$.

• Priors specification

As suggested by Wi [11], most priors are stock-specified. Informative priors were considered for $r, m,$ and $\varphi$, whereas other parameters were a priori distributed following weekly informative distributions.

Table 3. Priors of the Bayesian dynamic production model.

Priors	Distribution
$K$, carrying capacity	$\mathrm{L}\mathrm{o}\mathrm{g}\mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l} (\mathrm{30,000},0.55)$
q, catchability	Range [0–10]
$r$, intrinsic rate of stock growth	$\mathrm{L}\mathrm{o}\mathrm{g}\mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l} (0.3,0.1)$
$\varphi$, initial depletion rate (φ = B1974/K)	$\mathrm{L}\mathrm{o}\mathrm{g}\mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l} (0.9,0.1)$
$m$, shape parameter of the Pella–Tomlinson model	$\mathrm{L}\mathrm{o}\mathrm{g}\mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l} (0.5,0.1)$
${\sigma }_{\eta }^2$, process variance (default)	$1/\mathrm{g}\mathrm{a}\mathrm{m}\mathrm{m}\mathrm{a} (4,0.01)$

describes the priors specified for each parameter. For $r$, this distribution corresponds to the value calculated in FishBase (http://www.fishbase.org, accessed on 20 June 2022) with a FishLife package. Concerning $m$, it was established from previous analysis in pseudo-equilibrium. $K$ was implemented considering results from [7,9], which both described a value around 25,000 tons for the $K$ in their country. In the current study, we considered a value of 30,000 with a lognormal distribution of strong right asymmetry for a consideration of values above the average. All catchability parameters were formulated as uninformative uniform priors.

• Input fishery data

In addition to a total catch time series from the Gambia, Mauritania, and Senegal, four abundance indexes were made available from the Delta-GLM analysis (Figure 1) in the 1974–2018 period. Based on the empirical equation described by [21], abundance indices from commercial CPUE are corrected by assuming a yearly increase of mean fishing power ${\alpha }_{com}$.

$${\alpha }_{com}=13.8\cdot (Y_n-Y_{t0})-0.511$$

(4)

$${{IA}_{t\_corr}={IA}_t\cdot {(1+{\alpha }_{com})}^t-t}_0$$

(5)

where $t_0$ and $n$ are the first and last year of the data series, and ${\alpha }_{com}$ is the annual rate of increase in the artisanal fishery fishing power. A creep in fishing power allows us to include an intensification of fishing effort through improvements in techniques (e.g., experience of fishermen, organization of work, etc.) and/or technologies (e.g., evolution of gear, improvement of equipment such as GPS, motorization) leading to an increase in the efficiency of the nominal unit of fishing effort [22].

Considering Equation (4) given an increase in average creeping of vessels’ fishing power, three scenarios are computed to assess the sensitivity of the results depending on the value of ${\alpha }_{scenario}$: 1% for an optimistic case, $(2\times {\alpha }_{com}-1)$ for a pessimistic case, leading to 3% for Artisanal_SEN and 7% for Artisanal_MRT.

In both countries, the change of vessels, that occurred respectively in 1996 and 2000 in Mauritania and Senegal, did not result in an inter-calibration survey [14]. Hence, by defining two-time blocks (early and late) for each survey, the model will estimate a different catchability. In addition, by observing the strong variations in the Mauritanian survey abundance index, a bigger fixed CV is implemented for the ‘early’ period, as well as for the year 2004 and 2005, and the commercial CPUE (

Table 4. Deviance analysis for GLM models of abundance index.

Country	Data Used	Model	% Deviance
			Year	Season	Depth	Area	Fishing Gear	Year: Area	Area: Depth	Total
Senegal	Survey	IA 0/1	5.9	0.7	3.8	2.8		4.0		17.3
		IA +	10.5		12.4	5.8		7.2	4.8	40.8
	Artisanal	IA 0/1	10.4			2.4	1.2	2.8		16.9
		IA +	4.6	0.9		4.3	3.8	4.9		18.1
Mauritania	Survey	IA 0/1	5.9		5.8	0.7			2.6	15
		IA +	17.1							17.1
	Artisanal	IA 0/1	3.1			9.4	25.6			38.2
		IA +	8.5			36.3	29			63.5

). A fixed observation error of 0.1 is added for each abundance index.

• Retrospective and hindcast cross-validating analysis

To evaluate CPUE fits, the model-predicted CPUE indices were compared to the observed CPUE. JABBA residual plots were also examined, and the randomness of model residuals was evaluated by means of the Root Mean-Squared-Error (RMSE; [23]).

Finally, to verify systematic bias in the estimation of B or F, a retrospective analysis was conducted for the last 8 years of the assessment to evaluate whether there were any strong changes in model results based on data availability. The selected period for the retrospective was intended to cover the most recent trend identified in Figure S1 (2010–2018). In addition, hindcasting analysis was also conducted over the same period to assess the model predictive skills by computing the MASE score [24].

Yt is the one-step-ahead forecast of the expected value for the observation at time t based on the model conditioned with data up to time t − 1.

A MASE score > 1 indicates that the average model forecasts are worse than a random walk. Conversely, a MASE score of 0.5 indicates that the model forecasts twice as accurately as a naïve baseline prediction; thus, the model has prediction skill [25].

3.2. Length-Based Bayesian Biomass

LBB (LBB_13.R. accessed on 10 June 2022) is a new Bayesian size-based approach [12] for estimating stock status in data-constrained situations. LBB requires no input data in addition to size frequency data, but offers the user the possibility to specify a priori distributions for the asymptotic estimable parameter size ($L_{\infty }$), length at first capture ($L_c$) and relative natural mortality ($M/K$). In addition, relative fishing mortality ($F/M$) is estimated as the average of the age range represented in the size frequency sample. Using these parameters as inputs, standard fishing equations can be used to estimate depletion or the currently exploited biomass relative to the unexploited biomass ($B/B_0$). In addition, these parameters can be used to estimate the length at first catch that will maximize catch and biomass for a given fishing effort ($L_{c\_opt}$) and to estimate an approximation of the relative biomass capable of producing maximum balanced catches ($B_{MSY}/B_0$).

The following equation describes the framework for estimating stock status from $L_{\infty }, M/K, F/K$ and $Lc$ [26,27]. First, given the estimates of $L_{\infty }$ and $M/K$, $L_{opt}$, i.e., the size at which the cohort biomass is at its maximum, can be obtained from Equation (6):

$$L_{opt}=L_{\mathrm{\infty }}\left({\displaystyle \frac{3}{3+\frac{M}{K}}}\right)+{\displaystyle \frac{M}{K}}$$

(6)

Based on Equation (7) and a given fishing pressure ($F/M$), the average length at first catch, which maximizes the catch and biomass ($L_{c\_opt}$), can be obtained from:

$$L_{c\_opt}={\displaystyle \frac{L_{\mathrm{\infty }}(2+3\frac{F}{M})}{(1+\frac{F}{M})(3+\frac{M}{K})}}$$

(7)

Then, $L_{c\_opt}$ estimates are used to calculate an approximation of the relative biomass that can produce MSY [12].

A sensitivity analysis was performed on the Linf values. The values used are respectively 106 cm, 94 cm, and a value calculated by LBB by default. The first value comes from the results of growth analysis by the Shepherd method applied to the size frequency data collected by the Demerstem project between 2019 and 2020 in Mauritania and the second value comes from the application of the Shepherds method on data from Senegal.

4. Results

4.1. Delta-GLM Model and Abundance Indices

The Delta-GLM fitted to commercial CPUEs and to scientific surveys data indicates the existence of significant interannual change and provides consistent trends among countries and between data sources, except in recent years where dissimilarities are denoted (

Table 5. Abundance index used in JABBA.

Country	Abundance Indices	CV	Remarks
	Scientific_Survey_early	0.15	∅
Senegal	Scientific_Survey_late	0.15	∅
	Artisanal	0.25	Fish creep: 1–7%
	Scientific_Survey_early	0.2	∅
Mauritania	Scientific_Survey_late	0.15	CV of 0.5 for 2004–2005
	Artisanal	0.25	Fish creep: 1–3%

). In fact, AI reduce from 1975 to 2005, when minimal values are observed, followed by an increase until 2014 (Figure S1). Then, the signals are quite different, with a 2-year time lapse between Senegal when comparing the Mauritanian scientific survey with Senegalese artisanal CPUEs, which is ahead of Mauritania and where the artisanal CPUE indicates a constant increase.

Maps of mean density and biomass over a 10-year period denote the same signals (Figure 1), with higher values in Mauritania in the early period, before falling for 20 years and finally rising in recent years. It also appears the density and biomass are higher in the deeper strata in Senegal, except in the North, where the pattern is a mix between Senegal and Mauritania with the highest density in the first layer, that is increasing again to around a 60 to 100 m depth, hence highlighting a transition zone in the predicted spatial distribution between the two countries.

4.2. Evaluation

4.2.1. JABBA

The JABBA model selected as the base case for the white grouper in the studied area was established by including all CPUE indices available and splitting the Mauritanian and Senegalese surveys into two CPUEs, resulting in an improved RMSE and the model showed a significant difference in catchability between the defined time blocks in Mauritania and a slight one in Senegal (Figure S2). The model passed convergence and run test diagnostics (Figures S2 and S3), except correlated patterns in the ‘late’ Mauritanian scientific survey. Nevertheless, fitting presents some misspecifications (Figure 2).

For both scenarios (pessimistic and optimistic), models were able to converge adequately as judged by the [28,29] diagnostic tests implemented in JABBA [11]. Moreover, the visual inspection of trace plots of the key model parameters showed good mixing of the three chains (i.e., moving around the parameter space), also indicative of convergence of the MCMC chains and that the posterior distribution of the model parameters was adequately sampled with the MCMC simulations (Figure S5).

The RMSE value indicated a relatively low precise model (25.4%, Figure 2). The boxes indicated huge discrepancies between abundance indices mainly when the Mauritanian survey took place, denoting conflicting information. A Loess-smoother highlight slightly auto-correlated residual patterns in the beginning, and the residual run tests implemented in JABBA [25] showed there was no evidence (p > 0.05) to reject the hypothesis of randomly distributed residuals for all CPUE times series fit in both scenario models except for correlated patterns in the ‘late’ Mauritanian scientific survey, with a very few points exceeding the 3-sigma rule indicating outliers (Figure S3).

The fits of the abundance indices were varied but generally within restricted 95% confidence intervals (CIs). Overall, trends in the observed and predicted CPUE were notably consistent for most indices (Figure S4), with the exception of the late period (2010–2019) during which abundance trends shifted and, as stated previously, revealed conflicting information between countries and data sources (CPUE and survey). The routine JABBA output plot, showing the posteriors and the assumed prior distribution for the key parameters, provided no evidence of severe prior misspecification (Figure S2)

The management plot illustrated a stock severely over-exploited in both scenarios with a high fishing mortality in 2019 (with CI take values in a threshold between three- and eight-fold the $F_{MSY}$, Figure 3), and which has been constantly increasing since 2013. This has to be reviewed in conjunction with the biomass decreasing during the same period of 2013–2018. Reference points are presented in

Table 6. JABBA reference points.

Scenario	Reference Points
	K	B_msy	F_msy	MSY	F/Fmsy	B/Bmsy	B/K
Optimistic	43,578	11,811	0.27	3230	4.93	0.41	0.11
Intermediate	45,522	12,284	0.27	3329	5.79	0.34	0.09
Pessimistic	53,401	14,924	0.24	3609	6.84	0.26	0.07

.

The retrospective analysis of predicted fishing morality $F_t$ showed a consistently positive but small retrospective bias (Figure 4) with rho ρm = −0.04, which is within the acceptable thresholds for long-lived species [25]. In addition, the retrospective peels fall within the estimated 95% CI limits from the reference run, which confirms that the errors in biomass and fishing mortality estimates resulting from additional years of data being removed are consistent with estimate uncertainty. However, concerning predictive skills, in agreement with the issues raised previously, the model does not present predictive skills with a lack of consistency between CPUEs while abundance trends are shifting (Figure 2).

Hindcasting cross-validating (HCxval) was performed using one reference model and seven hindcast model runs (solid lines) relative to the expected catch-per-unit-effort (CPUE). The observations used for cross-validation are highlighted as color-coded solid circles with associated 95% CI (light gray shading). The model reference year refers to the endpoints of each one-year-ahead forecast and the corresponding observation (i.e., year of peel +1). The mean absolute scaled error (MASE) score associated with each CPUE and size composition time series is denoted in each panel (Figure 5).

4.2.2. LBB

An analysis was conducted on length frequencies of the commercial data. Considering the analysis done on the commercial data in Senegal and the Gambia (2004–2021), the estimate of ${\displaystyle \frac{F}{M}}>1$ confirms that the stock has been overexploited in recent years, while the estimate of ${\displaystyle \frac{B}{B0}}<0.25$ indicates that the current biomass has been extremely low since 2014 (

Table 7. LBB reference points.

Country	Sampling Period	Linf	Lmean/Lopt	Lc/Lc_opt	B/B0	B/BMSY	F/M	F/K	Z/K
Senegal and Gambia	2004–2006	95.2	0.72	0.47	0.83	1.8	0.1	0.1	1.1
	2007–2009	94.6	0.71	0.51	0.20	0.7	1.5	0.9	1.5
	2010–2012	94.5	0.72	0.43	0.81	1.9	0.1	0.1	1.1
	2013–2015	94.9	0.70	0.58	0.04	0.1	5.8	2.8	3.3
	2016–2018	91.9	0.59	0.47	0.1	0.2	2.4	2.4	3.4
	2019–2021	90.5	0.46	0.33	0.01	0.03	8.8	8.5	9.5
Mauritania	2019–2021	99.2	0.87	0.82	0.24	0.64	1.5	2	3.27

). Furthermore, the $L_{mean}/L_{opt}$ and $L_c/L_{c\_opt}$ ratios were less than unity, suggesting a truncated length structure and fishing of undersized individuals. The analysis in 2020 in Mauritania also indicates a similar signal in a less pronounced way (Table 7).

In addition, LBB outputs indicate great differences in mean length between countries, with bigger thiof catches in Mauritania (Figure 6).

In Senegal and the Gambia, both scientific survey and artisanal landings show a degrading mean length, in particular since 2014, reducing from 50 cm in early 2000 to 33 cm in 2020. On the contrary, mean length in Mauritania seems to keep a high value according to the scientific survey, which is higher in exploited sampling from artisanal fishery and will reach 60 cm in 2020. In fact, according to the LBB outputs using Demerstem data in 2020, Lc in Mauritania is about 50 cm while it only attains 30 cm in Senegal and the Gambia. The results of the LBB indicate a marked overexploitation of sizes in Senegal and the Gambia with a population dominated by average sizes twice as low as the optimal size (Lmean/Lopt = 0.46) unlike in the Mauritanian zone where Lmean/Lopt = 0.87 (therefore, an average size close to the optimal size; Figure S5).

The use of similar main gears in both countries excludes a difference in selectivity and highlights strong variations in the size structure between the Senegalese and the Mauritanian populations, which is consistent with the differences between F/M values, which are three-fold higher in Senegal and the Gambia.

However, the study data suffer from some information gaps: (1) The high and uneven distribution of missing data can undermine the reliability of the models; (2) The heterogeneity of data sources, which may introduce systematic bias. Nevertheless, the results obtained, overall, remain robust given the quantity and diversity of data and methods used. The advantage here is the availability of several series of abundance index data and over long periods (scientific campaigns in both countries, CPUEs of artisanal fishing in both countries) the model synthesizes the information contained in the different series and the gaps in certain series compensated by the information provided by other series. Heterogeneity can be seen as a factor of resilience of the results produced by the model.

5. Discussion

Our results indicate that management measures should differ from one country to the other. In fact, Delta-GLM models and associated maps allowed us to identify differences in spatial repartition between Senegal and Mauritania. In addition, exploitation stock status and indicators are different between countries with juvenile fish particularly subject to fishing pressure in Senegal, where the average catch size has recently dropped while it is still stable in Mauritania. This difference in exploitation patterns may have resulted from different origins such as fishing pressure in the different countries or population structure (spawning areas). The size segregation in the migration process has been suggested by biophysical modeling in West Africa [30].

Size-based approaches proved to be very good complementary assessment tools to catch-based methods. Indeed, the LBB model estimates optimal sizes at first capture (Lc_opt) of the order of 60 cm in Mauritania and 58 cm in Senegal and an optimal size (L_opt) of 69 cm in Mauritania and 61 cm in Senegal. These reference points seem to be high compared to the average size exploited in Senegal unlike in Mauritania where the average size exploited is equal to the Lc_opt estimated by the LBB. The minimum size authorized in Mauritania for the thiof is 40 cm in total length; in Senegal, the regulatory length is 30 cm.

Results from the dynamic model regardless of the scenario were consistent with length frequency analysis, both methods indicating a severe over-exploitation since 2014. The results obtained with the JABBA model indicate a strong fishing pressure with a fishing mortality five to seven times higher than Fmsy while the current biomass represents 34% of the biomass corresponding to MSY (Bmsy) and only 9% of the virgin biomass. These results show that the state of thiof has further deteriorated compared to the assessments conducted between 2005 and 2013 by [9] who estimated an excess effort between 40% and 80% compared to the effort in 2010 in Mauritania. This was also consistent with previous estimates for this species in Senegal, where the excess was estimated by [10] to vary from 35% to 55%.

However, at the level of the two countries, a severe overexploitation of biomass is revealed by the Bayesian biomass production surplus model (JABBA). Stock status indicators show a state of overfishing with high fishing mortality (B/Bmsy = 0.34 and F/Fmsy = 5.9).

In addition, the current F described using JABBA is similar to the one observed in Senegal with LBB. Even though this is mainly a consequence of important catches in Mauritania in recent years cumulated with decreasing abundance indices, similar results are obtained when processing a stock assessment only on the Senegal/Gambia scale (Figure 7). These results are particularly coherent with trends identified in the Artisanal CPUE in Senegal, specifying a drop in abundance around the same period hence denoting a change in the functioning of the fishery. In addition, abundance indices report a delay in the trends between countries with Senegal 2–3 years in advance.

One of the main benefits of the approach presented here is to show how analysis of various data sources and the combination of several methods can reinforce a stock assessment analysis, even in the context of poor data availability (Figure 7). This multi-pronged approach allows the use of various independent sources of data and the use of size-structured models in the present study reinforced the results obtained with the surplus production model JABBA, while providing further information on the disparity in exploited sizes in the two countries.

As stated by [31], in a review of data-limited methods, obtaining better data is as important as using care in acknowledging and interpreting uncertainties or developing harvest strategies (including control rules). For many small-scale fisheries, obtaining reliable time series on historical total catch is difficult, whereas sampling lengths from the catch is easier and for fisheries where the time series of catch are unavailable or catches are not consistently monitored and managed, using length-composition data can provide good approximations of the status of the stock, in particular for medium-lived species. Furthermore, many data-limited approaches have been developed to meet an increasing demand for science-based fisheries management of unassessed fisheries where data and resources are limited [22,32,33,34,35,36,37] and contribute further to advanced hindsight in the use of length-based assessment and catch-based models by comparing the performance in determining stock status [37,38,39]. For the LBB in particular, as reported by [40], relying on the assumption that the ratio of natural mortality (M) to the von Bertalanffy growth parameter (K; M/K) is typically around 1.5 while they argue there is strong evidence to support the claim that M/K is outside the narrow bounds of 1.2–1.8 for many exploited species is an important issue of this method. In addition, the estimations of L∞ can be misleading due to the approximation using the maximum observed length Lmax when individuals that are bigger than L+∞, so the LBB estimated higher values for this parameter than the true values [40,41]. However, in our case, the sensitivity analysis of the model to the asymptotic size (L+∞) did not reveal major changes in the results of the LBB model.

6. Conclusions

Overall, the trajectory of the state of the stock illustrated by the Kobe figure suggests an overexploitation of the biomass of the white grouper stock since the 1990s and that the fishing pressure has never been as high as it was over the recent period 2016–2018. This period coincides with an increase in overall fishing efforts in the Mauritanian zone since the implementation of the new management policy based on individual TACs and quotas for offshore and coastal fishing. Artisanal fishing remains for the moment within the framework of management by a global quota to fish for cephalopods. In Senegal, the fishing effort remains globally high, with a higher pressure from small-scale fishing than from industrial fishing. This species is particularly targeted by the different artisanal gears, i.e., lines, longlines, and passive nets. More than 80% of the landings are provided by small-scale fishery. Unfortunately, apart from the measures edited in the fishing code (first catch size of 30 cm TL), there is no management of this resource by TACs or quotas and the fishing effort is slowly monitored. However, given its market value, thiof is increasingly sought after, which explains the unprecedented level of fishing mortality during the past decade.

However, both evaluation methods (JABBA and LBB) have their limitations. The JABBA model uses annual landings data, aggregated at the national level. Specific landings by size or age class are therefore unknown. What is more, these data often do not represent real catches, as they are essentially declared landing statistics. As for the LBB model, it uses size frequency data collected at certain landing sites. This raises the question of the representativeness of the data in relation to real catches.

Therefore, given the different histories of exploitation of white grouper between Senegal (lines, nets, and trawls) and Mauritania (lines and trawls), it is likely that the exploited populations do not present the same situation.