nueva página del texto (beta)
Inglés (pdf)
Artículo en XML
Referencias del artículo
Traducción automática
Enviar artículo por email
Citado por SciELO 
Similares en
SciELO 
Permalink
INTRODUCTION
The purse seine fishery is the most important tuna fishery in the world, accounting for 66%-76% of the global tuna catch in 2010 (Dreyfus-León et al. 2015, McCluney et al. 2019). Of the global tuna catch, the contribution made by the Pacific Ocean fishery is by far the most important (Miyake et al. 2004). Around 21% of tuna catches within the Pacific are made in the eastern Pacific Ocean (EPO), where the purse seine fleet targets mainly yellowfin (Thunnus albacares) and skipjack (Katsuwonus pelamis) and tracks tuna schools using 3 main fishing indicators: dolphins, free-swimming schools, and floating objects (IATTC 2017). In the EPO most yellowfin tuna are caught when sets are made on dolphins and in 2 main high-catch areas, one near the Central American coast (~10º S, 10º N) and the other near the Baja California Peninsula (centered around 20º N, 115º W) (IATTC 2017). Most skipjack tuna, on the other hand, are caught in sets on floating objects south of 5º N and east of 110º W (IATTC 2017).
The tuna species targeted by the purse seine fleet are generally assumed to be capable of performing long-distance migrations (see, for example, Itoh et al. 2003), although the movement patterns of most species are not yet fully understood. Some studies, however, have demonstrated that some of the most important tuna species show some degree of fidelity to certain areas, a behavior that could result in patchy population distribution patterns. For example, tagging studies of yellowfin tuna have shown that individuals usually remain within 1,800 km of the release location (IATTC 2017). The discrete-population hypothesis for yellowfin tuna inhabiting the Pacific Ocean is also supported by the global genetic analysis conducted by Pecoraro et al. (2018).
On another note, the horizontal movements of the skipjack tuna appear to be restricted to a few hundred kilometers, although it is possible that there is only one skipjack population in the Pacific Ocean and that the exchange of individuals is thus occurring at local scales (IATTC 2017). The rate of exchange of skipjack individuals between the EPO and other areas in the Pacific Ocean cannot be quantified with the tagging data that is currently available (IATTC 2017).
Recent studies have shown that fishers track small-scale oceanographic features where important tuna congregations occur, such as thermal fronts and eddies (Scales et al. 2018). Torres-Orozco et al. (2005) suggested that yellowfin tuna occupied the warm side of the thermal fronts that occurred in the EPO off Cabo Corrientes (~20º N, 106º W), Mexico. Jiménez-Tello (2014) showed that most unassociated purse seine sets targeting yellowfin tuna in the Pacific Ocean, off northwestern Mexico, were carried out in waters with sea surface temperatures of ~22 ºC and negative sea surface height anomalies, indicating that yellowfin tuna schools are associated with cyclonic eddies. Mugo et al. (2010) suggested that the skipjack tuna preferred hábitat is bounded by sea surface temperatures between 20.5 and 26.0 ºC, relatively low chlorophyll a concentrations, and positive sea surface height anomalies, indicating that the skipjack tuna likewise inhabits zones with thermal fronts and eddies.
The purse seine fleet operating in the EPO is perhaps one of the most technologically advanced fishing fleets in the world because it uses state-of-the-art technology (helicopters, speedboats, bird sonars, fish aggregating devices equipped with GPS, and echo sounders) to track tunas (IATTC 2004). Despite the use of advanced technology, purse seine maneuvers often result in null sets (purse seine maneuvers where the retained tuna catch is zero), even when tunas are numerous and concentrated in small áreas (Fonteneau et al. 2008).
Previous works have demonstrated that fishing on free-swimming tuna schools is an important cause of the occurrence of null sets (Sarralde et al. 2005, Dreyfus-León and Mejía 2009, Chassot et al. 2013). However, and despite the obvious economic impact of zero-catch purse seine maneuvers, little research has been conducted to elucidate which environmental factors are influencing the occurrence of these phenomena. For the Spanish and French purse seine fleets operating in the Indian Ocean, Guillotreau et al. (2011) suggested that fisher preference for free-swimming or fish-aggregating-device (FAD) sets (and the associated percentage of null sets, which is higher on free-swimming schools) can vary with certain environmental events, such as El Niño/Southern Oscillation (ENSO), but they also observed that this variation was high with changes in available technology and the experience of fishers. Dreyfus-León et al. (2015) concluded that the deepening of the thermocline was the main cause of null sets in Pacific Ocean waters off Mexico and Central America, that these type of sets were more likely to occur when made on free-swimming schools during ENSO events, and that most of these sets occurred at the entrance to the Gulf of California.
Different statistical approaches have been used to link the distribution or presence/absence of large pelagic fishes to environmental features. For example, with data from the Japanese longline fishery and satellite-derived environmental data, Zainuddin et al. (2008) used a combination of generalized linear models (GLMs) and generalized additive models (GAMs) to predict potential albacore (Thunnus alalunga) fishing grounds. Using incidental catch data from the purse seine fleet in the eastern Tropical Pacific Ocean, Martínez-Rincón et al. (2012) found that GAMs and boosted regression trees had nearly identical predictive performance when modeling the probability of occurrence of wahoo (Acanthocybium solandri). Marín-Enríquez et al. (2018) used a presence/absence GAM approach to depict the potential hot spots and migration patterns of dolphinfish (Coryphaena spp.) in the EPO off Mexico, also using incidental catch data from the purse seine tuna fishery and satellite-derived oceanographic data.
The main working hypothesis of this study is that variations in the marine environment are the main cause of the occurrence of null sets made by the purse seine fleet that operated in the EPO from 2003 to 2015. A logistic GAM was fitted to data on the presence/absence of null cells in order to assess how and to what extent the probability of occurrence of null cells is influenced by variations in the marine environment and by spatial and operational factors inherent of the tuna purse seine fleet. The final GAM presented in this paper could serve as a decision-making tool to minimize the number of purse seine sets that are required to reach the fixed fishing quotas set for the purse seine fleet. The operational costs for the purse seine fleet are high; for example, the mean cost of a purse seine set carried out by an Ecuadorian vessel is around 4,120 USD (mean total operational cost of 642,700 USD per year per vessel, mean of 6 trips per year per vessel, mean of 26 sets per trip per vessel), and the mean catch per set is 20 t (Bucaram 2017). Therefore, besides the obvious economic impact (minimizing fishing effort would certainly result in lowering the operational cost of each fishing trip), minimizing the fishing effort exerted by the purse seine fleet could also have an important ecological impact because the bycatch from purse seine maneuvers, which includes several species of sharks, billfish, sea turtles, and marine mammals (IATTC 2017), would also be diminished.
MATERIALS AND METHODS
Study área
The study area covers the entire eastern Tropical Pacific Ocean (ETPO). In the northern area, large-scale circulation is dominated by the California Current, which is the Eastern extension of the North Pacific Subtropical Gyre (Badán 1997). Ocean circulation north and south of the equator is dominated by the Equatorial Current System. North of the equator, the North Equatorial Countercurrent flows eastward and the North Equatorial Current flows westward as the extension of the main branch of the California Current (Fiedler and Talley 2006). South of the equator, the South Equatorial Current is the main feature of the large-scale ocean circulation, and coastal circulation near South America is dominated by the Peru Current, also known as the Humboldt Current (Kessler 2006).
Several smaller-scale oceanographic features occur in the study area. In the northern zone the Eastern Pacific Warm Pool, centered at around 10º N and 100º W, is one of the main features and it is characterized by year-round high temperaturas (>27.5 ºC) (Fiedler and Talley 2006), as a result of strong stratification due to poor wind mixing (Wang and Enfield 2001). Three coastal mountain-gap wind jets occur seasonally in the study area, one in the Gulf of Tehuantepec (Mexico), one in the Gulf of Papagayo (Costa Rica), and the other one in the Gulf of Panama (Panama), and they are more evident during the boreal winter (Trasviña and Barton 2008) (Fig. 1). Centered at around 9º N and 90º W is the Costa Rica Dome, which is a ridge in the thermocline caused by anticyclonic circulation. This dome is affected by these coastal wind jets (Fiedler 2002). Associated with this feature, off Central America, is the Costa Rica Coastal Current, a northerly warm surface current that turns westward near the Gulf of Tehuantepec owing to the wind jets that occur in this región (Kessler 2006). The equatorial cold tongue, an extensión of the Peru Current, occurs south of the equator, producing strong thermal fronts in the equatorial upwelling system (Fiedler and Talley 2006).
Figure 1 Spatiotemporal distribution of fishing effort (number of sets) exerted by Class-6 tuna purse seine ships (>435 m3 storage capacity) from August 2003 to December 2015. Panels show data for the first (January-March) (a), second (April-June ) (b), third (July-September) (c), and fourth (October-December) (d) quarters of the year. Arrows in (a) and (b) depict the seasonal wind jets: Tehuantepec (TEH), Papagayo (PAP), and Panama (PAN). Color scale indicates number of sets. White color indicates zones where no fishing effort was observed.
Fishery database
We analyzed a database from the international tuna purse seine fleet operating in the EPO from August 2003 to December 2015. The database was obtained from the Inter-American Tropical Tuna Commission (IATTC) website (PublicPSTunaSetType 1958-2016). Data here were either gathered at sea by observers aboard the Class-6 purse seine tuna vessels (>435 m3 storage capacity) or extracted from vessel logbooks when observer data were not available. The database included the retained catch (in metric tons) for 6 tuna species, namely Pacific bluefin (Thunnus orientalis), black skipjack (Euthynnus lineatus), yellowfin (T. albacares), bigeye (Thunnus obesus), skipjack (K. pelamis), and albacore (T. alalunga); a tuna group labeled as Sarda, which included striped bonitos (Sarda chiliensis and Sarda orientalis); and a group classified as “unidentified tunas”. This database also included the number of purse seine sets, temporal (month, year) and spatial (latitude, longitude) variables, and the fishing indicator (SetType: dolphin, floating objects, and free-swimming schools), all aggregated in monthly 1º × 1º quadrants in a grid that covered the ETPO, from 24º S to 45º N and from 150º W to 70º W (Fig. 1). Because the IATTC database was aggregated in 1º × 1º quadrants, we were unable to analyze the occurrence of null sets. We therefore had to work at a coarser spatial level, defining null cells as those 1º × 1º quadrants where the total monthly retained catch was zero for all tuna species, including the unidentified group.
Environmental database
Monthly mean satellite-derived data of sea surface temperatura (SST, degrees Celsius), surface chlorophyll a concentration (Chla, milligrams per cubic meter), sea level anomaly (SLA, meters), and the Oceanic Niño Index (ONI) were used to assess possible environmental patterns in oceanic zones where null cells were more common. Both SST and Chla were obtained from the US National Oceanic and Atmospheric Administration’s ERDDAP data server (ERDDAP 2003-2016, 2003-2020), with a monthly temporal resolution and a 0.1º and 0.04º spatial resolution, respectively. Sea level anomaly data was obtained from the AVISO website (AVISO+ Satellite Altimetry Data 1993- 2020), with a weekly temporal resolution and a 0.25º spatial resolution. We decided to use SST because it affects the physiology, distribution, and vulnerability to certain fishing gears of different tuna species (Brill 1994). Surface Chla was used as a proxy for prey availability and to distinguish coastal from oceanic waters (Farrell et al. 2014), and SLA was used as an indicator of certain mesoscale ocean features that can influence thermocline depth, such as eddies and thermal fronts (Zainuddin et al. 2008). We also used ONI because El Niño conditions are known to alter thermocline depth (McPhaden 1999), and this variable has been highlighted as one of the main factors that are responsible for zero-catch tuna sets in northwestern Mexico (Dreyfus-León et al. 2015).
Statistical análisis
Model building
The general structure of a logistic GAM including continuous variables and a categorical variable is given by
where P is the binomial distribution probability of occurrence of null cells (1 if the retained catch in the i-th 1º × 1º cell was zero, and 0 otherwise), and it is modelled as the sum of n smooth functions (f) of the continuous variable x and the categorical variable z that has k levels.
For our model, the continuous variables were the environmental (SST, Chla, SLA, ONI), temporal (month, year), and operational (number of sets) variables and the categorical variable was the type of set (3 levels: dolphin, unassociated schools, floating objects). To minimize the problems associated with stepwise model building, we entered the covariates in an order that was established at the start of the modeling process, following the order suggested by Marín-Enríquez et al. (2018): (a) environmental variables; (b) operational variables (number of sets and type of set); (c) temporal variables, used to account for seasonal and interannual differences in the occurrence of null cells; and (d) spatial covariates, used to account for the spatial setting and to reduce the potential bias induced by spatial correlation. Additionally, we analyzed the interactions between latitude, longitude, and year to account for potential spatiotemporal differences in how the purse seine was set, which could affect the probability of occurrence of a null cell (Bigelow et al. 2002, Su et al. 2008). We retained those covariates that resulted in a decrease of at least 2 Akaike information criteria (AIC) units associated with significant statistical increase in explained deviance (Burnham and Anderson 2002). Because the use of covariates that are highly correlated can be confusing when interpreting modeling results (Zuur et al. 2009), we tested the models for potential multicollinearity amongst environmental variables using the concurvity function of the R mcgv package (Wood 2006). Model building was performed using the gam function of the mgcv package (Wood 2006) in the R environment (R Core Team 2016).
Model validation
Model validation was conducted using a random data splitting strategy. We split the database into 2 different datasets: the “training” dataset, which included 75% of the observations (chosen randomly), and the “validation” dataset, which included the remainder 25%. We fitted the GAM using the training dataset and then used the final model to make predictions based on the validation dataset. We used these predictions to evaluate model performance by calculating the área under the receiver operating characteristics (AUROC) curve, a value for the probability (0-1) of correctly discriminating absence from presence of null cell occurrence. This validation/ evaluation approach has been successfully used in similar studies (Farrell et al. 2014).
Spatial predictions
To highlight potential seasonal changes in spatial distribution of null cells, we made predictions of the probability of occurrence of these cells for the whole study area and for the period from August 2003 to December 2015. The environmental database was used in these predictions to ensure predicted values in the entire spatial range of operations by the purse seine fleet. For predictions, we used the mean number of total sets in each cell and a random type of set for the whole study period to detect if certain environmental conditions could be favoring the presence of null cells. Monthly maps for the most representative months of each quarter of the year (February, May, August, and November) were constructed to assess seasonal variation of null cell occurrence. These maps should be interpreted as quarterly means of the probability of occurrence of null cells for the period from August 2003 to December 2015. To highlight the importance of each type of fishing indicator, we also simulated 3 scenarios, as if all fishing sets were made on each of the different fishing indicators (dolphin, unassociated or free-swimming tunas, and floating objects). This resulted in a mean map of null cell occurrence for the whole study period for each type of set. Additionally, to depict the possible effect of ENSO events on the spatial distribution of occurrence of null cells, we created mean maps for 2011 and 2015, when mean ONI values were lowest and highest, respectively.
RESULTS
Fishing effort
During the first 2 quarters of the year, most fishing effort by the purse seine tuna fleet was distributed off Cabo Corrientes, Mexico (~20º N, 115º W), where a patch of >100 purse seine sets was observed. Isolated quadrants with high fishing effort (>300 sets) were observed in the northern part of the Gulf of Tehuantepec (~15º N, 98º W) during the first quarter and off the western coast of the Baja California Peninsula (~24º N, 117º W) during the second quarter (Fig. 1a, b). During the boreal summer (July-September) most fishing effort was concentrated near the Baja California Peninsula (quadrants with >300 sets) and around the Revillagigedo Archipelago (~18º N, 115º W; quadrants with >100 sets) (Fig. 1c). In the last quarter of the year, the most important fishing ground was observed just above the imaginary equator line, across a band of quadrants with 50-100 sets that extended from the coast of Central America to ~98º W (Fig. 1d).
Statistical modeling
Table 1 summarizes the model selection process. Concurvity for environmental variables was low (<0.3) in all cases. All variables produced a decrease of at least 2 AIC units and were statistically significant at 99% confidence level (Table 1). The model explained 21.50% of total deviance. In general, the most important variable was number of sets. The environmental variable with the lowest associated ΔAIC and highest cumulative deviance was Chla. On the other hand, ONI was the environmental variable with the highest associated ΔAIC and lowest cumulative deviance. Null cells were more likely to occur when SST was between 20 and 24 ºC and higher tan 27 ºC. A similar situation was observed when Chla concentration was relatively high (>1 mg·m-3), SLA values were positive, ONI values were close to zero, and fishing was done on unassociated tuna schools. Higher probability of occurrence of null cells was observed for the 2007-2012 period and 3 zones: one in the open ocean, around 100º W and 20º S, and two closer to the coast, one at a lower latitude (0-10º N) and the other in the northern section of the study area (>20º N) (Fig. 2).
Table 1 Summary of the logistic generalized additive model fitted to the presence/absence data of null cells for the purse seine fleet in the eastern Pacific Ocean. Abbreviations are sea surface temperature, SST; chlorophyll a, Chla; sea level anomaly, SLA; Oceanic Niño Index, ONI; number of sets, NumSets; type of set, SetType; longitude, Lon; latitude, Lat; Akaike information criterion, AIC.
| Explained deviance (%) |
Cumulative deviance (%) |
AIC | (AIC | P((2) | |
| Null | 0.00% | 0.00% | 34,653.76 | <0.01 | |
| +s(SST) | 1.37% | 1.37% | 34,193.49 | -460.27 | <0.01 |
| +s(log(Chla)) | 3.29% | 1.92% | 33,541.69 | -651.80 | <0.01 |
| +s(SLA) | 3.40% | 0.11% | 33,512.67 | -29.02 | <0.01 |
| +s(ONI) | 3.42% | 0.02% | 33,505.01 | -7.66 | <0.01 |
| +s(NumSets) | 11.90% | 8.48% | 30,577.53 | -2,927.48 | <0.01 |
| +SetType | 15.10% | 3.20% | 29,450.79 | -1,126.74 | <0.01 |
| +s(Month) | 15.20% | 0.10% | 29428.56 | -22.23 | <0.01 |
| +s(Year) | 15.50% | 0.30% | 29366.95 | -61.61 | <0.01 |
| +s(Lon) | 18.50% | 3.00% | 28321.47 | -1,045.48 | <0.01 |
| +s(Lat) | 20.10% | 1.60% | 27795.53 | -525.94 | <0.01 |
| +s(Lon, Lat) | 21.00% | 0.90% | 27494.49 | -301.04 | <0.01 |
| +s(Lon, Year) | 21.20% | 0.20% | 27464.13 | -30.36 | <0.01 |
| +s(Lat, Year) | 21.50% | 0.30% | 27400.44 | -63.69 | <0.01 |
| Total | 21.50% |
Figure 2 Partial effect plots from the logistic generalized additive model fitted to the fishery and satellite-derived data from August 2003 to December 2015. (a) Sea surface temperature (SST), (b) chlorophyll a (Chla, log-scale), (c) sea level anomaly (SLA), (d) Oceanic Niño Index (ONI), (e) number of sets, (f) month-year interaction, (g) longitude-latitude interaction, and (h) type of set (on dolphins, DEL; unassociated schools, NOA; and floating objects, OBJ). Gray shaded area depicts 95% confidence intervals, and lines perpendicular to the x-axis depict the density of observations for each covariate.
The final model performed well when discriminating absences from presences (AUROC ~ 82.22%). The seasonal prediction maps showed that the probability of null cell occurrence (>0.3) was higher in an oceanic zone centered at ~20º S, 110º W (hereafter oceanic zone), off the coast of Central America (hereafter southern zone), and at the entrance to the Gulf of California (hereafter northern zone). The 3 zones seemed to be present throughout the year, with slight seasonal variations (Fig. 3). The oceanic zone in the southern portion of the study area (~20º S, 110º W) spatially agrees with a zone where fishing effort was low (cells with <10 sets, Fig. 1) during 2003-2015, so this zone was not considered an important tuna fishing area; therefore, the rest of the análisis focused on the other 2 zones for which the model predicted high probability of null cell occurrence (the northern and southern zones).
Figure 3 Mean quarterly spatial prediction of the probability of occurrence of null cells for the period from August 2003 to December 2015. Predictions are made for the first (January-March) (a), second (April-June), third (July-September), and fourth (October-December) quarters of the year. Color bar depicts the probability of occurrence of null cells. Contours are printed in 0.1 increments.
Low probability of null cell occurrence (<0.1) was observed when the fleet fished using dolphins as fishing indicators (Fig. 4a). Fishing on unassociated tuna schools was the most likely cause for the 2 zones with high probability of null cell occurrence because our model predicted probabilities >60% under this scenario (Fig. 4b). Probability of null cell occurrence was relatively high (~0.2-0.3) when simulating that the fleet fished on floating objects (Fig. 4c).
Figure 4 Mean maps for the prediction of null cell occurrence in the period from August 2003 to December 2015 under different simulation scenarios. Simulations were made assuming all the purse seine sets were made on dolphins (a), unassociated (free-swimming) tuna schools (b), and floating objects (c). Color bar depicts the probability of occurrence of null cells. Contours are printed in 0.1 increments.
The lowest mean ONI values occurred during 2011 (yearly mean = -0.72, SD = 0.30) and the highest occurred during 2015 (yearly mean = 1.25, SD = 0.63). The zone with high probability of null cell occurrence near the Central American coast was broader during 2015, when probabilities of null cell occurrence were >0.6. The zone with high probability of null cell occurrence at the entrance to the Gulf of California remained practically unchanged when comparing the spatial distribution of the probability of null cell occurrence during La Niña (2011) and El Niño (2015) conditions (Fig. 5).
Figure 5 Mean yearly predictions of null cell probability of occurrence for those years with extreme Oceanic Niño Index (ONI) values in the period from August 2003 to December 2015: (a) 2011 (lowest values) and (b) 2015 (highest values). Color bar depicts the probability of occurrence of null cells. Contours are printed in 0.1 increments.
DISCUSSION
Spatiotemporal distribution of fishing effort
During the first half of the year, increased fishing effort was observed off Cabo Corrientes (~15º N, 110º W), the entrance to the Gulf of California, Mexico. In this region, intense wind-driven upwelling events occur during the second quarter of the year, with peak primary productivity during the late boreal spring and early summer (López-Sandoval et al. 2009). Moreover, this region is also affected by coastal trapped internal waves, which are capable of generating both cyclonic and anticyclonic eddies when they alter the local poleward currents as they interact with the shape of the coastline (Zamudio et al. 2007). Our results agree with those of Torres-Orozco et al (2005), who reported that the number of purse seine sets carried out off Cabo Corrientes by the Mexican tuna fleet increased during March and June. Torres-Orozco et al. (2005) mentioned that this region was an interesting yellowfin tuna fishing ground because water filaments detach from the coastline into the open ocean, as a result of winddriven upwelling events.
During the second half of the year, fishing effort was mostly observed during the boreal summer (July- September) near the Baja California Peninsula, off Mexico (~25º N, 118º W). The high biological productivity that takes place in this area makes the Baja California Peninsula an important fishing ground for the purse seine fleet. Here wind-driven upwelling events last up to 10 days and are more common during May-June (Cervantes-Duarte et al. 1993, Zaytzev et al. 2003). These upwelling events promote large aggregations of pelagic red crab (Pleuroncodes planipes), one of the favorite prey of the pelagic fish community in the region (Aurioles-Gamboa et al. 1994), including yellowfin tuna (Bocanegra-Castillo 2007). Ortega-García and Lluch-Cota (1996) found a 3-month lag between peak upwelling intensity and peak yellowfin catch rates in Pacific waters off southern Mexico and attributed this lag to the time required for predators higher in the food chain to arrive at an upwelling area. Therefore, during July- September, 2 to 3 months after the peak upwelling activity in the region, yellowfin, the main target species of the purse seine fleet that operates in the ETPO (Maunder and Watters 2003), arrives in the area near the Baja California Peninsula to feed on pelagic red crab, explaining the increased fishing effort observed during the boreal summer.
Modeling and spatial prediction of the probability of occurrence of null cells
In general, operational variables were more important in explaining the probability of null cell occurrence, compared with environmental variables. For example, the variable that accounted for most of the explained deviance was number of sets, with higher probability of null cell occurrence at low numbers of purse seine sets. This suggests that the probability of a null cell occurring decreased when Fisher insistency increased. Fishing indicator (type of set) was the variable that followed in importance. Increased probability of null cell occurrence was observed when purse seiners fished on unassociated (free-swimming) tuna schools. This result is consistent with previous reports for the Indian, Atlantic, and Pacific oceans (Fonteneau et al. 2000, Delgado De Molina et al. 2010, Floch et al. 2012), and it appears to be attributable to the fact that medium-sized tunas (10-15 kg), the most common size in free-swimming schools, are more evasive and therefore more difficult to catch in purse seine maneuvers (Fonteneau et al. 2000, Hallier and Gaertner 2008).
Of the environmental variables, surface Chla concentration was the most important covariate, and higher probabilities of null cell occurrence were observed at relatively high Chla values. One plausible explanation for this is that small- and medium-sized tunas, which occupy lower positions in the food web than their larger relatives, inhabit áreas with higher Chla concentrations, feeding on small prey that are closer to the lower trophic levels occupied by phytoplankton containing Chla (Ménard et al. 2006). Moreover, Brill and Lutcavage (2001) found that small juvenile bluefin tuna prefer zones with relatively high Chla concentrations. Since medium-sized tunas are more elusive than larger tunas (Fonteneau et al. 2000), the probability of null cell occurrence is expected to be higher in Chla-rich zones, where small- and medium-sized tunas are likely to inhabit.
Our final model predicted high probabilities of null cell occurrence in 2 zones that are environmentally different. Ocean dynamics in the southern zone are dominated by the North Equatorial Current, the Costa Rica Dome, and the Papagayo and Panamá wind jets. Mean SST values in this zone are >26 ºC, with very little seasonal variations (±2 ºC, Fiedler and Talley 2006). The thermocline in this zone is mostly shallow (<50 m) all year long (Fiedler 2002). The northern zone, on the other hand, lies in the transition zone where the cold, nutrient-rich waters from the California Current meet the warm and oligotrophic water masses of equatorial origin (Kessler 2006). Sharp seasonal variations in SST occur in this zone, with values of 20-24 ºC during the first quarter and >28 ºC during the third quarter. The thermocline in this zone is usually found at shallow depths (<60 m), although seasonal variations in thermocline Depth are more evident here than in the southern zone, and it is shallower during the second quarter of the year (Fiedler and Talley 2006).
Dreyfus-León et al. (2015) suggested that the main cause of zero-catch sets in the northern zone is the seasonal deepening of the thermocline, because tunas are able to escape through the bottom of the purse seine. Dreyfus-León et al. (2015) used a basic statistics technique (linear correlation) to highlight the relationship between null sets and environmental variables (the Multivariate El-Niño Index). We believe that our results are more robust because (1) the spatiotemporal coverage of our data is broader, (2) we used additional environmental variables (SST, Chla, SLA, ONI) as predictors, and (3) we applied an improved statistical technique to analyze nonlinear relationships between predictors and the logit of null cell occurrence. Environmental variables, including the variable that causes the deepening of the thermocline (ONI) and the one that can be used as a proxy for thermocline Depth (SLA), explained only a small percentage of total deviance, as opposed to operational variables. Moreover, most of the purse seine sets made on unassociated tuna schools were carried out in the 2 zones (near the Baja California Peninsula and near the Central American coast) for which our model predicted higher probabilities of null cell occurrence (Fig. S1). Therefore, operational factors seem to be more important for the occurrence of null cells than variations in the marine environment.
Economic and ecological impact of predicting areas with high probability of null cell occurrence
The final GAM quantifies the probability of occurrence of null cells for the tuna purse seine fishery operating in the EPO. The model could be used as a decision-making tool by crews in purse seine tuna vessels that operate in the EPO. For example, of the Ecuadorian purse seine fleet (Bucaram 2017), if we presume that a purse seiner performs half (13) of the yearly mean sets (26) in a fishing trip to one of the zones where the probability of null cells is ~30%, then 13 × 0.3 ~ 4 of those cells would be a zero-catch set. On the other hand, if the crew of this purse seiner decides to carry out these 13 sets in a zone where the probability of occurrence of null cells is ~10% (or to fish using a fishing indicator other than unassociated tuna schools), then only around 13 × 0.1 ~ 1.3 of the 13 sets would be unproductive, thus saving 4,120 × 3 ~ 12,360 USD (where 3 is the difference of unproductive sets in both theoretical scenarios) of the mean cost of the fishing trip.
Minimizing the fishing effort by the purse seine fleet could also have interesting ecological consequences. For example, for purse seiners with storage capacity >452 m3, Marín-Enríquez et al. (2018) reported a mean bycatch of ~55 dolphinfish (Coryphaena spp.) per purse seine set in the EPO during 2004-2013, and Cruz-Cosío (2018) estimated a mean bycatch of up to 0.7 individuals of the critically endangered oceanic whitetip shark (Carcharhinus longimanus) per set during 1993-2005. For the same 2 theoretical scenarios described above, the difference of 3 unproductive sets would imply a bycatch decrease of 55 × 3 × 6 ~ 990 dolphinfish per year per vessel and of 0.7 × 3 × 6 ~ 13 individuals of oceanic whitetip shark per year per vessel (3 sets per trip, 6 trips per year). Of course, these are mean rough estimates, and even though they are based on assumptions that could not be entirely true, they could give the reader an insight into the advantage of decreasing the fishing effort by the purse seine fleet in the EPO, from the marine conservation perspective.
