Effect of environmental conditions on the seasonal and inter-annual variability of small pelagic fish abundance off North-West Africa: The case of both Senegalese sardinella

The objective of this study was to assess the effect of environmental variations on the abundance of Sardinella aurita and Sardinella maderensis in Senegalese waters in the upwelling system. Monthly data indicating the abundance of sardinella were first estimated from commercial statistics, using Generalized Linear Model from 1966 to 2011. Abundance indices (AIs) were then compared with environmental indices, at the local scale, a Coastal Upwelling Index (CUI) and a coastal Sea Surface Temperature (SST) index, and on a large scale, the North Atlantic Oscillation (NAO), the Atlantic Multidecadal Oscillation (AMO) and the Multivariate El Ni ~ no Southern Oscillation Index (MEI), using correlations and times series analyses. The results showed that the abundance of sardinella is determined by a strong seasonal pattern and inter-annual fluctuations. The abundance of S. aurita peaked in spring and in autumn, whereas that of S. maderensis peaked in the warm season (July – September). The trend of the sardinella abundance was significantly correlated with the CUI, especially in autumn and spring. Interannual fluctuations of S. maderensis and S. aurita abundance are, respectively, driven by the precocity and the duration of the upwelling season that is attributed to distinct migration patterns. Both sardinella species also respond with a delay of around 4 years to the winter NAO index and the autumn CUI, and the AMO index, respectively, both related to migration patterns. The wide variations in sardinella biomass are caused by variations in environmental conditions, which should be considered in the implementation of an ecosystem-based approach in sardinella stocks management.


| INTRODUCTION
The North-West (NW) African coast, being one of the four major eastern boundary upwelling systems (EBUS), is characterized by an important biological productivity owing to its upwelled nutrients, which help to sustain large fish populations. Small pelagic fish constitute a resource of primary importance, which is heavily exploited in Senegal, Mauritania and Morocco. In Senegal, small pelagic fish mainly include, by order of importance in the average landings: sardinella (Sardinella aurita or round sardinella, and Sardinella maderensis or flat sardinella), mackerel (Scomber scombrus), anchovy (Engraulis encrasicolus), ethmalosa (Ethmalosa fimbriata), horse mackerel (Trachurus spp.) and sardines (Sardina pilchardus); they are all aggregative species occurring in a fish school (Brehmer et al., 2007). The sardinella accounts for more than 70% of the total landings that is mainly provided by the small-scale fishery (CRODT, 2009). On average, the yearly landings of sardinella, within e.g., the years [2002][2003][2004][2005][2006], were estimated at 284,000 tons/year (FAO, 2007).
The size of the sardinella populations is highly variable within and over the years, in connection with the environmental control on growth and recruitment processes (for example, Cury et al., 2000). These variations might be accentuated by the constant increase in fishing pressure (Sharp & Csirke, 1983). The sardinella stocks have been considered to be overexploited since 2006 (FAO, 2006) by small-scale fisheries, foreign industrial fishing vessels which operate within the framework of the bilateral agreements and the illegal non-reported fisheries (Belhabib, Koutob, Sall, Lam, & Pauly, 2014).
Sardinella species possess a relatively short life span of approximately 7 years (Chesheva, 1998(Chesheva, , 2006) and a high natural mortality rate (Camarena-Luhrs, 1986;Fr eon, 1988). The S. aurita reproduction is continuous with two annual reproductive peaks, the first one in May-June and the second between September and November (Fr eon, 1988;Ndiaye, 2013). Because of the spatial variability and strong seasonality of the coastal upwelling, S. aurita spawning is not uniform along the coast but occurs preferentially over the Arguin Bank (Mauritania), the south of Cap-Vert (Senegal) (Bo€ ely, Chabanne, & Fr eon, 1978;Bo€ ely, Chabanne, Fr eon, & St equert, 1982;Conand, 1977) and, to a lesser extent, in between the two regions (Fr eon, 1988). For S. maderensis, two spawning periods of S. maderensis were determined (Ba et al., 2016). The first spawning period was from April to October, with two spawning peaks, one in June-August and the other in October. The second spawning period occurred from early January to the end of February, with a single intense spawning peak. Sardinella maderensis almost spawned all year round, with spawning peaks at these periods.
Sardinella are widely spread along the NW African coasts. Bo€ ely (1980) described the large migrations of S. aurita between the Moroccan and the south of Senegal waters whereas, on the other hand, S. maderensis is believed to perform lesser migrations around their nursery area, suggesting the existence of independent stocks at the Arguin Bank, and at the southern coast of Senegal (Bo€ ely & Fr eon, 1979; Garcia, Tandstad, & Caramelo, 2012). These trophic and ontogenetic migrations are related to seasonal environmental variations, which have a stronger amplitude in the southern part of the eastern edge upwelling system of the Canary Current (Bo€ ely & Fr eon, 1979). The transboundary migrations complicate the fisheries management policies, as the S. maderensis and S. aurita landings reports are usually mixed, and the migration routes as well as the existence of sub-populations for each species remain poorly understood.
According to Bo€ ely et al. (1982), the adults of S. aurita, migrate from Senegal to the South of Morocco from April to September.
Starting at the end of September, S. aurita return to Senegal, thereby passing again through Mauritanian waters. Bo€ ely et al. (1982) further indicate areas of recruitment in Mauritania and Senegal, which are affected by the migration cycle after the respective stocks have spent their first year in the local nursery where the first spawning occurred. The seasonal primary production is likely to be the driver of such long-range migration (Bo€ ely et al., 1982;Cury & Roy, 1988).
The recent work by Bacha, Jeyid et al. (2016), Bacha, Jehid, Vantrepotte, Dessailly, and Amara (2016) clearly indicated that the Mediterranean Sea and Atlantic Ocean S. aurita populations are distinct. Unlike the genetic study conducted off NW Africa (Chikhi, Agn ese, & Bonhomme, 1997), the otolith shape of S. aurita suggests the existence of isolated groups of fish. The separation of these groups seems to be linked to oceanographic barriers and local retention areas. There is no evidence of such barriers between Senegal and Mauritanian waters. Thus, the variability of S. aurita abundance in Senegal in relation to local environmental fluctuations must be linked not only to population dynamics but also to migration processes.
In NW Africa, the influence of the environment on fisheries resources has been increasingly studied in recent decades (Cury & Roy, 1989;Faure, Inejih, Demarcq, & Cury, 2000;Caballero-Alfonso et al., 2010;Braham, Fr eon, Laurec, Demarcq, & Bez, 2014;Mbaye et al., 2015;). The changes in hydro-climatic conditions are likely to be the major cause of the strong variability of coastal pelagic resources, particularly for sardinella. Off NW Africa, the coastal upwelling intensity during the upwelling season in winter and spring is an important driver of primary production (Auger, Gorgues, Machu, Aumont, & Brehmer, 2016;Lathuili ere, Echevin, & L evy, 2008), which may affect the survival of early life stages of Sardinella spp. (for example, Bartolino, Colloca, Sartor, & Ardizonne, 2008;Faure et al., 2000;Gonz alez Herraiz, Torres, Farina, Freire, & Cancelo, 2009;Gr€ oger, Winkler, & Rountree, 2007). The mechanisms that determine the spatial and temporal dynamics of sardinella must be elucidated for the sustainable management of these resources. An analysis of the relationship between climate variations and living resources should contribute to improve our knowledge of sardinella population and its spatial dynamics, allowing for the establishment of a rational management plan for these fisheries.
This study aims at analyzing the relationship between the environment and sardinella abundances in Senegalese waters. The objectives of the study were (i) to test the empirical predictive models of fish abundance and (ii) to analyze the abundance-environment relationships at various time scales. Monthly abundance index (AI) of sardinella was estimated from commercial statistics over the period 1966-2011 using Generalized Linear Models (GLM) techniques.
Hereafter we use the term «abundance» to refer to this abundance index. Thereafter, the seasonal variability was isolated and the interannual variability of the trend of abundance was correlated against five environmental indices: (i) the Coastal Upwelling Intensity (CUI) i.e., the Ekman transport from a re-analysis of alongshore wind stress,

Coastal Upwelling Index (CUI)
We used outputs of wind stress from an atmospheric re-analysis from the National Center for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR, 2.5°resolution) over the period 1948-present to derive a monthly time series of CUI for the Senegalese coast. This low-resolution product is the only available product which covers the study period 1966-2011. As a first approximation, we considered the upwelling index owing to Ekman transport, omitting Ekman pumping due to wind-stress curl. Ekman transport is proportional to the wind stress, and inversely proportional to the sine of the latitude (Sverdrup et al., 1942). The volume of Ekman transport per meter of coastline (m 3 /s per meter of coast) was estimated using the following equation: where s is the along shore component of wind stress within 270 km of the coastline (positive southward), is the seawater density (1,025 kg/m) and f is the Coriolis parameter (=2 Ω sin(h), with Ω and h equal to the Earth's angular velocity and latitude, respectively).

Coastal Sea Surface Temperature (SST)
As an alternative index of coastal upwelling, a monthly index of coastal SST was derived from the AVHRR Pathfinder product , NOAA, 4 km of resolution) for the Senegalese coast by averaging SST from the coast to the isobath 200 m where upwelling-induced SST anomalies are generally maximum (Demarcq & Citeau, 1995). Alongshore winds generate coastal upwelling through offshore Ekman transport, which brings cold water from below the seasonal thermocline (50-200 m) to the surface, especially near the coast where the upwelling flow is at its maximum.
Because of the temperature difference between the coastal and offshore water masses, the surface temperature is a good descriptor of coastal upwelling. gov/_kew/MEI/)) to monitor the status of ENSO during the autumn season (October-December), before the winter-spring upwelling period (January-June) in the eastern tropical Atlantic. The MEI is a composite index of the six main observed variables over the tropical Pacific. As the MEI integrates more information than other indices, such as the SOI index based on Tahiti-Darwin pressure difference, it is thought to reflect the nature of the coupled oceanatmosphere system better (Wolter, 1987;Wolter & Timlin, 1993 The CPUE can be modeled as the result of an AI multiplied by the fishing power of the individual vessels, combined with a residual variability (Robson, 1966). This is usually turned into an additive relationship by logarithmic transformation, which makes it possible to use linear regression techniques to do an estimation of the unknown parameters.

Large-scale Atlantic climate indices
Linear models and their extension into GLM (McCullagh & Nelder, 1989) are nowadays widely used. They make it possible to consider several explanatory variables either categorically or continuously, but more interestingly, is their interactions. Monthly CPUEs were estimated using a GLM from the commercial industrial fishery statistics.
Based on the detailed data obtained from Senegalese industrial trawlers, mean CPUE, expressed in kg of sardinella per fishing day, were calculated per boat, fishing area, year and month. Based on their engine power, the boats were placed in categories that are referred to as "engine power-classes". The CPUE is the response variable of the model and its dependent factors are: the year, the month, the fishing area and the engine power-class of the fishing boats. Sardinella can be off-loaded in almost every fishing trip, and the occurrence of this species in the catches is therefore close to 1.
Hence, a Gaussian model is used to estimate monthly abundance.
The model is expressed as follows: ln CPUE y;m;z;i ¼ lnA y;z þ lnd m þ lnP f þ lne y;m;z;f where CPUE y,m,z,i is the catch per unit of effort of the year (y), the month (m), the area (z) and boat (i) (belonging to the engine powerclass (f), A y,z is the combined statistical effect of year and area, d m is the monthly effect, P f is the engine power-class effect, and e y,m,z,f is the normally distributed residual. It is noteworthy that A y,z can be interpreted as a monthly abundance index in the fishing area under the assumption of a constant year-to-year seasonality of CPUE, and a constant fishing efficiency per engine power-class. Conversely, such an index is not biased by changes in the spatial fishing patterns or by an increase in the engines' power. However, changes in the seasonal pattern cannot be investigated using abundance index from this method.
GLM models were fitted using a negative log-like loss function.

| Exploratory abundance data analysis
A boxplot analysis was carried out to describe the seasonal cycle of sardinella abundance estimated by GLM off Senegal. Monthly abundances were averaged over each season to obtain a time series of abundance per season.

| Time series analysis
The monthly time-series of sardinella abundance were analyzed to determine the main factors affecting their variability. First, a seasonal decomposition was conducted on AIs and environmental indices: CUI, SST, NAO, AMO and MEI. For example, as sardinella abundance (R t ) shows a strong seasonal cycle, the signal was decomposed into a combined trend and cycle component (p t ), a seasonal component (s t ) and a residue component (u t ) (Cleveland, Cleveland, McRae, & Terpenning, 1990) using the package STL in R software.
There are sophisticated procedures to carry out seasonal decomposition nevertheless in this study, we used the method of Census II (Makridakis, Wheelwright, & McGee, 1983) for its flexibility and simplicity. The seasonal decomposition is with a homogeneous variance, with or without long-term trend, and is applied on log-transformed and differentiated data (Gonz alez Herraiz et al., 2009).  Table 1). These are used T A B L E 1 Pearson's correlations between "seasonal time series" of the environmental indices (explanatory indices) and time series of the annually averaged trend components of (a) Sardinella aurita and (b) Sardinella maderensis abundance indices (AI) taking into account a lag of 0-10 years  For S. maderensis, the fitting GLM model explained 23.8% of the total deviance (Table 2b). The same factors are retained as before.

| Relation abundance versus environment
However, this species was equally distributed in the North (Grande Côte) and the Petite Côte. While this is in contrast with artisanal catches, the S. maderensis abundance was generally lower than that of S. aurita (Figure 4), which seems to be reflected in the catches.     The evolution of the abundances of both species of sardinella off Senegal was close, suggesting common drivers. However, both sardinella species presented different answers with regard to local environmental conditions and climate variability in the North Atlantic (see Table 1).

| Long-term variations for S. aurita
The interannual variability of the S. aurita abundance (annual mean) correlates positively with the CUI in spring at lag 0-2 (in years, p < .01, Table 1a) and also, although poorly (p > .01), with the autumn CUI at lag 1. However, the negative correlation found with the autumn The use of non-linear GAM models in place of linear regressions taking each environmental index separately as an explanatory variable did not demonstrate any additional significant relationship that our linear analysis could have missed (see Table 1a) with the S. aurita abundance (not shown). The GAM model was run considering the annual time-series of the environmental indices to identify the most significant environmental drivers of S. aurita abundance among those previously identified (see above). The results showed a significant correlation with the mean annual abundance (with lag, see Table 1). This model suggested that the S. aurita abundance is mainly explained by the spring CUI (lag 0) and the annual NAO index (winter-spring, lag 7), and, less, by the fall in CUI (lag 5) (Table 3a and Figure 9). The relationships with S. aurita abundance were mostly linear except that the effect of the NAO is overall evident during positive phases ( Figure S1).

| Long-term variations for S. maderensis
The interannual variability of S. maderensis abundance (annual mean) correlated positively with the CUI in spring at lag 0 (Table 1b) implying a direct effect of upwelling-favorable wind anomalies at the end of the upwelling season. A positive correlation, this time significant, was also found with the autumn CUI (Table 1b) at lag 1-3 which confirms that the duration of the upwelling season is important to explain interannual fluctuations of sardinella abundance, especially the precocity of the upwelling season for S. maderensis. As for S. aurita, the latter correlation was surprisingly accompanied by a negative correlation with the autumn NAO index at lag 1. In contrast, the annual abundance of S. maderensis was positively correlated, although poorly (p > .01), with the AMO index at lag 0-1 (Table 1b) in agreement with the analysis of fish catches by Alheit et al. (2014).
As for S. aurita, no additional environmental explanatory index of S. maderensis abundance emerged from the use of non-linear GAM models. However, a GAM modeling approach similar to that used for S. aurita (see Section 3.6.2) showed that the S. maderensis abundance was mainly explained by the annual NAO index (lag 1) and, less, by the effect of the spring CUI (lag 0) (Table 3b and Figure 10).
It is noteworthy that the relationship with the annual NAO index was non-linear as only weak negative anomalies of S. maderensis abundance were observed during positive phases of the NAO, with respect to negative phases ( Figure S2).  1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Abundance indices (Kg per fishing day) Year (  1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Abundance

| DISCUSSION
Results from this study will help to identify the relationships between the variability of sardinella abundance and the environmental conditions. We base our time series analysis on a 46-year data set of monthly observations of the S. aurita and S. maderensis fisheries off the Senegalese waters.
One limit of our methodology is that the small-scale fishing effort metric recorded in Senegal, expressed in the number of fishing trip, has decreased over the past decade despite an increase in catches. Indeed, the number of fishing trips did not translate well in the effective changes in fishing effort because the increase in distance traveled and sea fishing time were not taken into account.  1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Mean CUI in spring  1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Mean CUI in autumn  1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Yearly NAO

| Seasonal variability of sardinella
Monthly time series of S. aurita abundance showed a bimodal seasonal cycle with two peaks in spring and autumn, and an annual minimum in summer. The abundance index of S. aurita was maximal during the transitional phases of SST that is, from cold to warm (in spring, the highest abundance) and warm to cold (autumn) seasons ( Figure 7). The low abundance of S. aurita was observed when the SST remained stable in cold or warm season (Figure 7).
We propose two hypotheses to explain this bi-modal cycle of the abundance index. First, during those two seasons, the harvest rate is high owing to favorable climatic conditions which facilitate small-scale fishing operations. They also correspond to periods of migration of sardinella (southward, or "descent" and Northward or "ascent") during which fishermen easily detect fish shoals (com. pers. Moulaye MBAYE, "Comit e Local de Pêche Artisanale" CLPA Saint-Louis). Second, the spring might be the period of maximum sardinella accumulation in Senegal because, during this period, fish coming from Mauritania may end their southward migration and fish having previously migrated further south may begin their Northward migration (Bo€ ely et al., 1982). Conversely, the peak of abundance during the autumn might be due to the convergence of S. aurita, which continue to migrate from South to North (where the upwelling intensity was high last winter) with those in Mauritania that are beginning their Southward migration. Similar bimodal patterns were also found for other fish species such as octopus, sardine, shrimps and the whiter grouper, or "thiof" in the local name (Epinephelus aeneus) in NW Africa (Cury & Roy, 1988;Bacha, Jeyid et al., 2016;Bacha, Jehid et al., 2016). These authors suggested that the regional upwelling dynamics and local environmental effects might influence the migration of these species, i.e., the simultaneous relaxation and trigger of the coastal upwelling off Mauritania and Senegal, respectively.
The Senegalese small-scale fishery directed to sardinella is mainly carried out during the transition periods (May to June and November to December; Fr eon, 1991), which corresponds to the period of highest abundance of S. aurita. The increase in harvests during the cold season was then concomitant with a size increase of the population ( Figure 8). Indeed, the period of high catch of S. aurita corresponds to the period during which we observed the larger size of adult mode on the Petite Côte for both spring and autumn and for the Grande Côte in spring. As fecundity is proportional to the weight of the mature individuals, this is in line with the fact that there are two annual cohorts in the S. aurita populations of the Senegalese waters. These two cohorts are the result of two main annual reproductive peaks (Fr eon, 1988;Ndiaye, 2013).
For S. maderensis, the abundance was higher in summer (Figure 6), between June and September, than during the rest of the year. The juveniles were observed in autumn at the Grande Côte ( Figure 8). In contrast, they were present throughout the year (but especially plentiful in autumn) in the Petite Côte. It is known that this species tolerates strong environmental variations (Cury & Fontana, 1988), which would explain why S. maderensis is also found in the cold Mauritanian coastal waters (Bo€ ely & Fr eon, 1979;Corten & Sadegh, 2014) and even in the colder waters of the Sahara Bank (Ettahiri, Berraho, Vidy, Ramdani, & Do chi, 2003).
The highest catch of S. maderensis was observed in summer along the Senegalese coast ( Figure 6). At this season, the size mode of the S. maderensis population was highest than the S. aurita's on the Grande Côte (22.5 cm FL versus 20 cm). Both these differences in size and the reduction of S. aurita abundance owing to its Northward migration in summer (Bo€ ely, 1980) cause the fishermen to change their strategy by switching from the purse seine to the encircling gillnet to target S. maderensis (see section 3.2.).
In Senegal, the abundances indices for both sardinella species were higher in the Petite Côte than in the Grande Côte. The larger sizes of sardinella were encountered in the Northern area. In the Southern area, the sardinella encountered were generally small to medium in size. The juveniles of sardinella are always present in the Petite Côte and seasonally observed in the Grande Côte. The Petite Côte is the main spawning and nursery area for both sardinella species. This coastal area is characterized by a broad and shallow continental shelf, which leaves an inshore well-mixed zone acting as a retention area for fish larvae .
This spatial distribution is confirmed by the observation that the Southern area is a nursery area, and the especially large individuals migrate North in deeper areas (Krakstad, Sarr, Sow, Mbye, & Sk alevik, 2013).
The sardinella migration appears to be primarily driven by foraging needs and spawning preferences (Zeeberg, Corten, Tjoe-Awie, Coca, & Hamady, 2008). Sardinella adults move alongside, preferentially looking for convergence areas nearby upwelling of cold waters (Bo€ ely et al., 1982). In the cold season, most of S. aurita are found in Senegalese waters where the temperature remains above 21°C. The productivity of the Senegalese waters is high during winter and spring, as a result, according to Zeeberg et al. (2008), of river run-off after the rainy season, localized upwelling, and cyclonic eddies retaining productive waters.

| Environmental effect on sardinella abundance
The West Africa (Braham et al., 2014;Diankha et al., 2015;Mbaye et al., 2015). This pattern is also in accordance with the dynamics exhibited by other important resources in the area (for example, Farfantepenaeus notialis and Octopus vulgaris), which have similar periodicity (Thiaw, Gascuel, Thiao, Thiaw, & Jouffre, 2011;Thiaw et al., 2009). Changes in the recruitment from year to year that are due to fluctuations in environmental conditions are thought to especially affect the early life stages of several species (Bakun & Csirke, 1998;Caballero-Alfonso et al., 2010;Dawe & Warren, 1993). The latter studies suggest that food availability enhanced by coastal upwelling may be the primary controlling factor for larval sardinella survival and recruitment, which we find to translate into fish abundance.
In the NW African upwelling system, the variability of upwellingfavorable winds is positively related to the NAO (Meiners et al., 2010). The relationship between winds and small pelagic abundance depends on how wind variability affects the recruitment success, i.e., nutrient enrichment, concentration of larval food and retention of larvae, according to the "ocean triad hypothesis" (Bakun, 1996). In the Iberian upwelling, high dispersal of sardine eggs and larvae by enhanced coastal upwelling during the winter spawning season was actually proposed to explain a negative relationship between the NAO index and sardine abundance (Guisande et al., 2001;Borges et al., 2003;Santos et al., 2007). In contrast, the recruitment success and abundance of sardine are positively correlated with wind anomalies off Morocco (Roy et al., 1992;Kifani et al., 2008;Machu et al., 2009), and we found a similar relation with sardinella abundance in Senegalese waters. This suggests that the negative effect of wind speed on food concentration and larval retention does not drive the recruitment success of sardinella off Senegal. Moderate winds and a large continental shelf may explain why the latter processes are not so affected by wind variability.
An alternative or complementary explanation for the inter-annual changes in abundance may be that the sardinella abundance in Senegalese waters depends more on the inflow and outflow of sardinella due to migrations than on the local recruitment success. The S. aurita abundance in Senegalese waters seems mostly driven by the duration of the upwelling season (spring CUI at lag 0), whereas the S. maderensis abundance also depends on the precocity of the upwelling season (autumn CUI at lag 1) and the state of the NAO during the previous year (autumn and annual NAO at lag 1, Note that the apparent discrepancy of the correlations (see Table 1 | 597 Senegal, also correlates significantly though negatively with the NAO index and the intensity of upwelling-favorable winds, with a comparable time lag of 3 years (Meiners et al., 2010). For S. maderensis the abundance responds positively (but with poor significance, p > .1) to the AMO index and negatively to the winter coastal SST index, in line with enhanced coastal upwelling, with a similar delay. This suggests a primary role of food availability on S. aurita abundance, but in this case with a delay maybe depending on sardinella life span and migrations (Chesheva, 1998(Chesheva, , 2006, and on the size/age structure of the catches (Meiners et al., 2010). In contrast, S. maderensis is better related to SST according to a greater tolerance to strong environmental variations (Cury & Fontana, 1988), and may benefit from unfavorable conditions affecting S. aurita.
Wind observations during the last decades (Sydeman et al., 2014) and projections from coupled ocean-atmosphere general circulation models (IPCC AR5, Stocker et al., 2013) overall agree that global warming will widen the tropical band and shift the subtropical highs poleward in the future. Although large regional and seasonal uncertainties remain, this would tend to increase ( Indeed, the analysis of the landings by site show that most of the increase in sardinella landings in Senegal over the last decades was recorded in the northern and southernmost well connected commercial landing sites limits in Saint-Louis and Joal, respectively. We suggest that the increase in landings in these two sites was due to the increasing distance of the fishing trips toward the adjacent northern and Southern waters that are in Mauritania in the north, and in Gambia, Casamance and Guinea Bissau shared waters in the South. The precocity and duration of the coastal upwelling period off Senegal respectively influence interannual fluctuations of S. maderensis and S. aurita abundances, which is attributed to distinct migration patterns. Sardinella maderensis may mostly depend on the coastal upwelling intensity in autumn when the migrant S. aurita would still be absent to compete for food. In contrast, the migrant S. aurita would benefit from an anomaly of coastal upwelling intensity in spring when their population is full established off Senegal. Winds are a key driver of sardinella abundance and depends on the state of the NAO (significant relationship is found with the annual NAO). The relation is at low frequency so the decadal variability of sardinella abundance could be inferred from the low frequency variability in the NAO signal. Additionally, the winter/spring NAO could be related to the S. aurita abundance with predictive skills at around 4-7 years.
Integration of coastal primary productivity and migration patterns may drive such predictability. However, dedicated modeling approaches should be carried out to elucidate the main processes involved. Indeed, such information could be useful to further optimize the management of fish stocks and predict their response to climate change under various exploitation scenarios (Bartolino et al., 2014).
Integrating the effect of environmental variability on fish stocks dynamics is central to the ecosystem approach for fisheries management in North West Africa small pelagic fisheries as well as to estimate loss and damage in this fisheries sector due to climate change.