Contrasted optimal environmental windows for both sardinella species in Senegalese waters

We investigate Sardinella aurita and Sardinella maderensis recruitment success relative to the variability of oceanographic conditions in Senegalese waters using generalized additive models (GAM). Results show that recruitment of both species is marked by a strong intra-annual (seasonal) variation with minimum and maximum in winter and summer, respectively. Their interannual variations are synchronous until 2006 (recruitment decreasing), while from 2007 there is no synchrony. The model developed shows that sardinella recruitment variability is closely related to the tested environmental variables in the study area. However, the key environmental variables influencing the recruitment success are different for both species: the Coastal Upwelling Index and the sea surface temperature for S. aurita and S. maderensis , respectively. We report that recruitment success of S. aurita and S. maderensis are associated with distinct ranges of sea surface temperature, upwelling intensity, wind-induced turbulence, concentration of chlorophyll- a and north Atlantic oscillation index. Considering food security and socio-economic importance of both stocks, we recommend that consideration is given to the environmental variability in the small pelagic fish national management plans, particularly in the context of climate change.


| INTRODUCTION
Hydroclimatic conditions along Senegalese coasts are particularly favorable to many fishes, due to enrichment by one of the most productive upwelling systems (Auger, Gorgues, Machu, Aumont, & Brehmer, 2016;Binet, 1988;Roy, Cury, Fontana, & Belv eze, 1989). Indeed, their abundance is thought to be related to the enrichment of the euphotic zone which increases primary production (Bakun, 1997;Cury & Roy, 1987;Demarcq & Samb, 1991). This eastern boundary upwelling system is characterized by a strong seasonal and interannual variability, well known to impact the fluctuations of abundance and distribution of small pelagic fishes (Bakun, 1996;Teisson, 1982).
To date, it is well established that large marine ecosystems (LME) have been continuously warming during the 1982-2006 period, and to a lesser extent the California and Humboldt LME (Belkin, 2009), also including eastern boundary upwelling systems. In California, thermal trends are extremely variable within sub-regions, as noted by Mendelssohn and Schwing (2002) between 1945and 1990 warming is evident across the 1950-2000 period in this system (Sydeman et al., 2014). Based on in situ data, surface temperature in the Canary Current large marine ecosystem (CCLME) increased by 0.52°C from 1982 to 2006 (Belkin, 2009) and between 0.50 and 0.75°C from 1950 to 2000 (Harrison & Carson, 2007). Remote sensing observations also showed that this warming is spatially heterogeneous, particularly in the Southern part of the CCLME (Demarcq, 2009) Senegalese region. A depletion of phytoplankton biomass is usually associated with this warming (i.e., Richardson & Schoeman, 2004), even if moderate and spatially restricted to Mauritania and Senegal from 1998 to 2014 (Demarcq & Benazzouz, 2015). Therefore, such changes might influence the recruitment success of small pelagic fish populations (Chassot, Floch, Dewals, Pianet, & Chavance, 2011;Longhurst, 2010;Moore, Harvey, & Van Niel, 2009;Oliver & Irwin, 2008).
Round sardinella (Sardinella aurita) and flat sardinella (Sardinella maderensis), the only Sardinella sampled in the Senegalese area, account for more than 80% of the Senegalese artisanal fisheries total landings (Diankha et al., 2017;FAO, 2012), which are marked by strong seasonal and interannual fluctuations (Thiaw et al., 2017). Both species have short-lived, zooplankton feeders with similar shape and size (maximum fork length~30 cm; Ba et al., 2016), and occupy almost the same geographical areas over the Senegalese continental shelf (Cury & Fontana, 1988). However, these species have some physiological differences. Sardinella aurita undergoes an intensive seasonal migration from Morocco to Guinea passing through Mauritania, Senegal and Gambia due to the spatial variability and strong seasonality of the Senegalo-Mauritanian Upwelling System (Bo€ ely, 1980;Bo€ ely & Fr eon, 1979). This seasonal displacement makes its spawning heterogeneous, which occurs preferentially over the Arguin Bank (Mauritania) and the south of Cap-Vert (Senegal;Bo€ ely, Chabanne, & Fr eon, 1978;Bo€ ely, Chabanne, Fr eon, & St equert, 1982;Conand, 1977;Tiedemann & Brehmer, 2017). The migration of S. maderensis is less marked around its nursery area (Bo€ ely, 1982;Cury & Fontana, 1988), suggesting that this species is able to adapt more readily to environmental variations.
Sardinella maderensis is less sensitive to climatic/environmental fluctuations than S. aurita and has a less flexible adaptive strategy resulting in a smaller plasticity of their biological parameters (Ba et al., 2016).
Several studies attempted to investigate the dynamic of sardinella populations in relation to environmental change. Sabat es, Martin, Lloret, and Raya (2006) showed that abundance of S. aurita in West Mediterranean waters is positively correlated to sea surface temperature (SST). In Mauritanian waters, it has been demonstrated that high abundance of S. aurita is associated with SST below 21°C (Zeeberg, Corten, Tjoe-Awie, Coca, & Hamady, 2008). Diankha, Wade, et al. (2015) suggested that important catches of S. aurita in Senegalese waters occur with SST ranges of 22-24°C. Recently, Bacha, Jeyid, Vantrepotte, Dessailly, and Amara (2016) demonstrated that a high part of the catch variability of S. aurita in Mauritanian waters could be associated with SST variations. New investigations carried out by Thiaw et al. (2017) showed that the variations in sardinella biomass in Senegalese waters were driven by environmental conditions. Similar studies on the second sardinella, i.e., S. maderensis, are scarce.
However, it is worth highlighting that the assessment of reliable abundance indices of small pelagic fish from artisanal fisheries landings remains complex (Ould Talib Sidi, 2005). It relies on parameters difficult to estimate such as the fishing effort (Mangel & Bede, 1985), particularly in the artisanal sector targeting the sardinella which are mainly targeted when occurring in fish school structures (Brehmer et al., 2007).
The approach used in this study avoids misleading interpretation due to eventual bias in the catch per unit of effort (CPUE) estimation because the number of trips (outings) considered as fishing effort is not a proxy of the real effort. Moreover, the relationships between pelagic habitats and marine resources are fairly complex due to of their fundamentally non-linear characteristics (Bo€ ely, 1982;Rothschild, 2000). This is why links between the fluctuations of environmental conditions and the abundance of sardinella is still not well understood and difficult to predict for management and decision making support.
This study contributes to a better understanding of the relationships between environmental variability and the local abundance fluctuations of sardinella. It aims to define the environmental conditions associated with recruitment success of S. aurita and S. maderensis in Senegalese waters, based on the monitoring of landings per size classes. The influence of four environmental variables on the recruitment of both species is tested and quantified over one decade.

| Study area
The study area is the Senegalo-Mauritanian upwelling system ( Figure 1), a part of the Canary Current Large Marine Ecosystem (CCLME), influenced by the Canary Current (CC) flowing along the African coast from north to south between 30°N-10°N and offshore to 20°W (Fedoseev, 1970). This upwelling system begins in late autumn and ends in spring (Teisson, 1982). In the region north of Dakar, where the continental shelf is narrow, the upwelling occurs near the coast, while south of Dakar characterized by a larger continental shelf and the upwelling is a trapped cold water tongue in the middle of the continental shelf surrounded by warmed waters (Ndoye et al., 2014). The Senegalese upwelling is marked by great seasonal and interannual variability thought to have effects on small pelagic fish (Bakun, 1996;Cury & Roy, 1987;Fr eon, 1991;Fr eon & Mendoza, 2003).

| Biological fisheries data for both Sardinella
Landing data per size classes were used to estimate sardinella recruitment applying the virtual population analysis (VPA) cohort modeling technique widely used in fisheries science (Jones, 1984;Lassen & Medley, 2001;Pope, 1972). Monthly data were used in this work combining two data sets: (i) total landing of both sardinella F I G U R E 1 Map of the study area situated on the Senegalese coast showing the main sardinella landing ports. The landing ports are grouped by sub-region: the northern coast (e.g., Saint-Louis, and Cayar), the Dakar region (e.g., Yoff, Ouakam, Soumb edioune and Hann) and the south of the country (Mbour and Joal; reproduced from Thiaw et al., 2017) The number of trips per fishing gear was recorded on a daily basis, while landings data were randomly collected for about 5 days a week. After aggregating the data by port, gear and period (fortnightly), total landings per port were estimated by multiplying mean landings of sampled trips by the total number of fishing trips (Chaboud et al., 2015;Thiao, 2009). Note that landings were summed per month for each landing port (Chaboud et al., 2015;Thiao, Ngom, & Thiam, 2012). (ii) A secondary data set was only used for the VPA, the monthly size distributions of both species from the same landing ports within a period of 15 months between July 2014 and September 2015. The total number of measured individuals was 96,963 split between 51,295 individuals of S. aurita and 45,668 individuals of S. maderensis. It is worth mentioning that fish samples were randomly taken while individual weights as well as size were recorded. To estimate the recruitment from VPA, the number of individuals per size classes is first transformed into size frequencies. Missing data were then substituted by the average of the corresponding missing month of the time series for the same fishing port. We assume that all monthly landings were distributed according to this average size frequency per fishing port. Then we used a size-age key to convert size class data into seasonal age groups by applying the growth equation of von Bertalanffy (Matsinos & Wolff, 2001) using the growth parameters of sardinella (Fr eon, 1988). Seasonal time step (winter: January-March, spring: April-June, summer: July-September and autumn: October-December) was chosen to consider biological characteristics of both species, fast growth and short life cycle (Bo€ ely, 1980). Seasonal and interannual sums were also computed.

| Environmental variables
Monthly data concerning five environmental variables often applied to investigate relationships between fish abundance and environmental conditions (Bacha et al., 2016;Klemas, 2012;Thiaw et al., 2017) were used in our study: sea surface temperature (SST, in°C), chlorophyll-a concentration (Chl-a, in mg/m 3 ), windinduced turbulence (WTI, in m 3 /s 3 ) and a Coastal Upwelling Index (CUI, in m 3 /s 3 per meter of coast) deduced from wind data and North Atlantic Oscillation index (NAO). Except NAO, all variables were averaged from the coast to the 200 m isobaths, i.e., over almost all the Senegalese sardinella habitat, the continental shelf from 12 to 17°N ( Figure 1). Their seasonal and interannual means were also computed.
SST is commonly used to investigate relationships between environment and fish abundance (Kellogg & Gift, 1983;Ramos, Santiago, Sangra, & Canton, 1996). Chl-a is considered as an index of biological productivity (Lorenzen, 1970) and a proxy of food abundance. In this work, we used the Aqua-MODIS "Level 2" SST and Chl-a data from January 2004 to December 2013 obtained from the NASA Goddard Space Flight Center (GSFC) through the NASA web site (http://oceancolor.gsfc.nasa.gov). The spatial resolutions of these data were 4 km.
Wind speed data (m/s) from January 2004 to December 2013 were obtained from the historic reanalysis wind data, from the National Center for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR, 2.5°resolution). We defined the WTI as the cube of the wind speed, similarly to early ecological studies (Bacha et al., 2016;Ueyama & Monger, 2005). CUI were estimated from the equation: where s is the along shore component of wind stress within the coastline (from Saint-Louis to Joal), q w is the seawater density (1,025 kg/m 3 ) and f is the Coriolis parameter (2Ω sin (h), with Ω and h equal to the Earth's angular velocity and latitude, respectively).
The NAO is considered to be the most important mode of atmospheric variability over the North Atlantic Ocean, and plays a major role in weather and climate variations over the North Atlantic continent (Hurrell, 1995(Hurrell, , 1996

| Statistical modeling approach
The general additive model (GAM) of Hastie and Tibshirani (1986)  The "mgcv" package in the R software (Wood, Scheipl, & Faraway, 2013) was used. The Gaussian distribution and the smoothing function "ti" were applied because they provided the most natural fit for the transformed recruitment data after a stepwise procedure. It ("ti") excludes the basic functions associated with the "main effects" of the marginal smooth, plus interactions other than the highest order specified (Wood et al., 2013). The degree of freedom for each spline smoother was set to 4 to avoid overfitting. The time unit for environmental variables is season as recruitment was estimated by season in order to consider the biology of these species (Bo€ ely, 1982). The seasonal and interannual means of these five variables were also computed.
The GAM applied for each species was formulated as: The date included in the model consisted of seasonal data from 2004 to 2013, so the number of observations are n = 40.
The relative importance of each variable in the total deviance was determined from the "relaimpo" R package (Tonidandel & LeBreton, 2011). The application of this package allows the partition of the total explained deviance among the four predictors to better understand the role played by each one.
The variance inflation factors (VIFs) were calculated for all environmental variables in order to detect possible high dimensional collinearities (Zuur, Elena, & Chris, 2010). In fact, it was suggested by these authors that covariates with VIFs >5 are highly collinear. However, all VIF values calculated here were    data which strongly depends on fishing effort reliability, a parameter difficult to quantify (Mangel & Bede, 1985), especially for Senegalese artisanal fisheries where the unit of effort is expressed in number of outings. This does not reflect the real changes in fishing effort due to, for example, the non-consideration of the increase of distance traveled by the canoe and fishing activity duration, as reported in Ba et al. (2017). Moreover, the quick and important development of the Senegalese artisanal fishery sector, along with structural and conjectural changes, makes the choice difficult for the fishing effort unit.
The use of the VPA may be limited by several sources of potential uncertainties, mainly related to catch extrapolations, assessment of discards and estimation of fishing mortalities used to initiate the cohort analysis (Gascuel, 1994). In our case, because of their high demand in the fish market, discards of sardinella are not frequent.
Therefore, biases related to the distribution of catches by age class could be assumed to be negligible. Moreover, the sardinella are migratory species and the two main nursery areas are situated in Mauritania (banc d'Arguin) and Senegal (petite cote); thus the sampling had covered a complete annual cycle, and such facts confirm our assumption that the whole populations have been sampled.
Results show that sardinella recruitment, like the environmental variables, are characterized by strong temporal variability. Both sardinella show similar seasonal variation in recruitment with a minimum and maximum in winter and summer, respectively. Winter and summer are opposite hydrological seasons. In Senegalese water, winter is characterized by cold waters, strong wind-induced turbulence and upwelling intensity, while the opposite occurs in summer. These results seem to suggest that at this stage, sardinella prefer oceanographic conditions in the summer. These results are in accordance with previous studies which suggest that variability in abundance and distribution of sardinella is mainly under the influence of oceanographic changes, such as cooling or warming waters (Bacha et al., 2016;Sabat es et al., 2006;Thiaw et al., 2017;Zeeberg et al., 2008).
The use of GAM, which takes into account the potentially complex non-linear relationships between the covariates, allowed a better illustration of the links between S. aurita, as well as S. maderensis and the oceanographic conditions in off Senegal. The environmental variables considered in this work (SST, Chl-a, WTI CUI and NAO) were responsible for 76.7% and 66.1% of the variability of S. aurita and S. maderensis recruitments, respectively. The CUI and SST together accounted for 53.1% and 30.9% of the total explained variance for S. aurita and S. maderensis, respectively. Our results are in accordance with those found in Mauritanian waters, showing that SST and CUI were also the main environmental parameters explaining the variability of S. aurita (Bacha et al., 2016). The CUI played the major role in S. aurita recruitment variability, while for S. maderensis the SST was the key factor. However, SST was the second most important variable in S. aurita recruitment variability.
The role of the upwelling intensity on S. aurita abundance was already well established (Bacha et al., 2016;Braham et al., 2014;Mbaye et al., 2015). Nevertheless, to our knowledge, studies quantifying the effect of upwelling intensity on S. maderensis abundance have not previously been reported. The effect of SST and CUI does not hide the influence of Chl-a and WTI which had significant effect on both sardinella recruitment success in this study. These observed differences might be due to their physiological characteristics which are relatively distinct: S. aurita is more sensitive to temperature and salinity fluctuation than S. maderensis, which tolerate higher change in salinity and temperature (Ba et al., 2016;Cury & Fontana, 1988). This is also explained by the fact that, even if both species occupy the same area in Senegal (Cury & Fontana, 1988), their relative abundance strongly differs between Northern Senegal (in connection with Mauritania) and Southern Senegal (Capet et al., 2016;Ndoye et al., 2014), warmer and less sensitive to the detrimental effects of a strong upwelling season.
The role of temperature on small pelagic fish stock dynamics has been previously reported (Pepin, 1991). Temperature always affects fish populations at different stages of their life cycles, including during spawning and the development and survival of the eggs and larvae, as well as influencing their distribution, diet, migration pattern and schooling behavior (Gordoa, Maso, & Voges, 2000;Laevastu & Hayes, 1981;Sund, Blackburn, & Williams, 1981). Several studies have shown that abundance of sardinella is related to SST variability (Bacha et al., 2016;Diankha, Wade, et al., 2015;Sabat es et al., 2006;Thiaw et al., 2017;Zeeberg et al., 2008). However, S. maderensis is more associated with warmer waters (>24°C), as shown by Bo€ ely (1979), than S. aurita. Furthermore, the ideal balance between a moderate upwelling and warmer waters in Southern Senegal highly favors S. maderensis recruitment and its abundance, and makes it predominantly sensitive to environmental thermal variations and prey abundance with a low direct dependency of the upwelling intensity, whose detrimental effects are low in this region, characterized by a high physical retention capability of eggs and larvae (Demarcq & Faure, 2000;Mbaye et al., 2015;Roy et al., 1989).
The landings of small pelagic fish in the Pacific may be related to Chl-a (Ware & Thomson, 2005) and the influence of Chl-a is more important on sardine (Sardina pilchardus) than on S. aurita in Mauritania (Bacha et al., 2016). We also found a moderate influence of Chla on the recruitment success of S. aurita because its variability seems to be partly hidden by the upwelling intensity, whose effect is a direct increase of the primary production.  et al., 2008). Unfavorable conditions mainly include low temperature having a negative effect on their growth (Cole & McGlade, 1998). In addition, it was suggested that a marine environment with Chl-a above 0.2 mg/m 3 could support sustainable fisheries (FAO, 2003;Gower, 1986). The pioneering work of Cury and Roy (1989) suggested that in upwelling areas there is an optimal wind speed value outside which recruitment success is lower. However, such optimal wind speed values are different from those found in this study. Cury and Roy (1989) highlighted that wind speeds values outside 5-6 m/s (corresponding to WTI from 125 to 216 m 3 /s 3 ) had a negative effect on sardinella CPUE in Senegal, whilst we found a constantly higher recruitment success for low wind values for both sardinella species.
This difference is probably due to the use of distinct indices in Cury and , and because they have worked on the S. aurita CPUE as abundance index, while our study considered the recruitment.

| CONCLUSION
The results obtained on the estimation of S. aurita and S. maderensis recruitment in Senegalese waters showed how their success is related to direct environmental conditions and the existence of different ranges of environmental conditions (optimal environmental windows) associated with high recruitment of S. aurita and S. maderensis. Moreover, both clupeid species present distinct responses to these environmental factors probably because of their different physiological and ecological intrinsic characteristics. The explanations of our variables are unexpectedly high (76.7% and 66.1%). In other words, we obtained a good explanation of recruitment success with only environmental variables that describe the in situ environmental conditions, or habitat of the exploited fish.
As sardinella, mainly round sardinella (S. aurita), perform strong seasonal migration it suggests that national policy prerogatives for the management of such fisheries should be done at the regional level. Such an approach to facilitate common sub-regional management plans should allow a better mitigation of loss and damage in the fisheries sector in the context of overfishing and climate change.

ACKNOWLEDG EMENTS
An early draft of this paper was presented at an international confer- project. We also thank the spatial agencies, mostly the NOAA and ESA for providing continuous high quality environmental data sets.
We would also like to thank the anonymous referee comments and suggestions that greatly helped to improve this paper.