The Argentine shortfin squid, Illex argentinus, is a common neritic species occurring in waters off Brazil, Uruguay, Argentina, and the Falkland/Malvinas Islands in the Southwest Atlantic (Haimovici et al., 1998). It is the most economically important cephalopod species for China and many other countries (Chen et al., 2008). Illex argentinus is an opportunistic feeder (Ivanovic and Brunetti, 1994; Brunetti et al., 1998a) growing rapidly with a short life cycle and a high degree of intra-population differentiation (Arkhipkin, 1993, 2000). Meanwhile I. argentinus is migratory species, and their concentrations are usually found at 45°–46°S in January or February, while growing rapidly this squid subsequently migrate southward towards the Falkland Islands. Peak concentrations are found around the Falkland Islands between March and May. Toward the end of this period, this squid starts to migrate northward, ultimately to spawn and die in the shelf and slope waters off northern Argentina, Uruguay and Brazil around July or August (Basson et al., 1996; Brunetti et al., 1998b). The population of I. argentinus has been separated into four stocks based on their lengths at maturity, areas and timing of spawning, and the distribution of early juveniles and adults life stages: South Patagonic Stock (SPS), Bonaerensis-Northpatogonic Stock (BNS), Summer Spawning Stock (SSS) and Southern Brazil Stock (SBS) (Hatanaka, 1986). Of the four stocks, the SPS is the dominant commercial stock and is mainly exploited by some areas in China’s mainland (Lu and Chen, 2012), Chinese Taipei (Chen and Chiu, 2009), and Falkland (Waluda et al., 1999).

The oceanographic environment for I. argentinus is mainly influenced by joint Brazil and Malvinas/Falkland Currents around 33°–39°S, causing an important thermohaline front that separates the subtropical waters from the subantarctic waters (Legeckis and Gordon, 1982; Olson et al., 1988; Gordon, 1989). Strong thermal and saline gradients can be observed in this convergence area, due to interactions of the two currents and the influence of the Patagonian Current and the discharge of the Río de la Plata, which provides nutrients for high levels of primary production (Haimovici et al., 1988).

Previous studies suggest that I. argentinus tends to be highly susceptible to environmental changes during all its life histroy. For example, Waluda et al. (2001) suggested that large-scale oceanographic variability in the location of the spawning/hatching grounds during the early life stage of I. argentinus was important in determining recruitment to the fishery. Bazzino et al. (2005) identified environmental associations (depth, bottom temperature and bottom salinity) of shortfin squid in the Northern Patagonian Shelf, and found that squid distribution showed significant associations mostly with depth and bottom temperature. Sacau et al. (2005) developed generalized additive models (GAMs) of squid abundance in relation to physical and environmental conditions including sea surface temperature (SST), latitude, longitude, month, average fishing depth and year to be applied to fishery forecasting, and predictors retained in the optimal model. However, other environmental variables, such as sea surface height (SSH) and chlorophyll a (Chl a) are also important indicators of physical oceanographic processes, and their impacts on the squid distributions received little attention (Chen et al., 2012).

In this study, we evaluated the influence of SST, SSH and Chl a on the spatial distribution of I. argentinus in the Southwest Atlantic, and developed a neural network model for forecasting spatial CPUE distribution. We then linked spatio-temporal variability in these oceanographic variables with the migration pattern of I. argentinus. The framework developed in the study provides an approach for forecasting spatial CPUE distribution and migration pattern of I. argentinus in the Southwest Atlantic, which can also be used for the other oceanic squid species.

Understanding and projecting how fish species react to climate change and variability in the regional/local oceanographic environmental is essential for the effective management of marine resources (Waluda et al., 2001). Illex argentinus abundance and distribution are found to be significantly influenced by surrounding environmental conditions (Waluda et al., 2001; Bazzino et al., 2005; Chen et al., 2012; Sacau et al., 2005). In this study, the integrated approach which using GAM to select significant variables and using NNM to establish relationships may overcome disadvantages associated with the use of sub-models alone. The GAM can interpret the relationships between results and factors, but its inference ability may be somewhat problematical, such as hypothesis test or obtaining confidence intervals for the fitted values caused by fitting process (Venables and Dichmont, 2004); neural network model is a “black box” model of obtaining a good understanding of the underlying mechanisms difficultly, but has strong ability of fitting non-linear relationships (Paruelo and Tomasel, 1997). The aim of combining the GAM and the ANN is to obtain more accurate prediction with better explanatory between independent and dependent variables, though the ANN is more capable in dealing with nonlinear relationships than GAM.

In the GAM analysis, all the environmental variables (SST, SSH and Chl a) had significant impacts on squid abundance (Table 1). The highest CPUE was found in temperature between 10–16°C, this was similar to the conclusion obtained by different methods based on different fishery data in other studies (e.g., Sacau et al., 2005). The other two environmental variables (SSH and Chl a) are rarely utilized for analyzing the distribution of I. argentinus, but have been used to other species, such as neon flying squid Ommastrephes bartramii in the Northwest Pacific Ocean (Wang et al., 2015). The Chl a concentration maybe a good indicator of the food availability for squid with high Chl a concentration yielding good feeding environments (Nishikawa et al., 2014). The relationship between Chl a concentration and squid abundance is possibly linear in the GAM. The SSH filed may be effective for predicting a water mass front, which is a potential aggregation mechanism for planktons as well as their predators, such as squid (Polito et al., 2000). The relationship between SSH and squid abundance demonstrated non-linearity being the highest CPUE found at SSH between –60 and –20 cm.

For the NNM, one of the greatest advantages is predictive ability. Though there are some methods to try to interpret model mechanism (Özesmi and Özesmi, 1999), it is insufficiently specific in our research. For example, we just knew the relationship between squid abundance and environmental variables is non-linear in neural interpretation diagram, but did not know the exact form of non-linear. Thus, we just took full advantage of NNM to predict spatial CPUE of I. argentinus. In fact, we also did try to predict spatial CPUE distribution using GAM model fitted on the same data samples, it was found that the outcomes were not better than NNM’s. Additionally, the relationships between squid and environmental variables may be varied among different life-history stages. However, the current neural networks developed in this paper were general models those mixed the effect of environmental gradients (spatial effects) with life-history changes (temporal effects), it tends to be risky.

In this study, the annual maps of monthly CPUE distribution (Fig. 1), especially in monthly CPUE gravities shift (Fig. 6a), exhibit the migration pattern of I. argentinus during January to June. The results were consistent with the conclusion reported by Sacau et al. (2005) and Waluda et al. (2001). Previous work suggests that the life-cycle of I. argentinus is associated with the subtropical confluence of the Brazil and Falkland Currents during reproduction and the early life stages (Brunetti and Ivanovic, 1992; Hatanaka, 1988) and with the Falkland Current over the Southern Patagonian shelf during maturation, feeding and growth (Rodhouse et al., 1995). In order to evaluate this mechanism with environmental variables, we plotted maps of monthly catch distribution in 2005 overlapped it with SST, SSH and Chl a concentration, because the catch in 2005 was relatively steady (Figs 6, 7 and 8). From the maps, the catch locations were basically distributed in warm-cold (Brazil-Falkland) confluence in the SST map (Fig. 6), between high and low surface layer especially along low surface layer (about –80 cm) in the SSH map (Fig. 7), on the road of high Chl a concentration that interweaved with dark environment where almost no Chl a concentration in the Chl a concentration map (Fig. 8). The south-north migration pattern for I. argentinus was consistent with the change of warm-cold (Brazil-Falkland) confluence from January to June. However, the migration pattern was not obvious in SSH and Chl a maps, because the catch locations were looked like constant in a fixed environmental surrounding. This suggested that I. argentinus lives in special habitat from SSH and Chl a concentration perspective.

In summary, the integrated model was well developed for representing and predicting the spatial distribution of I. argentinus squid with approximately 82% average accuracy, the model could be further used for spatial habitat reconstructing. But more work should be done in further. For example, if the neural network models of I. aregentinus were developed by monthly or other environmental variables, such as sea surface salinity and sea depth temperature (Yu et al., 2015), can be obtained, the better model would be developed.

The authors thank the Chinese Squid-Jigging Technology Group of Shanghai Ocean University for providing the catch data and the National Oceanic and Atmospheric Administration for providing the environmental data.