Spatio-temporal analysis of commercial trawler data using General Additive models: patterns of Loliginid squid abundance in the north-east Atlantic

V. Denis J. Lejeune J. P. Robin General Additive Model (GAM) fitting techniques have been employed to understand and predict cephalopod abundance variations in the north-east Atlantic using data on commercial fisheries and geographic and climatic variables. Spatial patterns were studied for general average, seasonal average and monthly recruitment abundance. The capability of this method to model abundance in time and space was tested by comparing observed and calculated Catch per Unit of Effort for a 1 longitude by 0.5 latitude rectangular grid. The influence of the explanatory variables on Loliginid abundance was clearly shown. Climatic variable effects change with time scale. They vary during the year and are mainly important during the pre-recruitment month. GAM allows explanation of the main part of seasonal abundance variations of these species in time and space. GAM also provides the first means of predicting the main recruitment peak area by using previous-month climatic variables. This article demonstrates the advantages of using commercial-fisheries data for ecological studies. 1054-3139/02/060633+16 $35.00/0 - Introduction Understanding ecosystems in general and environmental effects on the variability in abundance and distribution of exploited fishery stocks in particular, are keys for future management strategies (Hofmann and Powell, 1998). Indeed the influence of environmental parameters on the distribution and abundance of marine species has been a question of major interest for scientists since the beginning of the 20th century. The development of climatic data recording allowed the first experimental research into this subject (Kemp, 1938) and since then numerous articles have addressed the relationships between climate, distribution and abundance (Russell, 1935; 1939; Southward, 1960, 1980; Cushing, 1982; Sinclair and Frank, 1995; Beamish, 1995; Southward et al., 1995). Target species are the subject of many studies: by-catch species are often overlooked. Research on the relationships between the abundance of cephalopods, such as loliginid squid, and environmental parameters in the north-east Atlantic has increased in the last 10 years in line with the increasing commercial interest. Loligo forbesi and Loligo vulgaris, which are mainly caught in this area, represent an important source of income for fishermen (Anonymous, 2000a). The life cycle and distribution of these two species have been described by Holmes (1974), Roper et al. (1984), Worms (1983), and their exploitation by the European fleet has been studied by Pierce et al. (1994), Robin and Boucaud-Camou (1993, 1995), Cunha and Moreno (1994), Guerra et al. (1992, 1994), and Denis (2000). Temporal and spatial changes in the relationship between squid abundance and climatic variables have been observed (Roberts and Sauer, 1994; Pierce, 1995; Pierce et al., 1998; Waluda and Pierce, 1998; Robin and Denis, 1999; Bellido et al., 2001). In these studies, Pearson correlation coefficients computed per ICES-rectangle showed that rectangles which have interannual common trends were grouped spatially. The location of these groups of rectangles with significant (positive or negative) correlation changed according to the month selected for climatic records. Ordinary linear regression techniques failed to describe such a complex system. Generalized Additive Models (GAM) techniques seem to provide a more powerful tool. They have already been used in the spatio-temporal stock assessment modelling of marine species such as Atlantic mackerel (Scomber scombrus), sprat (Sprattus sprattus), whiting (Merlangius merlangus), anchovy (Engraulis encrasicolus), and horse mackerel (Trachurus mediterraneus) (Borchers et al., 1997; Augustin et al., 1998; Daskalov, 1999). In all these studies the authors chose to use GAM techniques to model the spatio-temporal distribution of studied species abundance as a function of geographical and environmental variables. The application of GAMs has been described by Hastie and Tibshirani (1990). They show that the flexibility of non-parametric regression or smoothing when added to general linear models allows the uncovering of structure in the data that might otherwise have been missed with the usual linear assumptions. Many of the cited papers show results obtained by using GAM techniques on fish-survey data sets. This article presents for the first time GAM analysis fitted between a 10 year geo-referenced French commercial fishery data set and bathymetric and climatic data in the north-east Atlantic. No survey data are available for the study area so that commercial data are therefore the only means of obtaining information with these complete spatial and temporal scales. The objectives of this study can be summarized as: (1) Can a Generalized Additive Model explain and model the abundance of non-target species via commercial data? (2) What effects do geographic and climatic variables have on Loliginid squid ecology? Materials and methods This study makes use of existing data sets which were collected during the development of French GIS tools for cephalopod studies. Data sets used French trawlers monthly Landing Per Unit of Effort (LPUE) per ICES rectangle were computed for the period 19891998 with total landing and effort. The units of effort are hours fishing for a standardized 1000 kW boat. Commercial landing and effort data for French vessels were obtained from the Centre Administratif des Affaires Maritimes (CAAM). To test the difference between LPUE and Catch Per Unit of Effort (CPUE) for Loliginid squid, a series of on-board observations were made in order to estimate the discard rate of the French offshore trawlers since these vessels provide most of the cephalopod landings. Thirty-one days of shipboard observations were made on vessels from Port-en-Bessin (Normandy) between July and October 1997 in the western part of the English Channel and estimates made of cephalopod discards during this period from the total landings. The monthlymean weight calculated from fish market observations were used to assess the number of animals caught per cruise for Loliginids and cuttlefish. This assessment provides an estimate of the proportion of rejected animals during the 556 fishing hours. Climatic data were obtained from the Meteo-France Marine data base AVISO. Monthly-mean sea surface temperature was obtained as a 1 longitude by 1 latitude grid. Monthly averages were computed and the 1 by 1 climatic grids were converted into the same resolution as the ICES rectangle grid by a spatial linear interpolation. The original data sets given by AVISO are the results of models using satellites and in situ data from the European Centre for Medium-Range Weather Forecasts (ECMWF). Climatic parameters obtained are: Sea Surface Temperature (SST) in Celsius degree Solar Flux in Joules per squared metre Sea level pressure in Pascal Wind Speed in metre per second Wind direction in degree (...truncated)