Fisheries Research 79 (2006) 294–302 Using the bootstrap to investigate the effects of varying tow lengths and catch sampling schemes in fish survey Santiago Cervi˜ no a,∗ , Fran Saborido-Rey b a Instituto Espa ˜ nol de Oceanograf´ ıa, Cabo Estai, Canido s/n, 36200 Vigo, Spain b Instituto de Investigaciones Marinas (CSIC), Eduardo Cabello 6, 36208 Vigo, Spain Received 4 April 2005; received in revised form 8 March 2006; accepted 14 March 2006 Abstract In this paper we explore the application of bootstrap methods to analyse of errors in trawl survey indices of abundance at age, with focus on the relative participation of the three sampling levels: haul design, sampling of lengths and sampling of ages, and their implications for survey accuracy. The method consists of resampling these three sources of variability, independently and together, following the sampling scheme and comparing the results. Our results show that although the haul design component is the main source of variability of abundance at age, the importance of catch sampling (sampling of lengths and ages) may be substantial, especially at low abundance levels. Furthermore, we have used our method as a survey simulator; as an example we have performed a simulation with an alternative sampling scheme, i.e. reducing the tow duration and spending the saved time performing more hauls. The simulation shows that the alternative scheme improves the accuracy of abundance at age. This survey simulator may be used as a tool to evaluate other alternative sampling schemes. We have used the cod (Gadus morhua) data from the Flemish Cap survey as a case study but the method may be adapted to different survey procedures. © 2006 Elsevier B.V. All rights reserved. Keywords: Trawl survey; Catch sampling; Bootstrap; Simulations; Flemish Cap cod 1. Introduction Estimates of abundance of fish populations obtained from bottom trawl surveys provide a major source of fisheries inde- pendent information for management purposes. When the catch data are not reliable or do not exist, survey indices are the only source of information to assess the state of a fishery, and can be used for such (Pennington and Strømme, 1998; Korsbrekke et al., 2001). When there are good catch statis- tics and the age composition of the population is known, the virtual population analysis (VPA) is frequently used to assess the fishery. In these circumstances, the survey data are used to calibrate this model (Shepherd, 1999). In both cases, with and without catch data, the indices of abundance at age show the trend in the evolution of the population and ∗ Corresponding author. Tel.: +34 986 492111; fax: +34 986 498626. E-mail addresses: santiago.cervino@vi.ieo.es (S. Cervi ˜ no), fran@iim.csic.es (F. Saborido-Rey). the accuracy of these indices determines the quality of the assessment. Three different approaches have been applied to improve the quality of the survey results: one is based on the design of the sampling (Gavaris and Smith, 1987; Smith, 1990); the second consist of looking for more realistic statistical models like log-normal, gamma or delta log-normal to fit the abun- dance distributions (Myers and Pepin, 1990; Pennington, 1996; Steff´ ansson, 1996); and the third uses resampling meth- ods like Monte Carlo or bootstrap. Bootstrap methods were initially implemented to calculate the standard error of some statistics that otherwise would be difficult to perform (Efron and Tibshirani, 1993; Manly, 1997) and have been applied to fishery surveys in various studies (Sigler and Fujioka, 1988; Smith and Gavaris, 1993; Smith, 1997; Pennington et al., 2002; Schunute and Haigh, 2003). These three approaches to improve the survey results have focused mainly on the inter-haul errors but less attention has been directed to the intra-haul variability (Lai, 1993; Cotter, 1998; Pennington 0165-7836/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.fishres.2006.03.021