Elsevier

Fisheries Research

Volume 181, September 2016, Pages 34-47
Fisheries Research

Modeling landings profiles of fishing vessels: An application of Self-Organizing Maps to VMS and logbook data

https://doi.org/10.1016/j.fishres.2016.04.005Get rights and content

Highlights

  • Logbook data can be effectively combined with other data sources such as VMS.

  • A SOM was trained on a large Italian logbook dataset for which VMS were available.

  • SOM returned a pattern driven by fishing gears and ecology of the target species.

  • The VMS-determined depth of fishing activities plays also an important.

  • The trained SOM is also effective in predicting gear by landings profiles.

Abstract

Logbook data constitute a key element within the electronic recording and reporting system of the European Fisheries Control Technologies Framework and are used to record, report, process, store and send information about fishing operations, including landings and fishing gear. A relevant application of logbook data is to account for the heterogeneity of fishing practices (e.g., by gear or métier), which is a key aspect of the Common Fishery Policy. However, despite their importance, few published studies have explored the potential and pitfalls of logbook data, even in combination with other powerful data sources such as the Vessel Monitoring System (VMS). Here, a new approach to characterizing the composition of landings for the different types of gear based on the use of Self-Organizing Maps (SOMs − a particular type of Artificial Neural Network) is applied to the Italian fleet logbook dataset. The SOM is trained on the landings composition and the resulting patterns are interpreted using some measures obtained from the analysis of the corresponding VMS data. Namely, the mean sea bottom depth and the area of activity are obtained for each fishing trip. Moreover, the ability of the trained SOM to predict gear from landings is tested using a new dataset. The trained SOM classifies logbook records according to the ecological, taxonomical, and trophic characteristics of the species caught, and the depth of fishing activities plays an important role in diversifying the landings associated with certain widely used fishing gear such as the bottom otter trawl. The clustering of SOM units allows the identification of a set of 12 groups, which are strongly related to the types of gear used by the Italian fleet. Furthermore, the trained SOM shows a high ability to recognize gear from logbook data, thus confirming the robustness of the landings profiles detected.

Introduction

Modern fisheries science uses various sources of data to assess the status of exploited resources at the best level possible with respect to the economic and technical constraints of data collection (Acom, 2012, Bastardie et al., 2010, Cotter and Pilling, 2007, Marchal, 2008a, Russo et al., 2013, Russo et al., 2014b, Russo et al., 2015, Sampson, 2011, Tingley et al., 2003). The main distinction within data sources is between fishery-independent and −dependent information. The first basically refers to scientific surveys, which are designed to maximize spatial and biodiversity coverage (Bertrand et al., 2002). The second refers mainly to two sub-types of data: direct observations by scientific personnel either on commercial fishing vessels or at harbors. These two sub-sources of fishery-dependent data differ in their reliability, accuracy and power in terms of the absolute coverage of fishing activities and landings (Cotter and Pilling, 2007). Although observers are likely to be more reliable and accurate than fishers in collecting and annotating data about catches, landings, and so on, their coverage is generally limited to a small portion of the entire fleet (a sub-sample). In contrast, self-reported data are mandatory and are expected to cover the entire fleet. The core of the EU system for fisheries controls (http://ec.europa.eu/fisheries/cfp/control/technologies/ers/), defined within the European Fisheries Control Technologies Framework (Commission, 2009, Commission, 2011a, Commission, 2011b), is the Electronic Recording and reporting System (ERS), which is used to record, report, process, store and send fishery data (catches, landings, sales and transshipment). The key element of the ERS is the electronic logbook in which a member of the crew (generally the captain/skipper of a fishing vessel) keeps a record of fishing operations, including landings. This process is critical because monitoring fishing activity is the basis for fishery management (Cotter and Pilling, 2007, Gerritsen and Lordan, 2011, Tzanatos et al., 2008). Although logbook data remain the primary source of information on landings, few published studies have explored their potential and their issues, such as data quality. Recently, Sampson (2011) detected an inverse relationship between the consistency of information about fish location and the accuracy of catches, suggesting the reluctance of fishers to release complete information about fishing grounds positions and the abundance of resources. Currently, the availability of different tracking devices (e.g., VMS and Automatic Identification System—AIS) makes it possible, through appropriate methodologies and tools (e.g., VMSbase, Russo et al., 2014a), to overcome the uncertainty regarding fishing locations that may be introduced by skippers’ reporting (Chang, 2011). Contextually, according to Sampson (2011), the inaccuracy of reporting strongly supports the idea of using statistical tools that can minimize the distortion caused by mistakes or false information. Another relevant point is that logbook data are currently used to account for the heterogeneity of fishing practices (e.g., by gear or métier) (COUNCIL, 2014), as a key aspect of European Common Fishery Policy is the transition from single-stock management towards a fleet-based approach (EU, 2013). It is thus fundamental to define, at the best possible level of detail, a list of fishing operations (métiers) that differs in terms of specific gear used, target assemblages, and mesh size (Castro et al., 2010, Davie and Lordan, 2011, Pelletier and Ferraris, 2000). Although time and space are implicitly part of the definition of a métier, the gear and target species/assemblage represent the two main discriminators, with the variability due to time and/or space being more or less evident for the different gear types. However, given the importance of logbook data as a source of information on fishing activities, different methodologies have been used to group logbook records into métiers, to define métiers themselves and, ultimately, to compare and standardize a métier list across different fleets or countries. In this way, the Data Collection Framework for Fisheries (DCF) has progressively detailed the definition of métiers through a dichotomic key (Fig. 1), articulated at different hierarchical levels, in which the highest levels (i.e., 5–6) attempt to match the definition of métiers provided above. In general, raw logbook records do not allow métiers to be determined directly (under the assumption that DCF levels 5/6 are a good proxy for métiers) but instead require the reported information to be processed. According to Deporte et al. (2012), methods to handle logbook data and define métiers fall into two categories: (1) supervised (that is, performed by human experts with a priori knowledge about the investigated fishery) analyses of basic statistics (of catch composition, e.g., main species by weight) and (2) the unsupervised application of multivariate analyses of catch composition by trip or fishing operation to group similar catch profiles into métiers. Although the latter approach is more “objective” than the former, the clustering step also includes elements of subjective choice (Deporte et al., 2012).

In this paper, we present a new approach to logbook data processing based on the use of Self-Organizing Maps (SOMs). SOMs are a particular type of unsupervised Artificial Neural Network (ANN) that produce a synthetic representation of the information in the input data by approximating their probability density function (Quetglas et al., 2006). The representation produced by SOM is typically a two-dimensional pattern that preserves as much of the existing topology as possible (Chon et al., 1996). Although similar to classical multivariate approaches, SOM outputs are usually more comprehensible, and the method’s ability to extract information from complex datasets outperforms conventional approaches previously used in ecology for patterning purposes (Quetglas et al., 2006). SOM was thus explored because it offers a series of advantages with respect to standard multivariate analyses (Giraudel and Lek, 2001, Park et al., 2004, Russo et al., 2010, Russo et al., 2014c), overcoming issues regarding the subjectivity of results and the ability to classify new logbook observations. Among others, the potential advantages of the application of SOM in analyzing logbook data include the following: (1) the acknowledged ability to represent non-linear patterns without the assumptions characterizing the commonest scaling techniques such as principal component analysis (Giraudel and Lek, 2001, Hutchinson and Mitchell, 1973); (2) the possibility of obtaining a virtually infinite series of effective graphical outputs; and (3) the relatively lower sensitivity of SOM to outlier (biased) records, such as human error or incorrect data entry type errors (Latif and Mercier, 2010).

The aims of this paper are: (1) to present a methodology to process a large logbook dataset, allowing both the extraction and the prediction of relevant patterns; (2) to report the first analysis of logbook data (including a preliminary check for accuracy and precision) for the Italian fleet, the largest in the whole Mediterranean in terms of size and the spatial range of fishing activity; and (3) to characterize the composition of landings per métier for the Italian fleet. Given that only fishing vessels at least 10 m in length overall (LOA) are required to keep and transmit a fishing logbook (Commission, 2009) and that only fishing vessels at least 12 m in LOA are mandatorily equipped with VMS (Commission, 2011a), the object of this study is the subset of the Italian fishing fleet with LOA  12 m.

Section snippets

Logbook data: quality checks and pre-processing

A native logbook dataset, provided by the Italian Ministry for Agricultural, Food and Forestry Policies (MAFFP) and containing 16727 records corresponding to as many fishing trips of Italian vessels, was used in this study. This dataset covers part of the activity of 433 fishing vessels (11% of the whole fishing fleet with LOA greater than 12 m) during the years 2011–2012, in all seven Geographic Sub Areas (GSA) of which the Italian Seas are parts (Table 1

Quality checks and anosim

The 61 species/taxa retained after the first pre-processing steps are listed in Supplementary Table 1. They comprise bony and cartilaginous fishes, mollusks and crustaceans. The boxplot in Fig. 3a shows that the differences between logbooks and observer records are generally between −5 and 5 kg, with the exception of some large pelagic species (i.e., Thunnus alalunga and Xiphias gladius) and some species generally fished in large quantities (i.e., Chamelea gallina and Sardina pilchardus). These

Discussion

The identification and description of the assemblages exploited by fisheries are considered a prerequisite for any effective management action in multispecies and multi-gear fisheries (Garces et al., 2006) such as those in the Mediterranean. This study reports, to our knowledge, the first characterization of landings profiles for the main gears operating in Italian seas. Previous studies on the Italian fisheries addressed artisanal fishery in restricted areas (Battaglia et al., 2010, Colloca et

Acknowledgements

The authors would to acknowledge the Italian Ministry for Agricultural, Food and Forestry Policies, which provided the data for this study, and Dr. Elisabetta Betulla Morello (CNR ISMAR − National Research Council of Italy, Marine Sciences Institute) who critically revised and edited the manuscript. This research was supported within the activities for the application of the Italian National Program for the Data Collection in the Fisheries Sector 2014–2016. The authors also would to acknowledge

References (64)

  • L.R. Garces et al.

    Spatial structure of demersal fish assemblages in South and Southeast Asia and implications for fisheries management

    Fish. Res.

    (2006)
  • J.L. Giraudel et al.

    A comparison of self-organizing map algorithm and some conventional statistical methods for ecological community ordination

    Ecol. Model.

    (2001)
  • Y.-S. Park et al.

    Applications of artificial neural networks for patterning and predicting aquatic insect species richness in running waters

    Ecol. Model.

    (2003)
  • Y.-S. Park et al.

    Hierarchical community classification and assessment of aquatic ecosystems using artificial neural networks

    Sci. Total Environ.

    (2004)
  • Y.-S. Park et al.

    Application of a self-organizing map to select representative species in multivariate analysis: a case study determining diatom distribution patterns across France

    Ecol. Inform.

    (2006)
  • P.J. Rousseeuw

    Silhouettes: a graphical aid to the interpretation and validation of cluster analysis

    J. Comput. Appl. Math.

    (1987)
  • T. Russo et al.

    New insights in interpolating fishing tracks from VMS data for different métiers

    Fish. Res.

    (2011)
  • T. Russo et al.

    When behaviour reveals activity: assigning fishing effort to métiers based on VMS data using artificial neural networks

    Fish. Res.

    (2011)
  • T. Russo et al.

    Application of the Self-Organizing Map to the study of skeletal anomalies in aquaculture: the case of dusky grouper (Epinephelus marginatus Lowe, 1834) juveniles reared under different rearing conditions

    Aquaculture

    (2011)
  • T. Russo et al.

    Spatial indicators of fishing pressure: preliminary analyses and possible developments

    Ecol. Indic.

    (2013)
  • T. Russo et al.

    Modelling the strategy of mid-water trawlers targeting small pelagic fish in the Adriatic Sea and its drivers

    Ecol. Model.

    (2015)
  • D.B. Sampson

    The accuracy of self-reported fisheries data: Oregon trawl logbook fishing locations and retained catches

    Fish. Res.

    (2011)
  • S. Shanmuganathan et al.

    Self-organising map methods in integrated modelling of environmental and economic systems

    Environ. Model. Softw

    (2006)
  • D. Tingley et al.

    Estimating capacity utilisation in multi-purpose multi-métier fisheries

    Fish. Res.

    (2003)
  • E. Tzanatos et al.

    Catch length analysis, relation to minimum landing sizes and management implications from a Mediterranean small-scale fishery (Patraikos Gulf Greece)

    Fish. Res.

    (2008)
  • Acom, I.C.M., 2012. ICES WGECO REPORT 2012 Ecosystem Effects of Fishing Activities Report of the Working Group on the...
  • J.A. Bertrand et al.

    The general specifications of the MEDITS surveys

    Sci. Mar.

    (2002)
  • F. Biagi et al.

    Analysis of demersal assemblages off the Tuscany and Latium coasts (North-western Mediterranean)

    Sci. Mar.

    (2002)
  • M. Bottin et al.

    diatSOM: a R-package for diatom biotypology using self-organizing maps

    Diatom Res.

    (2014)
  • J.C. Bray et al.

    An ordination of upland forest communities of southern Wisconsin

    Ecol. Monogr.

    (1957)
  • Commission, E., 2009. Council regulation (EC) no. 1224/2009 of 20 november 2009 establishing a community control system...
  • Commission, E., 2011. Commission Implementing Regulation (EU) No. 404/2011 of 8 April 2011 laying down detailed rules...
  • Cited by (36)

    • Inferring ecosystem impacts of a small-scale snapper fishery through citizen science data, productivity and susceptibility analysis, and ecosystem modelling

      2022, Fisheries Research
      Citation Excerpt :

      The lack of data in Mexican fisheries is a common feature (Salas et al., 2007; Arreguín-Sánchez and Arcos-Huitrón, 2011), preventing stock assessments needed to implement management strategies. An alternative data source is the logbooks filled out by the fishing cooperatives or fishing permit owners (e.g., Russo et al., 2016; Mendoza-Carranza et al., 2018). In the present study, citizen science provides valuable data on six small-scale boats targeting snappers over 13 months.

    • Comparing life-history traits in two contiguous stocks of the deep-water rose shrimp Parapenaeus longirostris (H. Lucas, 1846) (Crustacea: Decapoda) in the Southern Tyrrhenian Sea (Central Mediterranean Sea)

      2022, Fisheries Research
      Citation Excerpt :

      Indeed, these factors (biotic and abiotic) can influence P. longirostris abundance and distribution, interacting with each other and causing synergic or contrasting effects with an impact on population structure (e.g., recruits/juvenile). Nevertheless, the fishing activity’s role is of primary importance (Russo et al., 2019; Russo et al., 2016a, 2016b), influencing the P. longirostris stock features (Ligas et al., 2011). Therefore, it is essential to improve conservation policy and fisheries management, shaped them in relation to fisheries heterogeneity and features of several communities’, which, in many species such as P. longirostris, they can change also inside the same population on a very small geographic scale.

    • Longline métiers and associated economic profiles in eastern Mediterranean fisheries: The case study of Kalymnos Island (South Aegean Sea)

      2020, Ocean and Coastal Management
      Citation Excerpt :

      Considering each métier separately, LLS2 that uses longlines to catch D. dentex was practiced as in other coastal areas of Greece (Tzanatos et al., 2006), while LLS3, LLS4, LLS5 and LLS6 were described from several Spanish and Italian coastal areas (Colloca et al., 2003; García-Rodríguez et al., 2006; Forcada et al., 2010). LLD that uses drifting longlines to catch X. gladius was also reported from Italy (Colloca et al., 2004; Russo et al., 2016). In the present study, the métier SLHP had exactly the same species catch and seasonality as in the Alicante Gulf (Forcada et al., 2010), whereas LHP had the same species composition but used different gears compared to other Mediterranean areas (Tzanatos et al., 2006; Forcada et al., 2010; Maynou et al., 2011).

    • Strategies and trends of bottom trawl fisheries in the Mediterranean Sea

      2020, Marine Policy
      Citation Excerpt :

      In particular, VMS data are used to characterize the yearly fishing effort pattern of each vessel on yearly basis (Fig. 1). Then Self-Organizing Maps (SOMs - [25,29,30]) are used to identify the main emerging patterns (i.e. strategies). After this, two complementary modelling approaches were applied in order to investigate different aspects of the decision-making process regulating the choice between different strategies and the relative importance of these strategies across several years.

    View all citing articles on Scopus
    View full text