Machine learning for characterizing tropical tuna aggregations under Drifting Fish Aggregating Devices (DFADs) from commercial echosounder buoys data
Introduction
Many marine species are known to naturally aggregate under floating objects. Although still poorly understood, this behaviour is widely exploited by fishermen, who deploy man-made floating objects (hereafter referred to as Fish Aggregating Devices or FADs) worldwide to improve their catches (Kakuma, 2001; Fonteneau et al., 2013; Albert et al., 2014). The use of drifting FADs (DFADs) in tropical tuna fisheries was first introduced in the late 1980s in the Eastern Pacific Ocean by the US purse seine fleet (Lennert-Cody and Hall, 2001) and was later extended to all oceans and fleets from the 1990s. The instrumentation of DFADs with GPS beacons and echosounder buoys, in the mid and late 2000s respectively (Lopez et al., 2014), led to major changes in fishing strategies and behaviour of purse-seine fleets (Torres-Irineo et al., 2014). By providing skippers with almost real-time remote information on the precise location of DFADs, and on the potential presence and size of the tuna aggregation, echosounder buoys reduced the search time between two successful DFAD sets (Lopez et al., 2014). As a result, modern DFADs have significantly increase fishing efficiency (Fonteneau et al., 2013). Consequently, their use has increased considerably in the past few decades. Recent studies indicate that in less than a decade, the number of DFADs deployed in the Atlantic and Indian Oceans have increased at least fourfold (Fonteneau et al., 2015; Maufroy et al., 2017). It is estimated that over half of the annual tropical tuna purse seine catches originate from fishing sets on DFADs (Dagorn et al., 2013; Fonteneau et al., 2013).
Aside from being highly efficient fishing tools, the large number and vast spatial distribution of DFADs, coupled with their constantly evolving technology (Lopez et al., 2014), mean that they can also potentially provide unprecedented scientific insights into pelagic communities (Moreno et al., 2016; Brehmer et al., 2018). The echosounder buoys attached to DFADs regularly produce and transmit biomass estimation data. This dataset potentially holds a major opportunity for improving the management of tropical tuna stocks through the development of fishery-independent abundance indices (Capello et al., 2016; Santiago et al., 2016). Currently, the main abundance indicators used in stock assessment for tropical tunas are derived through the standardization of Catch per Unit of Effort (CPUE) from commercial data (Fonteneau et al., 1998; Maunder et al., 2006). However, owing to the constant technological advances occurring in the purse seine fishery, it is extremely difficult to accurately standardize the CPUE time-series (Fonteneau et al., 1999). Traditionally, search time was used to quantify nominal fishing effort in this fishery, however, owing to its non-random nature, the DFAD-based fishery has made this metric inconsistent over time, thus introducing major biases and uncertainties in the relationship between tuna catches and abundance (Fonteneau et al., 1999; Gaertner et al., 2015).
The need for the consideration of non-traditional data sources to provide alternate abundance indices for stock assessment of tunas is becoming increasingly apparent. In this regard, the large amount of acoustic data autonomously collected by commercial echosounder buoys on DFADs is of undeniable value. However, the direct exploitation of these data remains challenging. The biomass estimate that a buoy produces is limited by the reliability and variability of the information provided, which depends on the hardware and software characteristics of the buoy, and varies between manufacturers (Lopez et al., 2014; Santiago et al., 2016). As a result, the data provided by echosounder buoys are heterogeneous in types and formats, with limited studies having provided an assessment of their accuracy for use in scientific investigations. (Lopez et al., 2016; Baidai et al., 2017; Orue et al., 2019a).
In recent years, fisheries scientists have shown a growing interest in machine learning methods for the processing of both passive acoustic data (Roch et al., 2008; Zaugg et al., 2010; Noda et al., 2016; Malfante et al., 2018) and acoustic data collected by scientific echosounders (Fernandes, 2009; Robotham et al., 2010; Bosch et al., 2013). Despite this trend, very few studies have been conducted on the implementation of automated classification methods for analysing the extensive datasets collected by commercial vessels (Uranga et al., 2017).
This paper presents a new methodology, based on machine learning, for processing the echosounder data collected from one of the main models of echosounder buoy used to equip DFADs worldwide (Moreno et al., 2019).
Section snippets
Echosounder buoy data
We used data from the Marine Instruments M3I buoy (https://www.marineinstruments.es), collected on DFADs deployed by the French purse seine vessels operating in the western Indian and eastern Atlantic oceans from 2013 to 2018. The dataset consists of more than 60 million data points collected by approximately 35,000 M3I buoys. This model of buoy includes a solar powered echosounder operating at a frequency of 50 kHz, with a power output of 500 W, a beam angle of 36°, and a sampling frequency of
Pre-processing of sampled depth layers
The clustering analysis carried out on the 3 m depth layers led in both oceans to the formation of six groups with similar layer compositions between the two oceans (Fig. 3). In each ocean, the comparison of the acoustic scores between the identified groups showed highly significant differences (p-value at Kruskal-Wallis test < 0.001 for both Indian and Atlantic Oceans). Scores declined strongly with depth (Fig. 4). The deepest group of layers (which also aggregated the greatest number of
Discussion
This study describes a new methodology for processing data collected by a commercial echosounder buoy commonly used in the DFAD purse seine fishery. The approach utilizes the acoustic scores (reflective of abundance) recorded at different depths and times of the day and combines data pre-processing procedures and machine learning algorithms to classify tropical tuna aggregations under DFADs. Although several models of echosounder buoys process data internally and generate abundance indices for
Conclusion
The methodology developed in this study provides an indicator of presence/absence of tuna schools at DFADs in both the Atlantic and Indian Oceans, from simplified acoustic data collected by one of the echosounder buoy models used in the tuna purse seine fishery. This approach has the potential to summarize and analyse a large amount of acoustic data, with an efficiency that obviously depends on the nature and quality of the data provided. Nevertheless, the rapid and continuous evolution in
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
CRediT authorship contribution statement
Y. Baidai: Conceptualization, Methodology, Formal analysis, Validation, Visualization, Investigation, Writing - original draft. L. Dagorn: Conceptualization, Resources, Validation, Writing - review & editing, Supervision, Funding acquisition. M.J. Amande: Conceptualization, Supervision. D. Gaertner: Conceptualization, Supervision, Writing - review & editing. M. Capello: Conceptualization, Supervision, Writing - review & editing, Funding acquisition.
Acknowledgements
This project was co-funded by “Observatoire des Ecosystèmes Pélagiques Tropicaux exploités” (Ob7) from IRD/MARBEC and by the ANR project BLUEMED (ANR-14-ACHN-0002). The authors are grateful to ORTHONGEL and its contracting parties (CFTO, SAPMER, SAUPIQUET) for providing the echosounder buoys data. The authors also thank all the skippers who gave their time to share their experience and knowledge on the echosounder buoys. The authors sincerely thank the contribution of the staff of the Ob7 for
References (60)
- et al.
Ultrasonic attenuation and propagation speed in mammalian tissues as a function of temperature
Ultrasound Med. Biol.
(1979) - et al.
Support vector machine under uncertainty: an application for hydroacoustic classification of fish-schools in Chile
Expert Syst. Appl.
(2013) - et al.
Evolution and current state of the technology of echo-sounder buoys used by Spanish tropical tuna purse seiners in the Atlantic, Indian and Pacific Oceans
Fish. Res.
(2014) - et al.
Fish aggregating devices (FADs) as scientific platforms
Fish. Res.
(2016) - et al.
Acoustic identification of small pelagic fish species in Chile using support vector machines and neural networks
Fish. Res.
(2010) - et al.
Theoretical estimation of the temperature dependence of backscattered ultrasonic power for noninvasive thermometry
Ultrasound Med. Biol.
(1994) - et al.
Changes in fishing power and fishing strategies driven by new technologies: the case of tropical tuna purse seiners in the eastern Atlantic Ocean
Fish. Res.
(2014) - et al.
Spatiotemporal distribution of fish schools around drifting fish aggregating devices
Fish. Res.
(2016) - et al.
Real-time acoustic classification of sperm whale clicks and shipping impulses from deep-sea observatories
Appl. Acoust.
(2010) - et al.
The contribution of nearshore fish aggregating devices (FADs) to food security and livelihoods in Solomon Islands
PLoS One
(2014)
Towards the Derivation of Fisheries-independent Abundance Indices for Tropical Tunas: Progress in the Echosounders Buoys Data Analysis. IOTC-2017-WPTT19-22
Effects of temperature on global patterns of tuna and billfish richness
Mar. Ecol. Prog. Ser.
Target strength of skipjack tuna (Katsuwonus pelamis) associated with fish aggregating devices (FADs)
ICES J. Mar. Sci.
Towards an autonomous pelagic observatory: experiences from monitoring fish communities around drifting FADs
Thalass. An Int. J. Mar. Sci.
Random forests
Mach. Learn.
Population assessment of tropical tuna based on their associative behavior around floating objects
Sci. Rep.
A general theory on fish aggregation to floating objects: an alternative to the meeting point hypothesis
Rev. Fish Biol. Fish.
Weighted kappa: nominal scale agreement provision for scaled disagreement or partial credit
Psychol. Bull.
Behavior of yellowfin (Thunnus albacares) and bigeye (T. obesus) tuna in a network of fish aggregating devices (FADs)
Mar. Biol.
Satellite-linked acoustic receivers to observe behavior of fish in remote areas
Aquat. Living Resour.
Is it good or bad to fish with FADs? What are the real impacts of the use of drifting FADs on pelagic marine ecosystems?
Fish Fish.
Fish communities associated with FADs
Sci. Mar.
Classification trees for species identification of fish-school echotraces
ICES J. Mar. Sci.
Vingt-cinq ans d’évaluation des ressources thonières de l’Atlantique : quelques réflexions methodologiques
Collect. Vol. Sci. Pap. ICCAT
An overview of problems in the CPUE-abundance relationship for the tropical purse seine fisheries
Collect. Vol. Sci. Pap. ICCAT
Global spatio-temporal patterns in tropical tuna purse seine fisheries on drifting fish aggregating devices (DFADs): taking a historical perspective to inform current challenges
Aquat. Living Resour.
Managing tropical tuna purse seine fisheries through limiting the number of drifting fish aggregating devices in the Atlantic: food for thought
Collect. Vol. Sci. Pap. ICCAT
Behaviour and vulnerability of target and non-target species at drifting fish aggregating devices (FADs) in the tropical tuna purse seine fishery determined by acoustic telemetry
Can. J. Fish. Aquat. Sci
Catch, effort, and ecosystem impacts of fad-fishing (CECOFAD)
Collect. Vol. Sci. Pap. ICCAT
Collecting Data on Board French Tropical Tuna Purse Seiners With Common Observers: Results of ORTHONGEL’s Voluntary Observer Program OCUP in the Atlantic Ocean (2013-2017). IOTC-2017-WPDCS-22
Cited by (23)
Exploring representative samples for modeling of wave buoy motion behavior
2024, Ocean EngineeringA fishing route optimization decision support system: The case of the tuna purse seiner
2024, European Journal of Operational ResearchLessons learnt from the first large-scale biodegradable FAD research experiment to mitigate drifting FADs impacts on the ecosystem
2023, Marine PolicyCitation Excerpt :Tuna presence and absence were estimated using a supervised classification of acoustic data from FAD buoy echo-sounder biomass data [28], but only employing one buoy model (i.e., M3i) for which the supervised classification was available. The data filtering process followed the protocols defined in [28,29]. Echo-sounder buoy data was also used to estimate the tuna biomass aggregation around FADs from acoustic energy values [30].
Transfer learning for modeling and prediction of marine buoy motion characteristics
2022, Ocean EngineeringExperimental and numerical studies on working parameter selections of a piezoelectric-painted-based ocean energy harvester attached to fish aggregating devices
2022, Energy for Sustainable Development