Vision and predictive analytics for monitoring sustainability of exploited fishing populations

Abstract
Global demand for seafood largely increased worldwide, and this high demand led to an unprecedented impact on the aquatic ecosystems, lowering ocean biomass content and the proportion of fish stocks that are within biologically sustainable levels. In the European Union (EU), fisheries are managed under the Common Fisheries Policy (CFP). According to a recent assessment, 69% of the 397 commercially exploited fish and shellfish stocks were overfishing. Sustainable management of fisheries imply the maintenance of populations of harvested stocks above levels that can produce the maximum sustainable yield (MSY). The Azores fisheries are considered as being artisanal and small-scale in nature, targeting multiple species. Despite all regional fisheries management measures implemented, signs of intensive exploitation of several fish species or fishing grounds are of increasing concern. A recent study show, that among the 138 species landed in the region between 2009-2019, twenty-two were selected as priority stocks according to the FAO and ICES criteria, showing a decreasing trend in their abundances and only four stocks are currently assessed using data limited approaches. As in all EU member states, most of the fisheries and biological data are collected under the European Data Collection Framework (DCF). However, despite the efforts made, for most species, information is usually deficient. Uncertainty inherent in data-limited stock assessments could compromise the ability to inform management. However, the most complex assessment methods require considerable amounts of data which are difficult and expensive to obtain. Increasing research has been dedicated to assessment methods for data-limited fisheries, and simple fishery indicators based on length composition provides useful fish stocks statuses indicators. Among the length-based methods the ICES identified the length-based indicator (LBI) and length-based spawning potential ratio (LBSPR) as the most appropriate to achieve reliable assessments. Collecting fish size samples manually in fishing vessels and in fish auctions is however, challenging and error prone. Statistical sampling assumptions are frequently difficult to be accomplished affecting the representativeness of the samples and only a small number of landed species and fishing harbors are usually covered. New existing solutions based on emerging vision technologies have shown to improve the quality of size sampling in a relatively inexpensive manner. This is the case of the 3D scanning of fish boxes landed, allowing the measuring of the fishes on the scanned images. Several of these systems were recently installed in several fish auctions and proved to overcome most of the well-known problems of the traditional manual size sampling methods.

The primary objective of this PhD is to leverage the state of art in machine learning and computer vision, and develop methods capable of allowing real-time monitoring of exploited fishing populations, and enable a more a fine-grained policy and stock management. Specifically, this PhD project aims to use the existing extensive image database of fish boxes already acquired in the Azores auctions and 1) develop machine learning algorithms to automatically obtain fish lengths and other indicators, 2) to develop algorithms that automatically classify the fishing “metiers” based on the vessel’s species composition landed, 3) to develop algorithms that automatically run several length-based methods and generate robust stock statuses indicators of several species including predictive functionalities using the time series of those indicators. Taking into consideration the state-of-the-art methods within this scope, we should point out that the successful completion of this PhD work will represent a breakthrough and an unprecedent advance for the application of length-based stock assessment methods, and a significant contribution to improve the management and sustainability of fisheries.

Candidate’s Profile
Preference will be given to candidates with advanced science degrees (Msc. Degree or equivalent), with a strong background on computer sciences or advanced computing and programming languages, advanced statistics, machine learning or Artificial Intelligence analyses. Scientific publications or previous works on the PhD topic are a plus.

Name(s) of supervisor(s)
Gui M. Menezes (Institute of Marine Sciencces – Okeanos, Universidade dos Açores)
Nuno Moniz (INESC TEC, Universidade do Porto, University of Notre Dame)

Identification of PhD Program

PhD in Computer Science, University of Porto, https://sigarra.up.pt/fcup/en/CUR_GERAL.CUR_VIEW?pv_ano_lectivo=2022&pv_curso_id=1021

Notice of the call (english version)

Notice of the call ( portuguese version)

We are no longer accepting applications for this scholarship. Thank you.