Content uploaded by Erik Sulanke
Author content
All content in this area was uploaded by Erik Sulanke on Sep 30, 2024
Content may be subject to copyright.
Amending the European shing eet segmentation based on machine
learning and multivariate statistics
E. Sulanke
a,*,1
, V. Rubel
b,1
, J. Berkenhagen
a
, M. Bernreuther
a
, T. Stoeck
b
, S. Simons
a
a
Thünen Institute of Sea Fisheries, Bremerhaven, Germany
b
RPTU Rheinland-Pf¨
alzische Universit¨
at Kaiserslautern, Landau, Germany
ARTICLE INFO
Keywords:
Classication
Data collection
Fisheries
Fisheries management
Fleet economic data
Random forest
ABSTRACT
Considering the critical issue of overexploited stocks due to overshing, the EU’s Data Collection Framework
(DCF) was established. Within the DCF, member states collect and analyze data relevant to sustainable sheries
management. To evaluate the status of sheries, it is necessary to categorize shing eets into eet segments.
However, the current DCF segmentation is primarily based on technical vessel parameters, such as vessel length
and predominant shing gear, which often do not accurately represent the shing activities of the vessels. To
address this, we developed an alternative eet segmentation approach that provides a more realistic overview of
shing activities. This approach utilizes multivariate statistics and is coupled with machine learning techniques
for automatization. Applying this approach to two decades of German sheries data resulted in a data set with
fewer segments compared to the DCF approach, which represented the actual shing strategies more closely. The
comparison of biological stock health indicators calculated for both the current and the novel segmentation
schemes revealed that the current scheme often misses signs of segments relying on overexploited stocks. The
machine learning technique applied showed high classication accuracy, with misclassications being rare and
only occurring in segments with overlapping catch composition. Since machine learning enables almost perfect
allocation to the revised segments, we expect a successful implementation of this protocol for future eet seg-
mentation. This approach is highly suitable for data collection and analysis procedures and can serve as a
standard tool. Therefore, this novel approach can contribute to the improvement of shing eet analyses and
policy advice for better sheries management.
1. Introduction
Fisheries are facing a period of global upheaval. Decades of indus-
trialized shing have extensively exploited the world’s oceans, leading
many sh stocks to critical depletion levels or placing them in slow re-
covery phases (FAO, 2024). Moreover, the impacts of climate change on
marine ecosystems are becoming increasingly evident, bearing consid-
erable challenges for sheries management (Miles, 2011). Changes such
as rising sea temperatures, ocean acidication, and shifts in ocean cir-
culation patterns, which are disrupting sh populations, altering habi-
tats, and unsettling the balance of ecosystems (Brierley and Kingsford,
2009). Many commercially important sh species are migrating in
response to new temperature regimes, causing noticeable distribution
shifts on a global scale (Cheung et al., 2010).
Effective sheries management plays a crucial role in addressing
these emerging challenges by implementing adaptive measures in
response to changing environmental conditions (Burden and Fujita,
2019). Key strategies include setting sustainable catch limits, modifying
shing gear, and adopting ecosystem-based approaches. By integrating
both ecological and socio-economic considerations, sheries manage-
ment can develop balanced strategies that support the recovery and
sustainability of sh stocks, protect marine environments, and safeguard
the livelihoods of shing communities (Frost and Andersen, 2006;
Gaines et al., 2018; Gourlie, 2017). To make informed decisions, it is
vital to assess the impact of these management strategies, ensuring they
effectively balance the diverse needs of society, economies, and eco-
systems (Thorpe et al., 2016). Impact assessments provide a systematic
method for evaluating the potential outcomes of management options
on sh stock health, the economic viability of the shing industry, and
the cost-effectiveness of management measures (Propst and Gavrilis,
* Correspondence to: Thünen Institute of Sea Fisheries, Herwigstraße 31, Bremerhaven 27572, Germany
E-mail address: erik.sulanke@thuenen.de (E. Sulanke).
1
Both authors contributed equally to the work
Contents lists available at ScienceDirect
Fisheries Research
journal homepage: www.elsevier.com/locate/fishres
https://doi.org/10.1016/j.shres.2024.107190
Received 12 June 2024; Received in revised form 23 September 2024; Accepted 23 September 2024
Fisheries Research 281 (2025) 107190
0165-7836/© 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ).
1987).
A crucial element in these assessments is the collection of compre-
hensive and detailed sheries data. Within the European Union, the
Fisheries Data Collection Framework (DCF) provides a robust system for
gathering, managing, and analyzing biological, environmental, social,
technical, and economic data related to sheries and aquaculture ac-
tivities in EU waters (D¨
orner et al., 2018). This framework mandates the
reporting of essential information by sheries stakeholders, including
data on catches, landings, effort, and eet structure. The DCF empha-
sizes the importance of data quality, promotes integrated data man-
agement, and supports scientic evaluations of sh stocks and the
economic performance of shing eets. This data-driven approach is
central to the success of the Common Fisheries Policy (CFP), ensuring
that European sheries resources are managed sustainably through
informed decision-making (Barkai et al., 2010; Simmonds et al., 2011).
The aggregated groups for which e.g. eet economic data are to be
provided are called eet segments. In the data assessment, the DCF eets
are partitioned based on dominant shing gear type (see Table 1), vessel
size categories (see Table 2), and area of operation (see Annex Table A1).
However, the current segmentation lacks components related to stock-
or catch proles (European Commission, 2021). There are many states
and organizations worldwide collecting sheries data and aggregating
vessels into eet segments (although the nomenclature differs), and
most of them apply technical characteristics of vessels for this purpose,
e.g., the Food and Agricultural Organisation (FAO). Adding
shery-based or stock-based aspects to the segmentation schemes is less
common. Nonetheless, this approach is more frequently adopted at na-
tional or regional levels, such as in the US North Pacic sheries
(Witherell et al., 2012), the United Kingdom (Seash, 2023), or Norway
(Fiskeridirektoratet, 2021).
The current eet segmentation under the EU sheries Data Collec-
tion Framework (DCF) is based on technical characteristics and thus
lacks a closer link to sheries or even sh stocks. This shortfall hinders
guidance on sustainability and economic efciency across individual
sheries (STECF, 2022).
While it is straightforward, it lacks the detail needed to accurately
represent the diversity of sheries in European waters (Guyader et al.,
2013). For instance, Pawson et al. (2008)and García-Fl´
orez et al. (2014)
both emphasize the need for clear denitions and segmentation of rec-
reational and artisanal sheries, respectively. Lloret et al. (2018) un-
derscores the evolving nature of small-scale coastal sheries, inuenced
by factors such as the expansion of recreational shing and the adoption
of more mechanized techniques. These studies collectively emphasize
the need for a more nuanced understanding of the European shing
eet, recognizing the unique characteristics of each segment.
In addition, the DCF segmentation scheme groups vessels with
similar technical features, yet these vessels may engage in distinctly
different sheries with differing catch compositions, shing practices,
and cost structures. The German demersal trawl eets operating in the
Baltic Sea and the North Sea are good examples of this. These vessels are
technically similar and thus belong to the same DCF segment, but they
target different species and exhibit variations in shing practices. Not
only do they sh on entirely different shing grounds, but they also
considerably differ in the time necessary to reach these shing grounds.
The time a vessel spends reaching shing locations is commonly referred
to as ‘steaming time’. Conversely, vessels performing the same shery
may have remarkable differences in technical characteristics, such as
vessel length, and consequently are assigned to different segments. For
instance, the German brown shrimp shery, which includes beam
trawlers ranging from 10 to 24 m, is segmented into different DCF
segments: TBBVL1012, TBBVL1218, TBBVL1824 (Goti-Aralucea et al.,
2021).
Thus, the current generalized eet segmentation employed by DCF
does not provide the detailed information required to fully comprehend
the diverse shing practices throughout European waters. This
mismatch between eet segmentation and actual shing practices can
lead to suboptimal sheries management (D´
epalle et al., 2020; Poos
et al., 2010), especially as many sheries management measures,
including nancial support, regulations, and quota allocations, are often
linked to DCF segments (Branch et al., 2006; Moura et al., 2016; Ulrich
et al., 2012). Consequently, eet segmentation plays a pivotal role in
deciding which management measures are applied to specic vessels,
impacting the effectiveness of these initiatives in achieving sustainable
shing practices.
Moreover, the International Council for the Exploration of the Sea
(ICES) as key advisory body for the European Union, denes sheries as
" […] a group of vessel voyages targeting the same (assemblage of)
species and/or stocks, using similar gear, during the same period of the
year and within the same area (e.g., the Dutch atsh-directed beam
trawl shery in the North Sea) […] " (ICES, 2003). This denition sug-
gests a more detailed eet segmentation than that used by the DCF,
highlighting a major granularity gap between ICES and DCF sheries
data systems. Because of this mismatch, ICES encounters difculties to
fully incorporate economic eet structures and dynamics into its pre-
dominantly biological advice (ICES, 2024; Smit, 1996). Since shing
measures will affect shers’ behavior and subsequently affect eet dy-
namics and sh stock development, it is crucial to consider detailed eet
data when evaluating sheries management measures (Li et al., 2021;
Branch et al., 2006; Salas and Gaertner, 2004).
To improve accuracy in sheries management, an innovative eet
segmentation approach based on multivariate statistics has been
developed that more closely aligns with the actual activities of eets.
Multivariate statistics have been commonly used in sheries research to
identify target sheries since the general availability of high-power
computers enabled researchers to analyze the necessary, large datasets
(Lewy, 1994; Rogers and Pikitch, 1992). Often originating in community
Table 1
DCF gear codes, corresponding gears, classication, and activity category of all
gear classes relevant to the German eet.
gear
code
activity
category
Description Category
DFN Passive
gears
Drift and/or xed netters Nets
DRB Mobile gears Dredgers Dredges
DTS Mobile gears Demersal trawlers and/or demersal
seiners
Trawls
FPO Passive
gears
Vessels using pots and/or traps Traps
FPO Passive
gears
Uncovered xed pound nets Traps
PGO Passive
gears
Vessels using other passive gears Other
TBB Mobile gears Beam trawlers Trawls
TM Mobile gears Pelagic trawlers Trawls
PG Passive
gears
Vessels under 12 m length using
miscellaneous passive gears
Nets, hooks
and traps
Table 2
DCF vessel length classes as applied in all European eets except for Mediter-
ranean eets. Depicted are the ofcial DCF length class codes, the corresponding
vessel length range and the prevalence of the length class in the German shing
eet. The prevalence is given in minimum and maximum percentage of the
segment in relation to the entire eet with regard to the number of vessels and
the gross tonnage for the years 2002–2022.
Vessel length class Vessel length [m] Share of German shing eet 2002–2022
(%)
number of vessels gross tonnage
VL0010 0 – 10 64.17 – 69.30 2.03 – 3.36
VL1012 10 – 12 5.87 – 7.33 1.17 – 1.99
VL1218 12 – 18 12.42 – 17.02 6.46 – 11.31
VL1824 18 – 24 6.64 – 9.03 8.51 – 13.43
VL2440 24 – 40 2.03 – 3.93 9.11 – 17–68
VL40XX >40 0.87 – 2.02 53.69 – 72.69
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
2
ecology (Faith et al., 1987), they are excellent tools for such analyses.
The analysis of a species community, specically of species diversity and
abundance, resembles the analysis of a sheries catch prole in nearly
every aspect. Yet, to our best knowledge, the multivariate approaches
used in shing eet analysis were primarily applied to specic sheries
or eets (Holley and Marchal, 2004; Jim´
enez et al., 2004; Lewy, 1994;
Murawski et al., 1983; Natale et al., 2015; Pelletier and Ferraris, 2000;
Pilar-Fonseca et al., 2009) and were not designed with regard to trans-
ferability and mechanistic consistency. In addition, they were not paired
with contemporary techniques of machine learning, which can boost the
applicability and effectiveness of such approaches.
This method enhances the consistency of catch data and cost struc-
tures. To facilitate the segmentation procedure, a machine learning-
based automation process using a random forest algorithm has been
employed. Random forest algorithms have been increasingly applied in
marine and sheries research to a variety of research questions. Early,
their general applicability in ecology was highlighted (Cutler et al.,
2007). Random forest algorithms have since been used in marine ecol-
ogy to forecast species abundances (Baba and Matsuishi, 2015), model
the distribution of rare species (Garcia et al., 2022), analyze population
structures (Berio et al., 2022; Zhang et al., 2016) and classify species
based on echosounder data (Rousseau et al., 2022). Current efforts are
underway to even incorporate new applications of random forest algo-
rithms for autonomous biomonitoring into legislation (Cordier et al.,
2018). In sheries research, the variety of applications is similarly
widespread, as random forest algorithms were applied for analyses of
spatial use patterns (Behivoke et al., 2021; Joo et al., 2013), harvest
control rules (Van Poorten et al., 2013), and bycatch misreporting
(Lennert-Cody and Berk, 2007). Even on the economic side of sheries
research, business risk assessment can be conducted by applying random
forest algorithms (Sethi et al., 2012). As highlighted by Cutler et al.
(2007), these algorithms inherit several crucial advantages over other
classication methods, most notably achieving very high classication
accuracies. Further advantages include the ability to account for in-
teractions of predictor variables and the robust identication of variable
importance (Cutler et al., 2007). Also, the random forest algorithm is
capable of performing accurate classication as well as regression tasks,
which makes it suitable for a variety of disciplines and problems
(Breiman, 2001).
Equipped with a training dataset, this method can efciently
segment and analyze extensive sheries data spanning multiple decades.
This advancement not only allows the user-friendly creation of eet
segments for management and advisory purposes but also reduces the
workload for individual data collectors within the DCF.
2. Material and methods
The is divided in the three major sections, which are
1) the method description and statistical background of the alter-
native eet segmentation approach based on multivariate statistics;
2) the description of the random forest algorithm which was applied
for automatically assigning alternative segments; and
3) the calculation of balance indicators for the evaluation of the
alternative segments.
They are structured as follows:
Section 2.1. and its sub-sections describe the data basis and prepa-
ration for the clustering approach; Section 2.2. describes the application
of the clustering approach and the formation of alternative eet seg-
ments. In Section 2.3., the random forest machine learning approach is
described, divided into sub-sections for data preparation (2.3.1.) and
model application (2.3.2). In the last Section 2.4., the theoretical
framework and calculation of the applied balance indicators used for
evaluation are described.
2.1. Clustering – data preparation
The newly developed eet segmentation approach focuses on a
detailed analysis of the catch composition for individual shing vessels
within the German shing eet from 2002 to 2020. This analysis in-
corporates landed catch weights, shing effort, and vessel characteris-
tics such as length, tonnage, and engine power, which were linked to the
landings data set.
2.1.1. Step 1: data segregation by gear classes
Initially, the dataset was categorized by gear classes (Table 1). From
the logbooks, the average trip duration and annual shing effort were
calculated and then used to assign the predominant shing gear to each
vessel based on the highest percentage of annual usage. For small-scale
coastal vessels without logbooks, estimates were derived from monthly
reports and assigned to a designated gear type labeled "PG".
2.1.2. Step 2: allocation of catches to stocks
Catches were then allocated to stocks (i.e. fractions of species
inhabiting dened geographic areas) according to species and catch
location, using species denitions from ICES, ICCAT (International
Commission for the Conservation of Atlantic Tunas), and GFCM (Gen-
eral Fisheries Commission of the Mediterranean). For areas lacking
specic ofcial stock descriptions, ‘pseudo-stocks’ were created based
on the combination of species codes from ICES (ICES, 1979) and FAO
shing areas (FAO, 2020). These are not ofcially dened stocks, yet,
they are necessary combinations of species information and geographic
areas to create a homogeneous data set.
2.1.3. Step 3: calculation of sock shares
The share of each stock in the total annual catch was calculated using
catch weight as a reliable variable due to its importance for stock
management (Davie and Lordan, 2011). Catch value was excluded from
consideration due to considerable uctuations in market prices
(Goti-Aralucea et al., 2021). Stocks contributing less than 5 % to a
vessel’s total catch were excluded to minimize distortions in the anal-
ysis, as they inate the time needed by the grouping algorithm and
inherit the risk of over-separation without adding any relevant infor-
mation. The 5 % threshold was set and tested by shing eet data ex-
perts in three eet segmentation workshops held prior to the preparation
of this article (RCG ECON, 2023, 2022 & 2021).
2.2. Clustering – model application
Based on the stock shares and the major shing gear, vessels were
then pre-separated, and a distance matrix was calculated using a metric
conversion of the Bray-Curtis-Dissimilarity Bray and Curtis, (1957),
included in vegan R-Package, version 2.6 – 4, (Oksanen, 2009) to analyze
the number of potential eet segments.
The Bray-Curtis dissimilarity D between the two catch proles j and k
is computed as
Djk =1−
2⋅∑
p
i=1
min(yij −yik)
∑
p
i=1(yij +yik)
Where p is the number of stocks occurring in both catch proles and
Σmin(y
ij
, y
ik
) is the sum of lesser shares of each stock on the total catch if
it occurs in both catch proles. The metric conversion M of the Bray-
Curtis-Dissimilarity D is then computed as
M=2D
(1+D)
The chosen distance measure for clustering adheres to the three
axiomatic principles of metrics - positive deniteness (i.e., the distance
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
3
between two points is always positive), symmetry (i.e., the distance
between point x and point y always equals the distance between y and
x), and triangle inequality (the sum of any two sides of a triangle formed
by the points x, y, and z is always greater than the third side) (Burago
et al., 2001). This adherence ensures robustness in post-hoc analysis and
cluster validation, as metric distances allow for the application of a
larger variety of methods which are also easier to interpret than
non-metric methods, e.g., multi-dimensional scaling (MDS) versus
non-metric multi-dimensional scaling (NMDS) (Borg and Groenen,
2007; Legendre and Legendre, 2012). Employing the distance matrix,
which reects the stock proportions from each vessel’s total catch, hi-
erarchical agglomerative clustering (HAC) is used. HAC starts by treat-
ing each vessel as an individual cluster and progressively merges them
into larger clusters using the UPGMA ("Unweighted Pair-Group Method
using arithmetic Averages") algorithm.
The selection of UPGMA as the optimal fusion algorithm was based
on its ability to maintain a high cophenetic correlation, which is the
linear correlation coefcient between the original distance matrix and
the clustering result (Sneath and Sokal, 1973). A high cophenetic cor-
relation signies a good reection of the underlying data by the clus-
tering result, underscoring the method’s efcacy. In UPGMA, the
distance between a single object and a cluster is calculated as the
average of distances between the object and every member of the group
(Sokal, 1958). This method is noted for its effective algorithm that
accurately determines distances during the clustering process,
enhancing the precision and relevance of the analysis.
d(AB)X=1
2(dAX +dBX)
In the clustering process, clusters are denoted by exemplary labels
such as A and B, which merge into a resultant cluster X with computed
distance d. Determining the ideal number of clusters within the data set
by applying the described procedure is challenging. To address this, a
combination of indices was used, including the silhouette coefcient
(Rousseeuw, 1987), the Mantel correlation (Borcard et al., 2011), and
the SD index (Halkidi et al., 2000). The silhouette coefcient and the SD
index are indicators based on the compactness and separation of clus-
ters, and therefore evaluate the quality of the clustering result
(Rousseeuw, 1987; Halkidi et al., 2000), whereas the Mantel correlation
between the original distance matrix and the binary partition matrices of
the clustering procedures evaluates the quality of the clustering result
with respect to the original data (Borcard et al., 2011). Even though
applying these indicators leads to selecting an appropriate number of
clusters, not all clusters accurately represented distinct eet segments,
with some being amalgamated based on similarities in vessel length,
catch composition, annual landed weight, and the Herndahl–Hirsch-
man index (HHI) of catch composition (Rhoades, 1993).
The HHI is a widely used economic index that measures the level of
competition or concentration within an industry. It ranges from 0 to 1,
where 1 indicates complete concentration while lower values indicate
greater competition. Applied to our sheries data, high HHI values near
1 suggest strong concentration of a shery on a specic stock, whereas
lower values indicate sheries with a diverse catch composition. This
nuanced application of the HHI helps understanding the degree of stock
concentration within different eet segments, facilitating more targeted
and effective management strategies.
The process for determining the optimal number of clusters involves
iterative loops to ensure accuracy, as shown in the owchart (Fig. 1).
Specic guidelines for applying the previously mentioned tests, select-
ing the appropriate number of clusters, and merging them are outlined
in the user manual of the R-package FleetSegmentation (Sulanke, 2021).
The clustering procedure segments each gear class dataset by reference
year, ultimately producing a comprehensively segmented eet dataset
that spans two decades.
2.3. Random forest approach
2.3.1. Random forest – data preparation
Before deploying machine learning techniques, Principal Component
Analysis (PCA) and correlation analysis were performed using the
prcomp and cor functions from the stats package (R Core Team, 2024).
PCA was chosen primarily for its ability to reduce the dimensionality of
the dataset, thereby simplifying the subsequent analysis while retaining
most of the variance present in the original data (Jolliffe and Cadima,
2016). This reduction in dimensionality helps in mitigating issues
related to multicollinearity, where highly correlated variables can
distort model predictions and make the model less interpretable. Addi-
tionally, PCA aids in noise reduction by focusing on principal compo-
nents that capture signicant variance, thus ltering out less
informative aspects of the data (James et al., 2013). This process
Fig. 1. Stepwise owchart of the developed eet segmentation approach. Dark sections highlight critical decision points for using the functions in the associated
R package.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
4
enhances the robustness and performance of the random forest model by
ensuring it is trained on the most relevant features.Variables with <5 %
contribution to distinguishing shing eet segments and those highly
correlated (>0.9 correlation coefcient) were excluded, resulting in
130,441 observations and 5 explanatory variables, including eet
segment categorization. This dataset was randomly split into a 70:30
learning-testing ratio iteratively.
2.3.2. Random forest – model application
A Random Forest (RF) model (Breiman, 2001) was applied to the
learning dataset using the randomForest package (Liaw and Wiener,
2002) with default tuning parameters for laymen applicability. The RF
model predicted eet segments for the testing dataset, and the results
were compared to the alternative segments. Confusion matrices were
generated to identify and quantify misclassications. Accuracy, Cohens
kappa values, and variable importance measures were determined. The
entire process was repeated in a 10-fold cross-validation, averaging
measures across bootstrapped models.
2.4. Balance indicators
To assess the improvements made by implementing the alternative
segmentation approach, both in terms of relevance and effectiveness for
sheries’ operational proles, a suite of both established and novel in-
dicators was used.
1. Cumulative Share Analysis: The cumulative share of segments on the
total catch of sh stocks was calculated, by calculating the shares of
the eet segments under both the DCF and novel segmentation
schemes on the total harvest of given stocks and then aggregating
these shares. The cumulative shares were subsequently analyzed
graphically.
2. Stocks Targeted by Each Segment: The number of stocks targeted by
each segment under both the novel and the DCF segmentation
schemes was calculated. In this part of the analysis, only stocks that
were harvested by more than one segment in at least one segmen-
tation approach were considered. Additionally, a stock had to ac-
count for at least 5 % of a segment’s total catch to be classied as
targeted.
3. Sustainable Harvest Indicator (SHI): The SHI was calculated for all
segments under both segmentation schemes. It quanties the reli-
ance of eet segments on potentially overshed stocks, taking into
account the health of the targeted stock as well as the eet eco-
nomics. This well-established indicator is an integral part of an
annual shing eet report of the STECF, in which the balance be-
tween the shing capacity and the shing opportunities in the EU
shing eet is assessed (STECF, 2024). It is calculated based on the
ratio between shing mortality (F) and shing mortality linked to
maximum sustainable yield (FMSY), where values above 1 (F / FMSY
>1) signify excessive economic reliance on vulnerable stocks. The
economic reliance is derived from the landing values. A detailed
description and discussion of the SHI methodology can be found in
the corresponding communication of the European Union (European
Commission, 2014).
4. Stocks at Risk (SAR): The number of SAR for each segment under
both segmentation schemes was determined. SAR accounts for the
number of stocks at low biological levels of biomass or productivity
that are economically important to the respective eet segment or
where the eet segment signicantly contributes to the overall
shing pressure on the stock. Like the SHI, it is a commonly used
indicator of the STECF (STECF, 2024). To be considered as an SAR
under the ofcial denition of STECF, a stock must meet at least one
criterion in each of two categories: 1) It must be assessed as being
below biological limits (B lim); subject to advice to close the shery
or reduce shing pressure; part of a regulation requiring catch and
release; or listed by IUCN or CITES. 2) It must constitute 10 % or
more of the catches by the eet segment or the eet segment must
harvest 10 % or more of the total catches of the stock.
These indicators collectively provide a comprehensive evaluation of
how well sheries are in line with their shing opportunities, how
reliant they are on stocks having a risk of being overshed and of how
well these sheries are represented by the underlying eet segmentation
approach. Therefore, they were chosen to illustrate how the new seg-
mentation approach aligns with the operational realities and conserva-
tion objectives of sheries, facilitating more informed management
decisions.
3. Results
In the analysis of the German shing eet from 2002 to 2020, fteen
distinct eet segments were identied, consisting of 2328 individual
vessels across ve different gear classes. Except for 2002, all identied
segments were consistently present throughout the years. The analysis
revealed a mix of multi-species sheries, predominantly targeting
various demersal sh assemblages, alongside single-species sheries
focused on demersal and pelagic sh, crustaceans, and bivalves. The
structure of these eet segments aligns with expert statements of the
German shing eet structure, which were collected before the analysis.
A comprehensive overview of all identied eet segments, the number
of vessels in each, and the type of sheries performed in the reference
year of 2018 is presented in Appendix Table A2. Fig. 2 illustrates the
relationship of DCF segments and alternative segments by displaying the
number of vessels per segment in the respective segmentation ap-
proaches as well as the relationship between segments formed in both
segmentation approaches. The details of the eet segments formed
applying the alternative approach and their relation to the DCF seg-
ments will be described in the following sections, which are structured
as follows. Sections 3.1 and 3.2 comprehensively describe the new eet
segments formed using the alternative approach; Sections 3.3 and 3.4
contain the results of the application of the random forest algorithm; and
Sections 3.5, 3.6, and 3.7 relate the alternative segments to the DCF
segments and describe the result of the evaluation based on the balance
indicators.
3.1. Small-scale and specialized sheries
Small-scale coastal sheries, which comprised about 70 % of the
vessels in the dataset, displayed a wide range of catch proles due to
their part-time or subsistence nature. Despite their operational diversity,
these vessels are often similar in cost structure, justifying their catego-
rization into a single eet segment labeled as "Small-scale passive gear
shery" for vessels using passive gear and less than 12 m in length.
However, few vessels under 12 m, actively engaged in the brown
shrimp shery but initially misclassied due to missing gear informa-
tion, were reassigned to “Brown shrimp shery”. This segment includes
the majority of beam trawlers (gear code TBB), and is noted for being the
largest and, in terms of revenue, most valuable coastal shery in the
German eet, accounting for about 18 % of the vessels annually.
Another small group of beam trawlers targeted plaice (Pleuronectes
platessa) and sole (Solea solea), with sole being the more lucrative spe-
cies. This segment was designated as "Sole shery". The remainder of
beam trawlers participated in the “ Blue mussel shery”, involving the
capture of seed mussels (Mytilus edulis) for cultivation. This "Blue mussel
shery" also contained all vessels using dredges (DRB).
3.2. Larger and diverse gear classes
Vessels longer than 12 m using passive gears – such as pots and traps
(FPO), drift and xed nets (DFN), and hools (HOK) – were grouped into a
single gear class, representing a smaller fraction (1–2 % annually) of the
eet.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
5
The demersal trawlers and seiners (DTS) represented the most
diverse gear class, distinguishing ve specic sheries. Two of those
were in the Baltic Sea: one primarily targeting herring (Clupea harengus)
and another mixed shery for plaice (Platichthys spp.), ounder (Plati-
chthys spp.), dab (Limanda limanda), and cod (Gadus morhua). In the
North Sea, one shery mainly targeted saithe (Pollachius virens) and cod,
and another is a mixed shery for plaice, sole, and Norway lobster
(Nephrops norvegicus). The fth is the high seas demersal shery oper-
ating Svalbard and Greenland, targeting Greenland halibut (Reinhardtius
hippoglossoides), cod, and redsh (Sebastes spp.).
The midwater trawlers (TM) also showed diverse activities, with one
segment comprising smaller vessels (<25 m) in the Baltic Sea catching
herring and sprat (Sprattus sprattus), while a second included larger
vessels (~40 m) in the North Sea targeting the same stocks, but also
sprat and sandeel (Ammodytes spp.). The last segment encompassed the
largest vessels (80–120 m) of the eet, operating in international waters
targeting pelagic species like mackerel (Scomber scombrus), herring, blue
whiting (Micromesistius poutassou), and Chilean jack mackerel (Trachu-
rus murphyi) in the Northeast Atlantic, as well as of the coast of West
Africa and in the Southern Pacic.
Overall, the novel approach to eet segmentation reduced the
number of eet segments from 24 to 15, with 65 % of DCF segments
containing vessels from two or more identied novel eet segments,
demonstrating more nuanced and efcient categorization reective of
operational realities.
3.3. PCA analysis and RF model implementation
The PCA (Fig. 3) of the data set resulted in 69.1 % of the variance
explained. Especially the variables ‘Vessel length’, ‘Gross tonnage’,
‘Engine power’, and ‘Mean trip length’ had a large contribution to
dimension 1, while ‘Gear type’, ‘Target assemblage ID’, and ‘Mean trip
length’ explained the separation of the observations along dimension 2
(see Fig. 3). The contribution of ‘Catch weight’ to the variance was
minimal, accounting for less than 5 % across both dimensions combined.
Closely spaced directional PCA vectors of the variables ‘Vessel
length’, ‘Gross tonnage’, and ‘Engine power’ already indicated a feature
correlation which was conrmed in a subsequent correlation analysis
(see Fig. 3). The correlation coefcient for all possible parameter com-
binations yielded between 0.9 and 0.96. Since the contribution of the
parameter ‘Vessel length’ (contrib.=13.6) to the PCA dimensions
exceeded the contribution of "Gross tonnage" (contrib.=13.3) and "En-
gine power" (contrib.=13.5), the highly correlating features ‘Gross
tonnage’ and ‘Engine power’ were omitted for downstream analysis to
streamline the dataset and focus on the most inuential parameters.
Hence, in addition to the reference label of shing eet segment, four
explanatory variables have met the previously set conditions and were
therefore used for RF model construction: i) ‘Vessel length’ (approx.
3.5–130 m), ii) ‘Mean trip time’ (approx. 30 mins-3.5 months), iii) ‘Gear
type’ (6 different gear types) and iv) ‘Target assemblage ID’ (14 different
IDs) (see Fig. 4). The RF models were able to correctly characterize the
shing eet segments in 99.18 percent of the observations in the testing
dataset (see Fig. 5). On average, the calculated kappa value was 0.98,
indicating a "perfect agreement" between prediction and reference
(Landis and Koch, 1977).
3.4. Error analysis and variable importance
Fishing eet segment classication errors were most frequent in the
segments ‘G_7’ (class error =7.1 %), ‘F_6’ (class error=3.2 %), ‘I_9’
(class error=3.1 %) and ‘C_3’ (class error=2.7 %). For all other shing
eet segment classications, the error rate was <1.5 %. For the
Fig. 2. Sankey owchart of the vessels in the German shing eet in 2018 classied according to DCF segmentation approach (left, orange) and the alternative
segmentation approach (green, right). The width of the rectangles and links is proportional to the segment size.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
6
prediction of ve shing eet segments (’E_5’, ’K_11’, ’L_12’, ’M_13’,
’N_14’), an error rate of zero was achieved, meaning that in all ten
bootstrap models, no single observation was assigned to an incorrect
shing eet segment. To determine variable contribution to the RF
model, variable importance measures for each feature were averaged
over the ten bootstrapped models. Overall, the feature ’Gear type’ was
detected to be the most important variable to enable correct classica-
tions of shing eet segments. The features ’Mean trip length’, ’Target
assemblage ID’, and ’Vessel length’ contributed in descending order to
correct classications. We found that G_7 and I_9 as well as F_6 and C_3
have overlaps in their catch composition with some vessels regularly
switching between the respective sheries. Therefore, the classication
error is the result of an actual overlap in shing strategies in a part of the
eet and not due to a mechanistic problem or a data issue. However,
they are not the only eet segments with overlapping catch proles
present in the data set. Those remaining segments with overlapping
catch proles had, as mentioned, very low classication errors. All
segments with zero classication errors are comprised of vessels with
unique combinations of technical features and catch proles with little
to no overlap with other segments.
3.5. Homogeneity of eet segments
Fleet segments are intended to reect and facilitate management
related to sheries and stock assessments. To compare the homogeneity
of segments in relation to sheries and stocks four specic metrics have
been employed: The SHI and SAR indicators, the analysis of the number
of stocks exploited per segment, and the number of segments exploiting
each stock. A full overview of the calculated SHI indicators and SARs can
be found in Appendix Table A3.
Multiple cases where vessels from two or more alternative segments
with considerably different SHIs and/or SARs were present within a
single DCF segment were detected. For example, three demersal trawler
length classes (DTS VL1218, DTS VL 1824, DTS VL 2440) encompassed
vessels from several alternative eet segments. Notably, the Baltic mixed
demersal and Baltic demersal forage sh (e.g., herring Clupea harengus
or sprat Sprattus sprattus), along with the North Sea mixed demersal
shery, were present across all these DCF segments. The SHIs for the
Baltic sheries were similar (1.85 and 1.81) and generally higher than
those of the corresponding DCF segments, whereas the North Sea mixed
demersal shery displayed a lower SHI of 1.18.
3.6. Challenges with DCF segmentation
The analysis highlights issues with the DCF segmentation, such as the
non-detection and misestimation of SHIs and SARs. For instance, the
Fig. 3. Relative contribution of labels to separation of observations among the two most discriminant PCA axis. Directions of the vectors indicate the axis to which
the factor contributes the most. The color coding shows the total relative contribution of the respective factor.
Fig. 4. Variable importance scores of the random forest, based on mean
decrease in accuracy and averaged from the ten cross-validation runs, scaled to
a total of 100.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
7
DTS VL1218 and DTS VL1824 were each assigned one SAR, but the DTS
VL2440 received none. This SAR was Western Baltic cod –
cod.27.22–24, which is only a relevant target species for the eet seg-
ments operating in the Baltic. Similarly, in the DFN VL2440 segment
(SHI: 1,85, no SAR), consisting of high seas passive and North Sea pas-
sive gear sheries, discrepancies in SHI and SAR assignments were
evident. While the former had no SAR and the SHI was not only low
(0.69) but also technically not applicable (assessed stocks accounted for
less than 40 % of the segments total catch), the latter had one of the
highest SHIs in the analysis (2.32) and one SAR (North Sea sole –
sol.27.4). For the DCF segment DFN VL 1218, two SARs were computed
(Western Baltic herring - her.27.20–24; North Sea sole – sol.27.4), yet
the herring SAR is only attributable to the fraction of the segment
assigned to the Baltic passive gear shery, while sole is only shed by
the fraction assigned to the North Sea passive gear shery. These cases
suggest that DCF segmentation may lead to overestimations or un-
derestimations depending on the stock and the applied metrics.
3.7. Stock exploration analysis
The analysis of how many segments harvest specic stocks showed
that Western Baltic herring (her.27.20–24) was harvested by eleven DCF
segments but only four alternative segments (see Fig. 6), underscoring
potential inefciencies in the DCF approach. In contrast, North Sea and
Northeast Atlantic herring (her.27.1–24a514a) were harvested by more
alternative segments than DCF segments, indicating a more rened
segmentation in the alternative model.
Graphical analysis of the number of segments necessary to assess
100 % of stocks’ total catch, six commercially important stocks revealed
that fewer alternative segments than DCF segments were needed to
cover a high share (>90 %) of total catch (see. Fig. 7). The mean
number of segments required to cover a high share of the analysed stocks
was considerably lower in the alternative segmentation (4.07) compared
to the DCF segmentation (5.6.).
4. Discussion
4.1. More effective management
The new eet segmentation method offers a more accurate division
of sheries data into segments that align closely with actual shing
activities compared to the current DCF segmentation. This novel
approach not only reduces the total number of eet segments but also
provides a closer link between groups of vessels, their targeted sh
stocks, and the shery performed. The latter became apparent in the
analysis of SHIs and SARs, as it exposed the missing connection of eet
segments and targeted shing stocks in the DCF segmentation as a major
downside. Results showed that this method of segmenting shing eets
can hide concerning imbalances in specic sheries. Additionally,
combining multiple shing strategies within a single DCF segment can
produce inaccurate indicator results, SHIs as well as SARs. If this indi-
cator is inaccurate, it may give a false impression of the actual condition
of the shery. For example, an inaccurate SHI might suggest that a eet
segment might excessively rely on an overshed sh stock, or vice versa
might imply an excessive reliance, which is actually not present. Inac-
curate SHIs derived from current DCF segmentation can misguide
Fig. 5. Confusion matrix resulting from RF predictions. The predicted eet segment classication for each observation is compared to the respective reference
classication. The confusion matrix diagonal shows the percentages of correctly classied observations among 10-fold cross-validation (green). Non-diagonal tiles
show the proportion of misclassied categories.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
8
policymakers into implementing measures that do not address the real
issues. For instance, a shery might be subject to unnecessarily strict
regulations if SHIs falsely indicate imbalances, or it might be under-
regulated if SHIs and SARs falsely suggest that eets are in balance. A
good example of such a case is the situation of German trawl sheries
(DTS) described in the results (see Section 3.6.). The SAR is assigned to
the demersal trawl DTS segments referred to Western Baltic cod
(cod.27.22–24), which was mainly targeted by the Baltic fraction of
German trawl sheries (‘Baltic mixed demersal shery’ in alternative
segmentation), yet, it was in a very critical state and has reached a
catastrophic stock collapse recently, critically jeopardizing the
continued existence of this shery (M¨
ollmann et al., 2021). This
example highlights how such inaccuracies in eet segmentation and
subsequent indicator calculation can contribute to the continued
degradation of sh stocks and the marine ecosystem, ultimately making
future recovery efforts more difcult and costly.
Moreover, effective sheries management involves the allocation of
resources, such as funding for conservation efforts, monitoring pro-
grams, and enforcement. Inaccurate SHIs can lead to the misallocation
of these resources, directing them away from areas that need them most
and towards areas that do not. Besides biological implications, this can
have signicant economic and social consequences for shing commu-
nities. Overly restrictive measures can harm livelihoods, while insuf-
cient measures can lead to the collapse of sh stocks, affecting long-term
sustainability and economic stability. Applying the novel segmentation
approach reduced the number of segments harvesting specic stocks.
Consequently, these rened segments can represent shing strategies
more accurately and precisely than DCF segments. With fewer segments
involved, it is easier to analyze the socio-economic impacts of stock-
specic management measures, such as TAC reductions, ensuring
more targeted and effective sheries management.
Fisheries management relies on accurate data to make informed
decisions. A database derived from a shery-based segmentation
scheme, which categorizes eets by vessel catch composition and stock
reference, is crucial as it acknowledges the variability in shing activ-
ities across different vessels with vessels engaged in multiple sheries
which may be impacted differently by management measures or stock
uctuations.
4.2. Integration with ecosystem-based management approaches
While the shift towards ecosystem-based management approaches is
growing (Scotti et al., 2022; Bonsdorff et al., 2015; M¨
ollmann et al.,
2014; Gascuel et al., 2012), such methods are still not the norm in
sheries management. For both stock-based and ecosystem-based
management, understanding which parts of the eet are shing which
stocks and to what extent is critical. Only then can well-informed de-
cisions be made and the consequences of management measures for
sheries be accurately assessed. The new approach offers a dynamic and
systematic method to link eet activities directly with current sheries
management practices, which predominantly focus on stock allocations
and technical measures. This not only applies to the European eet
segmentation in the DCF, in which the novel approach has been suc-
cessfully used for national eet data sets (Grigoraș et al., 2023). The
functions of the novel approach also can be and have been applied to a
variety of data sets on multiple aggregation levels, e.g., for the
Fig. 6. Bar chart of the number of segments harvesting specic stocks in the German shing eet in 2018. The number of segments is displayed on the X-axis; the
stocks on the Y-axis. The color signies the applied segmentation scheme (green =alternative, orange =DCF).
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
9
classication of haul compositions in the Icelandic demersal trawl eet
(Kasper et al., 2024). Beyond the European scope, it can enable re-
searchers and data globally to not only segment the shing eets of their
respective interest, but also serve as an explorative toolset in an earlier
stage. Participants of the workshops where the approach was tested
reported discovering shing strategies in the researched eets that they
did not know existed before. This highlights the universal applicability
of our work in well-established data collection systems as well as in
emerging management organizations, where ecosystem-based ap-
proaches are yet to be established.
4.3. Compliance with DCF segmentation
‘Fishery’ is an optional dimension that has been in the DCF data calls
for economic eet data. However, it is not included in the legislative
basis, is not clearly dened yet, and is not applied in the DCF environ-
ment (STECF, 2023). It is hence recommended to integrate the novel
approach with the existing DCF segmentation framework by applying
the novel approach to dene the ‘Fishery’-dimension of segments. To
avoid the described disadvantages of the current DCF segmentation, a
revision of the length and gear classes within the DCF would addition-
ally be required, for which a possible pathway is described in the
following sections. The integration of the novel approach in the DCF
framework would enhance the granularity and applicability of the seg-
mentation, allowing for a reduction in the overall number of segments
and improving management efciency.
4.4. Homogeneity in cost structures
Another possible advantage of the new segmentation method is its
ability to create eet segments with a more homogenous cost structure
then the DCF segmentation. Cost structure homogeneity is one of the
primary goals of the DCF data collection and analysis (STECF, 2023).
Cost data of shing eets is often sparse and heterogeneous and there-
fore difcult to analyze, yet, rst calculations of cost structures of eet
segments dened using the alternative approach showed promising re-
sults (Sulanke, 2020, internal DCF report). Homogeneous cost structures
attribute is particularly valuable when evaluating the economic per-
formance of different sheries or calibrating bio-economic models.
However, because cost data is often scarce and collected from a limited
sample of vessels, further research using a harmonized analysis pro-
cedure, broader data sources from different shing nations, and thor-
oughly analyzing the economic indicators applied by the DCF is
necessary to validate changes in cost structure homogeneity brought
about by the novel approach.
4.5. Spatial component and environmental considerations
By using stock-based catches as the basis of the clustering procedure,
the new method includes a spatial component, separating vessels that
might use the same gears and target similar species but operate in
different areas under varied environmental conditions and management
regulations. This feature is essential for accurately assessing the impacts
of area-specic management measures such as size restrictions or sea-
sonal closures. Using the shery-based segmentation procedure, three
Fig. 7. Ranked cumulative share of total catch of specic sh stocks under DCF and alternative segmentation. The number of segments is displayed on the X-axis; the
cumulative share of weight is displayed on the Y-axis. Segments were ranked in descending order according to their share of the total catch of the respective stock.
Stocks were selected because of their commercial importance for multiple segments under both segmentation schemes.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
10
eet segments were identied in a separate workow not described in
detail in this article. Each of these eet segments included vessels from
at least three different DCF segments. Comparing the haul positions
reported in the logbooks of vessels revealed spatially distinct mixed
sheries in the Baltic and in the North Sea, targeting substantially
different stocks and being affected by different regulations, like, e.g., the
Plaice Box, which is only relevant for North Sea sheries (Beare et al.,
2013). This example of demersal sheries in the German coastal zone
illustrates that a stock-based, multivariate analysis allows the identi-
cation of spatially distinct shing eets, something a purely technical
classication alone cannot ensure.
4.6. Future directions and recommendations
The new method is ready to use, as it is implemented in a user-
friendly R package. Feedback from scientic partners and workshops
(RCG ECON, 2023, 2022 & 2021) has highlighted the need for an
automated process within the eet segmentation package to decide
which clusters should be combined into eet segments without requiring
extensive expert knowledge. This improvement would simplify the
segmentation process and make it more easily applicable for users with
varying levels of expertise. Possible variables to consider in the algo-
rithm are catch composition and catch quantity, technical parameters of
the vessels, travel times and extent, and position data, such as VMS data.
In the current version of the novel eet segmentation approach, the
decision on which clusters to merge into valid eet segments is made by
the user with the aid of a set of diagnostic methods and comparisons.
These methods are not always unambiguous, and in some cases, the
decision requires in-depth expert knowledge of the shing eet under
consideration. While at least basic knowledge of the analyzed shing
eets substantially aids the application of the approach, the three
workshops held on the topic made clear, that only basic knowledge of
the R programming language is required to successfully run the package
containing the novel approach (RCG ECON, 2023, 2022 & 2021). The
machine-learning aspect is yet to be tested in a workshop. Yet, it’s
implementation in the R-package will not reduce the overall applica-
bility of the package and does not require any specic machine-learning
skills from the users, but in turn will reduce the overall requirements to
the user. Once a sufcient number of years (about 3–4) are segmented
using our described protocol, the machine learning approach can auto-
matically assign them to any additional data sets., e.g., older or
incoming year classes of eet data. The presented case study is based on
the German eet, which is, compared to other European eets, of me-
dium size and complexity (STECF, 2023). The application of the
approach to data sets of larger and more diverse eets during the
workshops, e.g., the Spanish, Danish, or French eets, highlighted the
principal applicability to such eets, yet, possible improvements were
also identied. These included the reduction of highly complex catch
composition data sets, e.g., by applying principal components, and a
novelization of the gear classication, which will be discussed in the
following.
Adjustments in the allocation of the main shing gear and the po-
tential inclusion of special gear classes should be considered to enhance
the accuracy and relevance of eet segmentation. In the novel eet
segmentation approach, vessel gear class is a pivotal yet complex factor.
Vessels often use multiple shing gears, with variations across different
seasons. Currently, the main shing gear is based on its usage over 50 %
of the logged shing time, as documented in the vessel’s logbook.
However, the gear with the highest usage does not necessarily reect its
signicance in terms of catch composition. For example, the Baltic Sea’s
demersal swamsh shery segment consists of vessels categorized as
DTS, as they predominantly used bottom trawls according to logbook
records. Yet, the primary catch of these vessels included herring and
sprat, typically caught with pelagic trawls – specically pair trawls
during limited seasonal windows. Despite their shorter time of usage,
these pelagic trawls contribute the majority of the total catch, leading to
a potential misclassication in eet segments.
This discrepancy underscores the need for revising the main shing
gear assignment procedure within the DCF. The current system merges
gear types like demersal trawls and seines into a single category (DTS),
which does not accurately reect their distinct operational patterns and
cost structures. For instance, demersal seines generally consume less fuel
compared to other gear types, affecting operational costs signicantly
(Cheilari et al., 2013). To address these issues, it’s proposed that vessels
switching between gears such as pelagic and demersal trawls, or be-
tween active and passive gears should be classied into the following
categories: mixed active and passive gears (PMP) and polyvalent active
gears (MGP).
Though these gear classes exist in the DCF, they are very rarely
implemented in current practice. They could be incorporated into the
DCF’s segmentation scheme as specic shing techniques (level 1.b, see
Annual Economic Report Metadata Protocol, (STECF, 2024)) or as a
combination of shing technique and gear (levels 1.b and 2.b, see
Annual Economic Report Metadata Protocol, STECF, 2024).
To facilitate the identication of these specialized gear classes, a
clustering procedure could be performed before applying a protocol
similar to the eet segmentation approach presented. This preliminary
step would help ensure that gear classications are based on actual
operational proles, enhancing the accuracy of eet segmentation and
ultimately supporting more precise sheries management practices.
Judging by the experience made in the development and testing
stages of the novel eet segmentation approach, we consider a separa-
tion of small-scale sheries (SSF, under 12 m, following EU denition
(Natale et al., 2015)), large-scale sheries (LSF, over 12 m and less than
500gt) and distant water sheries (DWF, over 500gt, following the
denition used by DCF and widely applied in the shing sector) instead
of the currently applied six length classes. As the analysis of variable
contribution to the random forest classication revealed, the gear class
is by far the most important variable for the random forest algorithm to
assign alternative eet segments, and the vessel length is among the four
most important variables as well. This highlights that these variables,
which are the core of the DCF classication scheme, should not be fully
abolished. Rather, the classication scheme described above would be
benecial in two ways, as it would ease the application of the novel
approach and at the same time make it more closely related to the
original DCF segmentation scheme. In addition, revising how gear
classications are determined and applied in eet segmentation will
better align the segmentation process with the real-world complexities
of shing operations. These technical improvements together with the
demonstrated overall advances in connectivity of eet segments and
biological sh stock assessment facilitated by the novel approach will
lead to considerable improvements in effective and targeted sheries
management at the European level and beyond. As global changes
continue to reshape marine environments, an adaptive and
well-segmented eet structure will be indispensable for the future of
sustainable sheries management. Therefore, integrating
catch-prole-related aspects into eet segmentation in a mechanistic,
standardized, and easily appliable way has the potential to considerably
enhance the effectiveness of sheries management, as it supports the
development of strategies tailored to precisely address today’s unique
challenges.
CRediT authorship contribution statement
Thorsten Stoeck: Writing – review & editing, Supervision, Project
administration, Methodology, Funding acquisition. Sarah Simons:
Writing – review & editing, Supervision, Project administration, Fund-
ing acquisition. J¨
org Berkenhagen: Writing – review & editing, Meth-
odology, Funding acquisition, Conceptualization. Matthias
Bernreuther: Writing – review & editing, Formal analysis. Erik
Sulanke: Writing – original draft, Visualization, Methodology, Formal
analysis, Data curation. Verena Rubel: Writing – original draft,
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
11
Visualization, Methodology, Formal analysis, Data curation,
Conceptualization.
Funding
Verena Rubel received funding from the Deutsche For-
schungsgemeinschaft (DFG), grant STO414/15–2 awarded to TS. Erik
Sulanke has been funded by the Thünen Institute of Sea Fisheries, Bre-
merhaven, and also received funding from the European Union’s Hori-
zon 2020 Research and Innovation Program under grant agreement No.
869300 (FutureMARES project).
Declaration of Competing Interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Data availability statement
Since the data set used contains condential individual vessel data, it
can’t be made publically available.
Appendix
Table A1
Geographic indicators to distinguish eet segments operating in outermost regions or exclusively in non-EU waters, either in-
ternational waters or in third countries’ waters based on sheries partnership agreements.
code Name denitions
NEU Non EU waters more the 50 % of activity occurs in non-EU waters
IWE International waters exclusively 100 % of activity occurs in non-EU waters
NGI No geographical indicator National waters, EU waters
P2 Madeira Portuguese outermost region (autonomous region)
P3 Azores Portuguese outermost region (autonomous region)
IC Canaries Spanish outermost region (autonomous community)
MA Morocco Coastal Most of the activity occurs in 34.1.1
GF French Guiana French outermost region (overseas department)
GP Guadeloupe French outermost region (overseas department)
MQ Martinique French outermost region (overseas department)
MF Saint-Martin French outermost region (since 2009) (overseas community)
RE Reunion French outermost region (overseas department)
YT Mayotte French outermost region (overseas department)
Table A2
Fleet segments of the German shing detected using the alternative segmentation approach. Included are the name and size of the new eet segment, vessels main
shing gear of the vessels, and length, engine power, and mean trip duration, all described by minimum, maximum, and mean in parentheses. Also depicted are the
main target stocks (ofcial ICES stocks are indicated in bold, otherwise species abbreviation and FAO area are indicated), and the DCF segments of the vessels included
in the new segment.
Fleet segment Number of
vessels
Main shing
gear
Main target stocks Vessel length (m) Engine power (kW) Mean trip
duration
(d)
DCF segments
included
Blue mussel
shery
4–14 (9) Dredges
(DRB)
−Blue mussel (Mytilus edulis,
MUS, 27.4b)
9.2–45.65 (37.33) 58–1200 (594) 1.13 VL2440DRB,
VL40XXDRB,
VL2440TBB
Brown
shrimp
shery
173–253 (214) Beam trawls
(TBB)
−Brown shrimp (Crangon
crangon,CSH, 27.4abc)
4.10–35.67 (16.59) 5–1103 (182.39) 1.46 VL0010PG,
VL0010TBB,
VL1012TBB,
VL1218TBB,
VL1824TBB,
VL2440TBB
Sole shery 5–20 (9) Beam trawls
(TBB)
−North Sea sole (Solea solea,
sol.27.4)
−North Sea plaice (Pleuronectes
platessa, ple.27.4.20)
15.69–41.82 (32.59) 184–1471 (820) 4.02 VL2440TBB,
VL40XXTBB
North Sea
passive
gear
shery
3–12 (6) Gillnets and
stow nets
(DFN)
−North Sea cod (Gadus
morhua, cod.27.47d20)
−North Sea sole (Solea solea,
sol.27.4)
−Smelt (Osmerus eperlanus,
SME, 27.4b)
5.3–31.78 (16.38) 7–485 (167) 3.28 VL0010_TBB,
VL1218DFN,
VL1824TBB,
VL2440DFN
Baltic
passive
gear
shery
3–17 (10) Gillnets
(DFN)
−Western Baltic herring
(Clupea harengus,
her.27.20–24)
−Western Baltic ounder
(Platichthys spp.,
bwq.27.2425)
6.6–32.36 (16.43) 18–441 (136) 0.93 VL1218_DFN,
VL1218_FPO
(continued on next page)
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
12
Table A2 (continued )
Fleet segment Number of
vessels
Main shing
gear
Main target stocks Vessel length (m) Engine power (kW) Mean trip
duration
(d)
DCF segments
included
−Baltic plaice (Pleuronectes
platessa,ple.27.24–32)
High seas
passive
gear
shery
3–6 (5) Gillnets
(DFN), Pots
and traps
(FPO)
−Anglersh, North Sea and
British isles (Lophius
budegassa, & piscatorius,
anf.27.3a46)
−Blackbellied anglersh,
Keltische See und Biscaya
(Lophius budegassa,
anf.27.78abd)
−Red deep-sea crab (Chaceon
quinquedens, KEF, 27.6. u.
27.7)
26.72–32.36 (29.67) 404–442 (425) 35.88 VL2440_FPO,
VL2440_DFN
Small-scale
passive
gear
sheries
662–1012 (860) Mixed
passive gears
(PG)
−Western Baltic herring
(Clupea harengus,
her.27.20–24)
−Western Baltic cod (Gadus
morhua, cod.27.22–24)
−Plaice, Kattegat, Belt Seas
and the Sound (Pleuronectes
platessa, ple.27.21–23)
−Eel (Anguilla anguilla,
ele.2737.nea)
−Pike-Perch (Sander lucioperca,
FPE, 27.3)
3.10 – 11.99 (6.66) 0–221 (26) <1 VL0010_PG,
VL1012_PG
Baltic
pelagic
herring
shery
1–9 (4) Midwater
trawls (TM)
−Western Baltic herring
(Clupea harengus,
her.27.20–24)
11.30–37.31 (21.56) 100–735 (209) 1.14 VL1824_TM,
VL2440_TM
Coastal
pelagic
forage sh
shery
1–4 (2) Midwater
trawls (TM)
−Baltic sprat (Sprattus sprattus,
spr.27.22–32)
−Norway spring-spawning
herring (Clupea harengus,
her.27.1–24a514a)
−North Sea sprat (Sprattus
sprattus, spr.27.3a4)
−Sandeel – Area 4 (Ammodytes
spp., san.sa.4)
17.35–53.55 (41.34) 219–2309 (1025) 7.36 VL40XX_TM
High seas
pelagic
shery
3–6 (4) Midwater
trawls (TM)
−Northeast Atlantic Blue
whiting (Micromesistius
poutassou, whb.27.1–91214)
−Norway spring-spawning
herring (Clupea harengus,
her.27.1–24a514a)
−Sardine, Mauritania (Sardina
pilchardus, PIL, 34.1.3)
−Northeast Atlantic Mackerel
(Scomber scombrus, mac.27.
nea)
−North Sea herring (Clupea
harengus, her.27.3.a47d)
62.22–140.8 (108.68) 1764–8640 (4756) 31.8 VL40XX_TM
Baltic
demersal
forage sh
shery
3–54 (30) Demersal
trawls (DTS)
−Baltic sprat (Sprattus sprattus,
spr.27.22–32)
−Western Baltic herring
(Clupea harengus,
her.27.20–24)
−Eastern Baltic herring (Clupea
harengus, her.27.25–2932)
−Plaice, Kattegat, Belt Seas
and the Sound (Pleuronectes
platessa, ple.27.21–23)
−Baltic whiting (Merlangius
merlangus, WHG, 27.3b,c)
9.23–39.48 (16.99) 50–961 (203) 1.13 VL1012_DTS,
VL1218_DTS,
VL1824_DTS,
VL2440_DTS
Baltic mixed
demersal
shery
15–64 (33) Demersal
trawls (DTS)
−Plaice, Kattegat, Belt Seas
and the Sound (Pleuronectes
platessa, ple.27.21–23)
−Western Baltic ounder
(Platichthys spp.,
bwq.27.2425)
−Western Baltic cod (Gadus
morhua, cod.27.22–24)
−Baltic dab (Limanda limanda,
dab.27–22–32)
8.38–32.34 (16.43) 37–588 (183) 1.47 VL0010_DTS,
VL1012_DTS,
VL1218_DTS,
VL1824_DTS
(continued on next page)
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
13
Table A2 (continued )
Fleet segment Number of
vessels
Main shing
gear
Main target stocks Vessel length (m) Engine power (kW) Mean trip
duration
(d)
DCF segments
included
North Sea
mixed
demersal
shery
7–19 (13) Demersal
trawls (DTS)
−North Sea plaice (Pleuronectes
platessa, ple.27.420)
−Norway lobster in the areas 5
und 33 (Nephrops norvegicus,
nep.fu.5, nep.fu.33)
14.72–40 (24.38) 159–1440 (282) 4.41 VL1218_DTS,
VL1824_DTS,
VL2440_DTS
Saithe & cod
shery
6–17 (10) Demersal
trawls (DTS)
−Saithe, North Sea and
adjacent waters (Pollachius
virens, pok.27.3a4a)
−Northern hake (Merluccius
merluccius,
hke.27.3a46–8abd)
−North Sea cod (Gadus
morhua, cod.27.47d20)
15.11–40.26 (32.65) 130–1720 (641) 5.13 VL2440_DTS,
VL40XX_DTS
High seas
demersal
shery
4–6 (5) Demersal
trawls (DTS)
−Cod, Northeast Arctic (Gadus
morhua, cod.27.1–2)
−Western Greenland halibut
(Reinhardtius hippoglossoides,
ghl.27.561214)
−Eastern Greenland cod
(Gadus morhua,
cod.2127.1f14)
−Pelagischer redsh (Sebastes
mentella, reb.27.1–2.sp und,
reb.27.1–2.dp)
57–92.1 (77.96) 1764–4500 (3242) 51.18 VL40XX_DTS
Table A3
SHI and SAR of both the alternative and the DCF segmentation approaches for the reference year 2018.
segment name segmentation SHI number of SAR comment
Baltic passive gear shery alternative 1.85 1
Baltic pelagic herring shery alternative 1.89 1
Baltic mixed demersal shery alternative 1.85 1
Baltic demersal forage sh shery alternative 1.81 1
Blue mussel shery alternative NA 0
Small-scale passive gear shery alternative 2.35 2
Sole shery alternative 2.04 1
Saithe & cod shery alternative 1.37 0
North Sea mixed demersal shery alternative 1.18 0
North Sea passive gear shery alternative 2.32 1
Brown shrimp shery alternative NA 0
High seas passive gear shery alternative 0.69 0 SHI not relevant as proportion of shed stocks with F/Fmsy present is below 40 %
High seas demersal shery alternative 0.91 0
Coastal pelagic forage sh shery alternative 1.05 0
High seas pelagic shery alternative 0.94 0
PG VL0010 DCF 2.44 2 SHI not relevant as proportion of shed stocks with F/Fmsy present is below 40 %
PG VL1012 DCF 2.28 2
DFN VL1218 DCF 2.28 2
DFN VL2440 DCF 1.85 0
FPO VL1218 DCF 1.84 1
FPO VL2440 DCF NA 0
TBB VL1012 DCF NA 0
TBB VL1218 DCF 1.34 0 SHI not relevant as proportion of shed stocks with F/Fmsy present is below 40 %
TBB VL1824 DCF 0.95 0 SHI not relevant as proportion of shed stocks with F/Fmsy present is below 40 %
TBB VL2440 DCF 2.11 1
TBB VL40XX DCF 1.93 1
DTS VL1012 DCF 1.65 2
DTS VL1218 DCF 2.10 1
DTS VL1824 DCF 1.35 1
DTS VL2440 DCF 1.38 0
DTS VL40XX DCF 1.07 1
TM VL 1218 DCF 1.95 2
TM VL1824 DCF 1.88 1
TM VL2440 DCF 1.65 1
TM VL40XX DCF 0.95 0
DRB VL2440 DCF NA 0
DRB VL40XX DCF NA 0
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
14
References
Baba, S., Matsuishi, T., 2015. Pacic saury shing forecasting by using random forest.
Nippon Suisan Gakkaishi 81 (1), 2–9.
Barkai, A., Felaar, F., Geggus, K., Meredith, G., 2010. A complete data recording and
reporting system for the EU commercial shing eets. IEEE 1–9.
Beare, D.J., Adhuri, D., Evans, L., Garces, L., Sulu, R., Adrianto, L., & Andrew, N. 2013.
Implementing an ecosystem approach to sheries (EAF) in small-scale tropical
marine sheries. WorldFish, Penang, Malaysia. First interim report for EU Action
DCIENV/2011/221352 (Interim report for EU Action DCI-ENV/2011/221352 No.
1). Worldsh, Penang, Malaysia.
Behivoke, F., Etienne, M.-P., Guitton, J., Randriatsara, R.M., Ranaivoson, E., L´
eopold, M.,
2021. Estimating shing effort in small-scale sheries using GPS tracking data and
random forests. Ecol. Indic. 123, 107321. https://doi.org/10.1016/j.
ecolind.2020.107321.
Berio, F., Bayle, Y., Baum, D., Goudemand, N., Debiais-Thibaud, M., 2022. Hide and seek
shark teeth in Random Forests: Machine learning applied to Scyliorhinus canicula
populations. PeerJ 10, e13575. https://doi.org/10.7717/peerj.13575.
Bonsdorff, E., Andersson, A., Elmgren, R., 2015. Baltic Sea ecosystem-based management
under climate change: Integrating social and ecological perspectives. Ambio 44,
333–334.
Borcard, D., Gillet, F., Legendre, P., 2011. Numerical ecology with R. Springer.
Borg, I., Groenen, P.J., 2007. Modern multidimensional scaling: Theory and applications.
Springer Science & Business Media.
Branch, T.A., Hilborn, R., Haynie, A.C., Fay, G., Flynn, L., Grifths, J., Marshall, K.N.,
Randall, J.K., Scheuerell, J.M., Ward, E.J., 2006. Fleet dynamics and shermen
behavior: lessons for sheries managers. Can. J. Fish. Aquat. Sci. 63 (7), 1647–1668.
Bray, J.R., Curtis, J.T., 1957. An ordination of the upland forest communities of southern
Wisconsin. Ecol. Monogr. 27 (4), 326–349.
Breiman, L., 2001. Random forests. Mach. Learn. 45, 5–32.
Brierley, A.S., Kingsford, M.J., 2009. Impacts of climate change on marine organisms and
ecosystems. Curr. Biol. 19 (14), R602–R614.
Burago, D., Burago, Y., Ivanov, S., 2001. pp. xiv+. In: A course in metric geometry, Vol.
33. American Mathematical Society, Providence, p. 415.
Burden, M., Fujita, R., 2019. Better sheries management can help reduce conict,
improve food security, and increase economic productivity in the face of climate
change. Mar. Policy 108, 103610.
Cheilari, A., Guillen, J., Damalas, D., Barbas, T., 2013. Effects of the fuel price crisis on
the energy efciency and the economic performance of the European Union shing
eets. Mar. Policy 40, 18–24.
Cheung, W.W., Lam, V.W., Sarmiento, J.L., Kearney, K., Watson, R.E.G., Zeller, D.,
Pauly, D., 2010. Large-scale redistribution of maximum sheries catch potential in
the global ocean under climate change. Glob. Change Biol. 16 (1), 24–35.
Cordier, T., Forster, D., Dufresne, Y., Martins, C.I.M., Stoeck, T., Pawlowski, J., 2018.
Supervised machine learning outperforms taxonomy-based environmental DNA
metabarcoding applied to biomonitoring. Mol. Ecol. Resour. 18 (6), 1381–1391.
https://doi.org/10.1111/1755-0998.12926.
R. Core Team. 2024. _R: A Language and Environment for Statistical Computing_. R
Foundation for Statistical Computing, Vienna, Austria. 〈https://www.R-project.
org/〉.
Cutler, D.R., Edwards, T.C., Beard, K.H., Cutler, A., Hess, K.T., Gibson, J., Lawler, J.J.,
2007. Random forests for classication in ecology. Ecology 88 (11), 2783–2792.
https://doi.org/10.1890/07-0539.1.
Davie, Sarah, Lordan, Colm, 2011. Denition, dynamics and stability of m´
etiers in the
Irish otter trawl eet (S). : Fish. Res. 111 (3), 145–158. https://doi.org/10.1016/j.
shres.2011.07.005.
D´
epalle, M., Th´
ebaud, O., Sanchirico, J.N., 2020. Accounting for eet heterogeneity in
estimating the impacts of large-scale shery closures. Mar. Resour. Econ. 35 (4),
361–378.
D¨
orner, H., Casey, J., Carvalho, N., Damalas, D., Graham, N., Guillen, J., Holmes, S.J.,
Natale, F., Osio, G.C., R¨
atz, H.-J., 2018. Collection and dissemination of sheries
data in support of the EU Common Fisheries Policy. Ethics in Science and
Environmental Politics, 18, 15–25.
European Commission. 2014. Communication from the European Parliament and the
Council. Guidelines for the analysis of the balance between shing capacity and
shing opportunities according to Art 22 of Regulation (EU) No 1380/2013 of the
European Parliament and the Council on the Common Fisheries Policy. COM(2014)
545 nal. Publications Ofce of the European Union, Luxembourg, 2014. https://
eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2014:0545:FIN:EN:PDF.
European Commission. 2021. Commission Delegated Decision (EU) 2021/1167 of 17
April 2021 establishing the multiannual Union programme for the collection and
management of biological, environmental, technical and socioeconomic data in the
sheries and aquaculture sectors from 2022. Publications Ofce of the European
Union, Luxembourg, 2021. https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?
uri=CELEX:32021D1167.
Faith, D.P., Minchin, P.R., Belbin, L., 1987. Compositional dissimilarity as a robust
measure of ecological distance. Vegetatio 69, 57–68.
FAO 2020. FAO Major Fishing Areas. Accessed last 08/02/2024. https://www.fao.org/
shery/en/area/search.
FAO. 2024. The State of World Fisheries and Aquaculture 2024 – Blue Transformation in
action. Rome. https://doi.org/10.4060/cd0683en.
Fiskeridirektoratet. 2021. Lønnsomhetsundersøkelse for skeåten 2019/Protability
survey of the Norwegian shing eet 2019.
Frost, H., Andersen, P., 2006. The common sheries policy of the European Union and
sheries economics. Mar. Policy 30 (6), 737–746.
Gaines, S.D., Costello, C., Owashi, B., Mangin, T., Bone, J., Molinos, J.G., Burden, M.,
Dennis, H., Halpern, B.S., Kappel, C.V., 2018. Improved sheries management could
offset many negative effects of climate change. Sci. Adv. 4 (8), eaao1378.
Garcia, J., Pasqualini, V., Vanalderweireldt, L., Bisgambiglia, P.A., Marengo, M.,
Lejeune, P., Aiello, A., Durieux, E.D.H., 2022. Global patterns and environmental
drivers of suitable habitat for Dentex dentex and Sciaena umbra along the Corsican
coast. ICES J. Mar. Sci., Artic. fsac184. Adv. Online Publ. https://doi.org/10.1093/
icesjms/fsac184.
García-Fl´
orez, L., Morales, J., Gaspar, M.B., Castilla, D., Mugerza, E., Berthou, P., de la
Fuente, L.G., Oliveira, M., Moreno, O., del Hoyo, J.J.G., 2014. A novel and simple
approach to dene artisanal sheries in Europe. Mar. Policy 44, 152–159.
Gascuel, D., Merino, G., D¨
oring, R., Druon, J.N., Goti, L., Guenette, S., Mackinson, S.,
2012. Towards the implementation of an integrated ecosystem eet-based
management of European sheries. Mar. Policy 36 (5), 1022–1032.
Goti-Aralucea, L., Berkenhagen, J., Sulanke, E., D¨
oring, R., 2021. Efciency vs resilience:
the rise and fall of the German brown shrimp shery in times of COVID 19. Mar.
Policy 133, 104675.
Gourlie, D., 2017. Reeling in Uncertainty: Adapting Marine Fisheries Management to
Cope with Climate Effects on Ocean Ecosystems. Environ. Law 47, 179.
Grigoraș, D., P˘
aun, C.V., Țiganov, G., Danilov, C.S., Diaconu, D., 2023. Alternative
Approach to the Segmentation of Fishing Fleets. Cercet˘
ari Mar. -Rech. Mar. 53 (1),
92–912.
Guyader, O., Berthou, P., Koutsikopoulos, C., Alban, F., Demaneche, S., Gaspar, M.,
Eschbaum, R., Fahy, E., Tully, O., Reynal, L., 2013. Small scale sheries in Europe: A
comparative analysis based on a selection of case studies. Fish. Res 140, 1–13.
Halkidi, M., Vazirgiannis, M., Batistakis, Y. 2000. Quality scheme assessment in the
clustering process. In Principles of Data Mining and Knowledge Discovery: 4th
European Conference, PKDD 2000 Lyon, France, September 13–16, 2000
Proceedings 4 (pp. 265-276). Springer Berlin Heidelberg.
Holley, J.-F., Marchal, P., 2004. Fishing strategy development under changing
conditions: examples from the French offshore eet shing in the North Atlantic.
ICES J. Mar. Sci. 61 (8), 1410–1431. https://doi.org/10.1016/j.
icesjms.2004.08.010.
ICES 1979. ICES List of Species Items. ICES Expert Group reports (until 2018). Report.
https://doi.org/10.17895/ices.pub.19260197.v1.
ICES 2003. Report of the Study Group on the Development of Fishery-based Forecasts.
ICES Document CM 2003/ACFM: 08 Ref. D., 37.
ICES 2024. Scoping workshop 3 on next generation of mixed sheries advice
(WKMIXFISH3). ICES Scientic Reports. Report. https://doi.org/10.17895/ices.
pub.25897327.v1.
James, G., Witten, D., Hastie, T., Tibshirani, R., 2013. An Introduction to Statistical
Learning: with Applications in R. Springer.
Jim´
enez, M.P., Sobrino, I., Ramos, F., 2004. Objective methods for dening mixed-
species trawl sheries in Spanish waters of the Gulf of C´
adiz. Fish. Res. 67 (2),
195–206. https://doi.org/10.1016/j.shres.2003.09.048.
Jolliffe, I.T., Cadima, J., 2016. Principal component analysis: a review and recent
developments. Philosophical Transactions of the Royal Society A: Mathematical,
Physical and Engineering Sciences 374 (2016), 20150202.
Joo, R., Bertrand, S., Tam, J., Fablet, R., 2013. Hidden Markov models: the best models
for forager movements? PloS One 8 (8), e71246.
Kasper, J.M., Oostdijk, M., Baranowska, E., Sturlud´
ottir, E., 2024. Implement. a M. ´
eTier.
-Based Dyn. Fish. Model Atlantis Model Icel. Waters.
Landis, J.R., Koch, G.G., 1977. An application of hierarchical kappa-type statistics in the
assessment of majority agreement among multiple observers. Biometrics 363–374.
Legendre, P., & Legendre, L. 2012. Numerical ecology (Vol. 24). Elsevier.
Lennert-Cody, C.E., Berk, R.A., 2007. Statistical learning procedures for monitoring
regulatory compliance: an application to sheries data. J. R. Stat. Soc.: Ser. a (Stat.
Soc. ) 170 (3), 671–689.
Lewy, P., 1994. Identication of Danish North Sea trawl sheries. ICES J. Mar. Sci. 51
(3), 263–272. https://doi.org/10.1006/jmsc.1994.1027.
Li, Y., Sun, M., Ren, Y., Chen, Y., 2021. Fisher behavior matters: Harnessing spatio-
temporal shing effort information to support China’s sheries management. Ocean
Coast Manag. 210, 105665.
Liaw, A., Wiener, M., 2002. Classication and regression by randomForest. R. N. 2 (3),
18–22.
Lloret, J., Cowx, I.G., Cabral, H., Castro, M., Font, T., Gonçalves, J.M., Gordoa, A.,
Hoefnagel, E., Mati´
c-Skoko, S., Mikkelsen, E., 2018. Small-scale coastal sheries in
European Seas are not what they were: ecological, social and economic changes.
Mar. Policy 98, 176–186.
Miles, E. 2011. Fisheries management and governance challenges in a changing climate.
M¨
ollmann, C., Cormon, X., Funk, S., Otto, S.A., Schmidt, J.O., Schwermer, H., Quaas, M.,
2021. Tipping point realized in cod shery. Sci. Rep. 11 (1), 14259.
M¨
ollmann, C., Lindegren, M., Blenckner, T., Bergstr¨
om, L., Casini, M., Diekmann, R.,
Gårdmark, A., 2014. Implementing ecosystem-based sheries management: from
single-species to integrated ecosystem assessment and advice for Baltic Sea sh
stocks. ICES J. Mar. Sci. 71 (5), 1187–1197.
Moura, T., Fernandes, A., Alpoim, R., Azevedo, M., 2016. Unravelling the dynamics of a
multi-gear eet–Inputs for sheries assessment and management under the Common
Fisheries Policy. Mar. Policy 72, 219–230.
Murawski, S.A., Lange, A.M., Sissenwine, M.P., Mayo, R.K., 1983. Denition and analysis
of multispecies otter-trawl sheries off the northeast coast of the United States. ICES
J. Mar. Sci. 41 (1), 13–27. https://doi.org/10.1093/icesjms/41.1.13.
Natale, F., Carvalho, N., Paulrud, A., 2015. Dening small-scale sheries in the EU on the
basis of their operational range of activity The Swedish eet as a case study. Fish.
Res. 164, 286–292.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
15
Oksanen, J. 2009: Ordination and analysis of dissimilarities: tutorial with R and Vegan.
In: University Tennessee: Knoxville, TN, USA.
Pawson, M.G., Glenn, H., Padda, G., 2008. The denition of marine recreational shing
in Europe. Mar. Policy 32 (3), 339–350.
Pelletier, D., Ferraris, J., 2000. A multivariate approach for dening shing tactics from
commercial catch and effort data. Can. J. Fish. Aquat. Sci. 57 (1), 51–65. https://doi.
org/10.1139/cjfas-57-1-51.
Pilar-Fonseca, T., Campos, A., Afonso-Dias, M., Fonseca, P., Mendes, B., 2009. Fleet
segmentation of the Portuguese coastal trawl shery: a contribution to sheries
management. ICES CM 2009/O 29, 1–17.
Poos, J., Quirijns, F., Rijnsdorp, A., 2010. Spatial segregation among shing vessels in a
multispecies shery. Ices J. Mar. Sci. 67 (1), 155–164.
Propst, D.B., Gavrilis, D.G., 1987. Role of economic impact assessment procedures in
recreational sheries management. T Am. Fish. Soc. 116 (3), 450–460.
R.C.G.E.C.O.N., Regional Coordination on Economic Issues. 2021. Report of the
workshop on an alternative approach to the segmentation of shing eets (RCG
ECON) March 29 - 31, 2021. 2021. Available at: https://datacollection.jrc.ec.europa.
eu/docs/other-meetings. DOI: 10.13140/RG.2.2.30895.82083.
RCG ECON, Regional Coordination on Economic Issues. 2022. Report of the second
workshop on an alternative approach to the segmentation of shing eets. 2022.
Available at: https://datacollection.jrc.ec.europa.eu/docs/other-meetings. DOI:
10.13140/RG.2.2.30895.82083.
RCG ECON, Regional Coordination on Economic Issues. 2023. Report of the 3rd ISSG
workshop on an alternative approach to the segmentation of shing eets. Sulanke
E., and Berkenhagen, J., editor(s). https://www.sheries-rcg.eu/wp-content/
uploads/2024/04/2023_05_ISSG_Segmentation_WS-Report.pdf.
Rhoades, S.A., 1993. The herndahl-hirschman index. : Fed. Res. Bull. 79, S. 188.
Rogers, J.B., Pikitch, E.K., 1992. Numerical Denition of Groundsh Assemblages
Caught Off the Coasts of Oregon and Washington Using Commercial Fishing
Strategies. Can. J. Fish. Aquat. Sci. 49 (12), 2648–2656. https://doi.org/10.1139/
f92-293.
Rousseau, S., Gauthier, S., Neville, C., Johnson, S., Trudel, M., 2022. Acoustic
Classication of Juvenile Pacic Salmon (Oncorhynchus spp) and Pacic Herring
(Clupea pallasii) Schools Using Random Forests (Article). Front. Mar. Sci. 9, 857645.
https://doi.org/10.3389/fmars.2022.857645.
Rousseeuw, Peter J., 1987. Silhouettes: a graphical aid to the interpretation and
validation of cluster analysis. : J. Comput. Appl. Math. 20, S. 53–65.
Salas, S., Gaertner, D., 2004. The behavioural dynamics of shers: management
implications. Fish Fish 5 (2), 153–167.
Scotti, M., Opitz, S., MacNeil, L., Kreutle, A., Pusch, C., Froese, R., 2022. Ecosystem-
based sheries management increases catch and carbon sequestration through
recovery of exploited stocks: The western Baltic Sea case study. Front. Mar. Sci. 9,
879998.
Seash. 2023. Economics of the UK shing eet 2022. Sea Fish Industry Authority, 18
Logie Mill, Logie Green Road, Edinburgh, EH7 4HS.
Sethi, S.A., Dalton, M., Hilborn, R., 2012. Quantitative risk measures applied to Alaskan
commercial sheries. Can. J. Fish. Aquat. Sci. 69 (3), 487–498. https://doi.org/
10.1139/f2011-170.
Simmonds, E., D¨
oring, R., Daniel, P., Angot, V., 2011. The role of sheries data in the
development evaluation and impact assessment in support of European sheries
plans. Ices J. Mar. Sci. 68 (8), 1689–1698.
Smit, W., 1996. An economic approach to measuring shing effort: application to a
Dutch cutter eet. Mar. Resour. Econ. 11 (4), 305–311.
Sneath, P.H.A., Sokal, R.R., 1973. Numer. Taxon. Princ. Pract. Numer. Classif.
Sokal, R.R., 1958. A statiscal method for evaluating systematic relationships. : Univ.
Kans. Sci. Bull. 38, S. 1409–1438.
STECF, S., 2022. Technical and Economic Committee for Fisheries. In: Stock
Assessments: demersal stocks in Adriatic, Ionian and Aegean Seas and straits of Sicily
(STECF-22-16), 2022. Publications Ofce of the European Union, Luxembourg.
STECF, S., Technical and Economic Committee for Fisheries. 2023. The 2023 Annual
Economic Report on the EU Fishing Fleet (STECF 23-07), Prellezo, R., Sabatella, E.,
Virtanen, J., Tardy Martorell, M. and Guillen, J. editor(s), Publications Ofce of the
European Union, Luxembourg, 2023, doi:10.2760/423534, JRC135182.
STECF, S., Technical and Economic Committee for Fisheries. 2024. Assessment of
balance indicators for key eet segments and review of national reports on Member
States efforts to achieve balance between eet capacity and shing opportunities
(STECF-23-13), Casey, J. and Virtanen, J. editor(s), Publications Ofce of the
European Union, Luxembourg, 2024, https://data.europa.eu/doi/10.27.60/958072,
JRC1336331.
Sulanke, E. 2021. FleetSegmentation: A Package For The Segmentation Of Fishing Fleets_
. R package version 1.0.3, commit c0220e23b3350dbc70d61c12cdb1c4cd9922302a,
〈https://github.com/ESulanke/FleetSegmentation〉.
Thorpe, R.B., Dolder, P.J., Reeves, S., Robinson, P., Jennings, S., 2016. Assessing shery
and ecological consequences of alternate management options for multispecies
sheries. Ices J. Mar. Sci. 73 (6), 1503–1512.
Ulrich, C., Wilson, D.C., Nielsen, J.R., Bastardie, F., Reeves, S.A., Andersen, B.S.,
Eigaard, O.R., 2012. Challenges and opportunities for eet-and m´
etier-based
approaches for sheries management under the European Common Fishery Policy.
Ocean Coast Manag. 70, 38–47.
Van Poorten, B.T., Cox, S.P., Cooper, A.B., 2013. Efcacy of harvest and minimum size
limit regulations for controlling short-term harvest in recreational sheries. Fish.
Manag. Ecol. 20 (2-3), 258–267.
Witherell, D., Fey, M., & Fina, M. 2012. Fishing Fleet Proles. Anchorage, Alaska, USA.
Zhang, C., Ye, Z., Li, Z., Wan, R., Ren, Y., Dou, S., 2016. Population structure of Japanese
Spanish mackerel Scomberomorus niphonius in the Bohai Sea, the Yellow Sea and
the East China Sea: evidence from random forests based on otolith features. Fish. Sci.
82 (2), 251–256. https://doi.org/10.1007/s12562-016-0968-x.
E. Sulanke et al.
Fisheries Research 281 (2025) 107190
16
