Content uploaded by Daniel Grigoraș
Author content
All content in this area was uploaded by Daniel Grigoraș on Feb 02, 2024
Content may be subject to copyright.
Content uploaded by Tiganov George
Author content
All content in this area was uploaded by Tiganov George on Feb 01, 2024
Content may be subject to copyright.
92
Alternative Approach to the Segmentation of
Fishing Fleets
Daniel Grigoraș, Cătălin Valentin Păun,
George Țiganov, Cristian Sorin Danilov,
Dragoș Diaconu
“Cercetări Marine“
Issue no. 53
Pages 92-105 2023
DOI: 10.55268/CM.2023.53.92
ALTERNATIVE APPROACH TO THE SEGMENTATION
OF FISHING FLEETS
Daniel Grigoraș*, Cătălin Valentin Păun, George Țiganov,
Cristian Sorin Danilov, Dragoș Diaconu
National Institute for Marine Research and Development “Grigore Antipa",
300 Mamaia Blvd., Constanta 900581, Romania
*Corresponding author: dgrigoras@alpha.rmri.ro
ABSTRACT
The appropriate segmentation of fishing fleets is controversially discussed in fisheries
research and management and a variety of approaches has been introduced. The present
approach, developed in a pilot study funded by the European Commission – Data Collection
Framework (DCF), introduces a standardized multivariate approach for characterizing
fisheries fleet segments by hierarchical agglomerative cluster analysis (HAC) of their catch
composition. We chose data from 2021 of the Romanian fishing fleet as the basis of our
analysis. Statistical analyses were performed using the program RStudio V3.6.1 by running
the fleet segmentation package script. The specific indices, tests, and visual validation
methods of the package were applied to determine the optimal number of clusters. The
procedure was finalized by a post-hoc validation of the clustering result to identify the actual
fleet segments. From the basic data, 6 fleet segments for Vessels using active and passive
gears (PMP) were highlighted, representing 52 boats, where it was noted that fishing at Rapa
whelk (RPW) prevailed with 91.94% of the total catches on the segment, respective 5 fleet
segments for Vessels using passive gears (PG) only for vessels <12m, representing 78
vessels where the main catches or recorded at European anchovy (ANE) 25.16%, turbot
(TUR) 13.08%, horse mackerel (HMM) 8.75%, thus 130 vessels from two classes of
different gears in total. We detected mixed fishing, especially on various assemblages of
demersal and pelagic fish, as well as target fishing on demersal and pelagic fish, Rapa
whelk, and mussels. For a better understanding of the approach, further research is needed.
Keywords: Romanian fleet segmentation, catch composition, Black Sea
AIMS AND BACKGROUND
As of the end of the year 2021, the fishing fleet comprised 163 vessels
with a total capacity of 1575.26 GT and 6198.29 kW. Out of these, 130
vessels were active, and they were categorized into two fishing techniques:
52 vessels using Vessels using active and passive gears (PMP) and 78
vessels using Vessels using passive gears (PG).
93
For an accurate and realistic analysis of the fishing fleets, it is crucial
to adopt a new approach to fleet segmentation, different from the current
one. The existing segmentation approach relies only on the dominant gear
class and vessel length (LOA).
The new approach to fleet segmentation should involve using
statistical programs capable of handling large volumes of data. Specific
scripts can be developed to process, analyze, and interpret the data,
generating comprehensive and detailed sets of information.
By implementing this approach, member states can ensure the quality
of data transmitted on the data collection platforms in the fisheries sector,
meeting the requirements of the European Union.
EXPERIMENTAL
Statistical analyses were conducted using the RStudio V3.6.1 program
by executing the fleet segmentation package script (Fig. 1). This package
includes functions for a multivariate approach to fleet segmentation in target
fisheries. It was developed by Erik Sulanke, a research associate at the
Thuenen Institute for Marine Fisheries in Bremerhaven, Germany,
economics subunit. The pilot project that led to the development of this
package was established in the DCF of Scientific, Technical and Economic
Committee for Fisheries (STECF).
Fig. 1. Updated stepwise flowchart of the newly developed fleet segmentation
approach (Sulanke E. and Berkenhage J., 2022)
Preparing and Loading fleet data. This first step describes uploading
data about our fishing fleet. The respective data were structured in a
framework of generated data that will be needed later for grouping the data
into clusters.
94
The vessels were grouped according to the type of fishing gear they
use in two fishing techniques, combining mobile and passive gears (PMP)
and vessels with passive gears (PG), this separation was necessary because
most of the vessels in the fishing fleet in Romania use a combination of
types of gear, both the active ones and the most active ones.
This segmentation procedure uses two input data frames, one frame
containing catch data for all vessels of a certain gear class and the second
frame containing information about the length of the vessels (Sulanke,
2022).
Examine the best number of clusters to use. In the second step,
technical knowledge related to the segmentation of fishing fleets is necessary
to correctly interpret the data and determine the appropriate number of
clusters.
The script developed by Sulanke 2022, utilizes several indices and
tests in this step to make an accurate choice of the number of clusters. To
obtain the most reliable grouping results, it is necessary to run several scripts
with different numbers of clusters. This allows for result comparison and
facilitates the final decision-making process.
The indices presented in (Table 1 and 2) act as measures for evaluating
different cluster configurations and help identify the most suitable number of
clusters for the given dataset.
Table 1. Indices optional of clusters for PMP vessels
Indices optional
No. of clusters
Index value
Average silhouettes*
8
0.788
Mantel test*
9
0.949
Davis_Bouldin index*
14
0.088
SD index*
5
2.884
Calinski-Harabasz
indexv*
15
1129.278
*Average silhouettes – The approach measures the quality of a clustering.
That is, it determines how well each object lies within its cluster. A high
average silhouette width indicates good clustering.
*Mantel test – The matrices must be of the same dimension; in most
applications, they are matrices of interrelations between the same vectors of
objects.
*Davi_Bouldin index – This is an internal evaluation scheme, where the
validation of how well the clustering has been done is made using quantities
and features inherent to the dataset.
*SD index – The standard deviation index is a measurement of bias (how
close your value is to the target value)
95
*Calinski-Harabasz index – Clustering validation has been recognized as one
of the important factors essential to the success of clustering algorithms.
How to effectively and efficiently assess the clustering results of clustering
algorithms is the key to the problem
Table 2. Indices optional of clusters for PG vessels
Indices optional
No. of clusters
Index value
Average silhouettes*
9
0.559
Mantel test*
15
0.861
Davis_Bouldin index*
11
0.531
SD index*
6
2.11
Calinski-Harabasz index*
14
44.184
For our segmentation, we selected the average silhouette index as the
most representative measure for both PMP and PG fishing techniques,
resulting in 8 and 9 clusters, respectively.
Test plot clustering diagnostics. By applying this hierarchical
agglomerative clustering (HAC) procedure (Fig. 2 and 3), with various
graphical tests, indices, and methods, the correct number of clusters in a
dataset can be determined.
Fig. 2. Gridded diagnostic plots of the HAC procedure – PMP vessels
96
Fig. 3. Gridded diagnostic plots of the HAC procedure – PG vessels
A) screen plot showing the total within sums of squares of the clustering
vs. the number of clusters used.
B) average silhouette width of the clusters vs. the number of clusters.
C) mantel test, i.e. the Pearson correlation between the clustering and the
original distance matrix vs. the number of clusters.
D) dendrogram of the HAC procedure with the linkage distance of the
clusters on the y-axis. The y-axis shows the linkage distance of
clusters (Sulanke, 2022).
In the plots generated by the script for both fishing techniques (Fig. 2
and 3) at the average silhouette index, it can be observed that the number of
clusters remains 8 for PMP and 9 for PG, which is the same number of
clusters as in the previous step.
Test dendrogram clustering diagnostics. In the continuation of the
Hierarchical Agglomerative Clustering (HAC) procedure, dendrograms are
used (Fig. 4 and 5). Dendrograms are graphical representations that display
the hierarchical clustering results obtained during the agglomeration process.
As the HAC algorithm iteratively merges similar data points or clusters, the
dendrogram visually illustrates these merging steps.
97
Fig. 4. Grided plot of three dendrograms of the HAC procedure – PMP vessels
Fig. 5. Grided plot of three dendrograms of the HAC procedure – PG vessels
A) is unmodified.
B) is cut at a linkage distance of 0.75 and resulting branches are
individually colored and the resulting number of clusters is labeled.
C) has cutting lines at linkages distance 0.5 and 0.9 and the range of
cluster numbers resulting from those cutting heights are labeled
(Sulanke, 2022).
Based on the information provided in the dendrograms, it becomes
apparent that they both support the conclusion that there are 8 to 9 distinct
clusters in the dataset. A dendrogram visually represents the hierarchical
98
clustering process and illustrates how data points or clusters are grouped
together based on their similarity or distance from each other.
Plot clustering tree. During the various tests for analyzing the number
of clusters, we experimented with different combinations of cluster
configurations and evaluation measures. This was done to assess the quality
and suitability of the clustering solutions at each step of the analysis. Toward
the end of the test, the clustering tree procedure was introduced. A clustering
tree is another method of clustering analysis, and it serves as a visual
representation of the hierarchical clustering process.
Fig. 6. Tree plot of the HAC-procedure – PMP vessels
Fig. 7. Tree plot of the HAC-procedure – PG vessels
99
The clustering tree makes it easy to trace back changes that occur when
new clusters are introduced. When a sixth group is added, cluster number 1
is divided into two new groups: one contains 37 vessels, and the other
contains 4 vessels for PMP, while for PG, one contains 16 vessels and the
other contains 57 vessels. These newly formed groups may represent valid
fleet segments.
RESULTS AND DISCUSSION
Following the HAC procedure, several variants of cluster selection
resulted from the entered data set. After the correlation and analysis of each
variant separately, it was established that the most correct variant is 8
clusters for the PMP vessels and 9 clusters for the PG vessels.
Catch composition of clusters. According to the script, the first step
for cluster characterization is to evaluate their catch (Fig. 8 and 9). The
average percentage that each stock contributes to the catches of a cluster can
be tabulated and graphed (Sulanke, 2022).
Fig. 8. Bar plot of the average percentage of each stock on the total catch of each
cluster, PMP vessels
The average percentage is depicted on the x-axis, the clusters on the y-
axis. The colors indicate the magnitude of the stock's contribution to the total
average catch. The stock names are indicated with labels, stocks are ordered
by their average percentage contribution. The plot illustrates the differences
in catch composition among the clusters (Sulanke, 2022).
From the basic data, 6 fleet segments for Vessels using active and
passive gears (PMP) were highlighted, representing 52 boats, where it was
100
noted that fishing at Rapa whelk (RPW) prevailed with 91.94% of the total
catches on the segment, respective 5 fleet segments for Vessels using passive
gears (PG) only for vessels <12m, representing 78 vessels where the main
catches or recorded at European anchovy (ANE) 25.16%, turbot (TUR)
13.08%, horse mackerel (HMM) 8.75%, thus totaling 130 vessels from two
classes of different gears. We detected mixed fishing, especially on various
assemblages of demersal and pelagic fish, as well as target fishing on
demersal and pelagic fish, Rapa whelk and mussels.
Fig. 9. Bar plot of the average percentage of each stock on the total catch of each
cluster, PG vessels
Plotting a 2-dimensional. This method of ordering in 2 dimensions
helps with a better visual identification of the overlaps and therefore, the
clusters can be united by matching them in the specific segment of the fleet
(Fig. 10-11). In this step of the script, a metric MDS was used because it
reflects real distances, not ranked distances (Sulanke, 2022).
Points represent individual vessels, colored according to their cluster
affiliation. Clusters are labeled, and labels are colored like points. The
goodness of fit (GoF) is labeled. This MDS reveals some of the relationships
in the catch composition and shows a good fit (Sulanke, 2022).
As can be seen in the PMP vessels (Fig. 10), a matching of the vessels
that form cluster 1 (which includes most of the vessels) and cluster 4 can be
distinguished, because they are consistent with their catch composition that
fish the same stocks, but in different amounts. Regarding the PG vessels
(Fig. 11), it can be clearly observed that most of the clusters are in the right
area of the MDS forming 2 representative clusters that have most of the
vessels in their competence, are cluster 2 and 3, it can also be observed that
101
cluster 1 is at a great distance from the other clusters, having turbot and
common stingray in its composition.
Fig. 10. MDS of the transformed vessel catch data – PMP vessels
Fig. 11. MDS of the transformed vessel catch data – PG vessels
Vessel and cluster properties. This last step characterizes the
clusters obtained based on the available data: the number of vessels in the
clusters, the length distribution of the vessels in the cluster, the total catch of
individual vessels, and the total catch of each cluster (Fig. 12-13), (Sulanke,
2022).
102
Fig. 12. Grid of mixed plot types displaying vessel characteristics – PMP vessels
Fig. 13. Grid of mixed plot types displaying vessel characteristics – PG vessels
A) number of ships in each group is plotted on the y-axis versus the group
number on the x-axis. The number above each bar gives the number of
vessels in the group.
B) vessel length in meters is plotted on the y-axis versus the group
number on the x-axis.
C) annual catches by vessels in tones on the y-axis versus group number
on the x-axis.
D) total catch in tons is represented on the y-axis versus the group number
on the x-axis (Sulanke, 2022).
103
This graphical grid contains a lot of information about the clusters and
allows us to group them according to their structure.
Table 3 provides information on fishing vessel activity based on two
fishing techniques, PG and PMP. Table 4 highlights the activities of the
fishing vessels belonging to the most representative clusters within the PMP
fishing technique. More precisely, it shows the distribution of vessels in
three clusters, with cluster 1 having 37 vessels, and clusters 2 and 4 each
having 3 vessels. Similarly, (Table 5) presents the activities of fishing
vessels within the PG fishing technique, focusing on the most representative
clusters. The table indicates that cluster 1 consists of 16 vessels, cluster 3 has
25 vessels, while clusters 5 and 7 comprise 3 vessels each.
Only the relevant information was extracted based on which an
analysis of the fishing fleet can be made in 2021, Cluster 1 from PMP
vessels and cluster 3 from PG vessels add up to the largest total catches in
the fishing fleet 2 837.219 tons representing 90.72%.
Table 3. Activity of fishing vessels by fleet segment in 2021
PMP
PG
Vessels
Catches
(tons)
Amount
(euro)
Vessels
Catches
(tons)
Amount (euro)
no.
52
2973.056
1880725
no.
78
154.0905
355759
%
40.00
95.07
84.10
%
60.00
4.93
15.90
Table 4. Activity of fishing vessels on combining mobile and passive gears (PMP)
clusters in 2021
CLUSTER 1 - 93.12%
CLUSTER 2 - 3.08%
CLUSTER 4 - 2.21%
Stock*
Catch (t)
Vessels
Stock*
Catch
(t)
Vessels
Stock*
Catch
(t)
Vessels
RPW
2693.279
37
SPR
38.865
3
RPW
31.515
3
MSM
33.352
4
MUT
32.215
3
MSM
31.551
3
TUR
29.825
21
TUR
14.101
3
HMM
5.374
9
HMM
3.860
3
BLU
5.141
9
BLU
1.179
3
Total
37
TOTAL
3
TOTAL
3
DOMI
NANT
MIN
MAX
DOMI
NANT
MIN
MAX
DOMI
NANT
MIN
MAX
15 m
7 m
26 m
15 m
15 m
20 m
8 m
7 m
8 m
*FAO fish species code: RPW - Rapana venosa; MSM - Mytilus galloprovincialis; TUR -
Scophthalmus maximus; HMM - Trachurus mediterraneus; BLU - Pomatomus saltatrix;
SPR - Sprattus sprattus; MUT - Mullus barbatus; JDP - Dasyatis pastinaca; SHC - Alosa
pontica; RJC - Raja clavata; MGA - Liza aurata; MBF - Mesogobius batrachocephalus;
GPA – Gobiidae; ATB - Atherina boyeri; ANE - Engraulis encrasicolus; WHG -
Merlangius merlangus.
Table 5. Activity of fishing vessels on passive gears (PG) clusters in 2021
CLUSTER 1 - 12.81%
CLUSTER 3 – 44.00%
CLUSTER 5 – 5.36%
CLUSTER 7 – 5.36%
Stock*
Catch
(t)
Vessels
Stock*
Catch
(t)
Vessels
Stock*
Catch
(t)
Vessels
Stock*
Catch
(t)
Vessels
TUR
18.256
16
TUR
1.635
6
RPW
11.402
2
WHG
2.940
2
JDP
1.045
2
SPR
6.957
6
MSM
39.527
3
HMM
1.413
3
SHC
1.419
12
RPW
0.848
2
RJC
0.185
2
MUT
1.229
9
MGA
0.880
4
MBF
0.594
8
JDP
0.963
6
HMM
10.946
21
GPA
1.961
11
BLU
0.898
2
ATB
1.256
2
ANE
37.842
25
TOTAL
16
TOTAL
25
TOTAL
3
TOTAL
3
DOMI
NANT
MIN
MAX
DOMI
NANT
MIN
MAX
DOMI
NANT
MIN
MAX
DOMI
NANT
MIN
MAX
7 m 5 m 11 m 11 m 5 m 11 m 5 m 4 m 5 m 7 m 6 m 7 m
104
105
CONCLUSIONS
In 2021, there was a decrease in both the total quantity landed and the
value in euro compared to 2020. The total quantity landed in 2021 amounted
to 3.127.146 tons, while the corresponding value in euro was 2.236 484. This
represented a decline of 29.93% in the share of landings and a decrease of
19.31% in the value when compared to the previous year.
Rapa whelk is represented in 2021 by (87.82 %) of the total catches
in Romania, followed by mussels with (4.00%) and turbot with (2.38 %).
The fishing vessels grouped by fishing techniques combining mobile
and passive gears (PMP) used during the activities of both active and passive
fishing tools beam trawl (TBB), midwater otter trawl (OTM), gillnet set
(GNS), beach seine (SB) and manual to collect Rapa whelk and mussel.
Such activities were performed by almost all ship categories 24-40 PMP, 18-
24 PMP, 12-18 PMP, and 06-12 PMP.
The fishing vessels with passive gears (PG) fishing techniques were
those of the 00-06 PG and 06-12 PG class segments, the main tools used
were: set gillnet (GNS), stationary uncovered pound nets (FPN), hand and
pole lines (LHP) and set longlines (LLS).
Acknowledgment. This study case was part of the workshop organized by
Erik Sulanke and Jorg Berkenhagen from the Thünen Institute for Sea
Fisheries in Bremerhaven, Germany under the coordination of data collection
framework (DCF).
REFERENCES
Sulanke E. and Berkenhage J. (Eds) (2022). Report of the workshop on an
alternative approach to the segmentation of fishing fleets, 29-30
March 2021, doi: 10.13140/RG.2.2.15792.23043
Sulanke E. (2022), Fleet segmentation R-package script. Available online at
https://github.com/ESulanke/FleetSegmentation/blob/main/vignettes/
FleetSegmentation_vignette.Rmd
MADR (Ministerul Agriculturii și Dezvoltării Rurale), 2022. Romania’s
annual report on efforts to achieve a sustainable balance between
fishing capacity and fishing opportunities for 2021. Available online
at https://oceans-and-fisheries.ec.europa.eu/system/files/2022-
09/2021-fleet-capacity-report-romania_en.pdf
RStudio V3.6.1. Avaible online at https://cran.r-
project.org/bin/windows/base/old/3.6.1/
