Introduction
Chile extends across more than four thousand kilometres of productive coastline within the Humboldt Current upwelling system, one of the highest pelagic production zones globally (Van der Meer et al., Reference Van der Meer, Arancibia, Zylich and Zeller2015). It sustains at least two extractive economies that operate from many of the same ports but follow different biological targets, technologies and market logics. The Chilean industrial fleet operates primarily on high-biomass pelagic resources such as jack mackerel (Trachurus murphyi), anchoveta (Engraulis ringens), and Chilean sardine (Strangomera bentincki). A substantial share of these landings is directed to fishmeal and fish oil production, and these pelagic resources are managed under nationally defined annual catch limits. The link between industrial pelagic landings, reduction to meal and oil, and quota-setting for species such as jack mackerel and anchoveta is documented in official statistical reporting and technical quota justifications produced by the Chilean fisheries authority (SERNAPESCA, 2022, 2023, 2024). In Chile, the Undersecretary of Fisheries and Aquaculture assigns catch quotas, while SERNAPESCA is the Chilean public agency responsible for enforcing compliance with fisheries and aquaculture regulations.
The artisanal fleet operates with different target resources, territorial access arrangements, and reporting categories. It works from several hundred officially registered landing coves (caletas), where reported landings include high-value benthic invertebrates (e.g. sea urchin, king crab, luga), commercially harvested macroalgae, and coastal finfish destined for direct human consumption. Much of this activity is carried out under spatially explicit access regimes, in particular Áreas de Manejo y Explotación de Recursos Bentónicos (AMERB; Managed Benthic Areas and Exploitation Regime), often described as territorial user-rights systems. These regimes assign harvest rights over specific benthic grounds and algal beds to local artisanal organisations and have been described as a structural feature of Chile’s contemporary artisanal fishery (Gelcich et al., Reference Gelcich, Hughes, Olsson, Folke, Defeo, Fernández, Simon, Gunderson, Rodríguez-Sickert, Scheffer, Steneck and Castilla2010; Aburto et al., Reference Aburto, Gallardo, Stotz, Cerda, Mondaca-Schachermayer and Vera2013).
The figures used in this study come from official landings reported by the National Fisheries and Aquaculture Service (SERNAPESCA, 2022, 2023, 2024). These statistics distinguish three components: the artisanal subsector (declared catches by registered artisanal vessels, shore collectors, and management areas), the industrial subsector (landings by industrial vessels and factory ships), and aquaculture harvest, which includes production from marine grow-out sites and land-based systems. The ‘total landings’ published by the authority correspond to the sum of these three components at the annual and regional scale, and they form the operational baseline for catch control and access allocation in Chile.
Existing descriptions of fisheries production often report landings by species, fleet, or administrative area separately, which limits the ability to compare the multivariate composition of regional landing systems in a transparent and reproducible way. This study addresses that descriptive gap by representing each administrative region through proportional multispecies landing profiles and by applying an exploratory clustering workflow to identify similarities among regional landing compositions.
The objective is not to test whether Chilean artisanal and industrial fisheries are institutionally or spatially separated. These subsectors already operate under differentiated access rules, reporting systems, and management arrangements. Instead, the objective is to quantify how multispecies landing compositions vary across administrative regions within each subsector, and to assess whether this variation can be summarised through reproducible regional landing-composition types.
The contribution of the study is therefore threefold. First, it provides a transparent workflow for converting official landings into regional proportional species profiles, applying k-means clustering, and using explicit diagnostics to select the number of clusters. Second, it shows that the degree of regional compositional heterogeneity differs between the industrial and artisanal subsets: the industrial subset is summarised by a small number of broad composition types, whereas the artisanal subset requires a finer partition. Third, it reframes routine administrative landings as a basis for exploratory spatial comparison and hypothesis generation, rather than as direct evidence of biological stocks, causal governance mechanisms, or formal management units.
Because the workflow uses standard administrative aggregates by species, subsector, and region, it can be applied to other fisheries systems where comparable landings data are available. Its value lies in producing an auditable baseline of regional landing-composition types that can guide subsequent analyses incorporating biological, environmental, fleet, market, or institutional variables.
Data and methods
All steps were carried out independently for the industrial subsector, the artisanal subsector, and the aggregate total (industrial + artisanal + aquaculture). The spatial interpretation in the Results section focuses on the industrial-only and artisanal-only outcomes, since these subsectors operate under distinct regulatory regimes in Chile. The Metropolitan Region (R.M.) – whose reported ‘total landings’ are driven largely by freshwater and land-based aquaculture rather than marine extraction – was retained in the ‘total’ dataset for completeness but excluded from the coastal spatial maps to avoid misleading inland artefacts.
The empirical basis for the analysis consists of official landing statistics published by the Chilean National Fisheries and Aquaculture Service (SERNAPESCA, 2022, 2023, 2024). These records report annual landings by species, by subsector (industrial or artisanal), and by administrative region. The same figures sustain quota allocation, catch control and fleet enforcement in Chile, and they are used operationally by the fisheries authority to monitor high-biomass pelagic fisheries such as anchoveta and jack mackerel, as well as territorially managed benthic fisheries under AMERB in the artisanal subsector. The AMERB system has been described as one of the most extensive national-scale applications of territorial user-rights in benthic fisheries, in which exclusive access to specific benthic grounds and algal beds is granted to local artisanal unions. This territorial design has been characterised as a structural feature of recent Chilean artisanal fisheries, and it is closely linked to the specialisation of particular caletas in benthic invertebrates and commercially harvested macroalgae.
From these records, a multispecies signature was constructed for each administrative region. This signature is defined as the proportional composition of that region’s average reported landings over the 2022–2024 window, which corresponds to the post-pandemic period used here. The working assumption is that what defines a regional fishing system is not only total landed volume, but the mix of resources that sustains that activity. This reading is consistent with how Chilean fisheries practice already distinguishes broad pelagic macro-zones dominated by high-volume species such as anchoveta, Chilean sardine and jack mackerel – managed through national quotas and industrial–artisanal allocations – from austral fisheries in which high unit value species such as Patagonian toothfish or southern king crab are landed in relatively small spatial areas under more specific control regimes.
Exploratory data analysis
Data were cleaned prior to any clustering. Species names were first standardised to a single canonical form: case differences and diacritics were removed, and duplicate entries referring to the same taxon under minor orthographic variants were merged. Rows corresponding to administrative subtotals or overall totals were dropped, since they do not represent identifiable landings of a specific resource. Entries indicating absence of landings (e.g. blank cells or non-numeric markers) were explicitly set to zero, and all tonnage fields were then cast to numeric format to ensure comparability across years and regions.
From this cleaned base, a composition signature was generated for each administrative region. This signature is defined as the share contributed by each species to that region’s cumulative landings, calculated by pooling reported landings across 2022–2024. The interest lies not in absolute landed volume but in the relative species mix that characterises each region during the post-pandemic window. To prevent the classification from being driven by species that are effectively absent at the national scale, ultramarginal species were removed from the matrix. ‘Ultramarginal’ here refers to any species whose mean contribution to total national landings was below 0.5%. This dimensionality reduction is intended to stabilise the identification of regional patterns and to suppress noise in the clustering, without altering the species that actually sustain the fishery.
Determining the number of clusters
k-Means is an unsupervised clustering method that groups observations into a predefined number of clusters without requiring prior labels (MacQueen, Reference MacQueen1967). The algorithm assigns each observation to the nearest centroid and iteratively updates those centroids to minimise within-cluster variation. In this study, each observation corresponds to an administrative region represented by its proportional multispecies landing profile. Thus, two regions are assigned to the same cluster when their relative landing compositions are more similar to each other than to those assigned to other clusters. The list of species retained in the analysis, together with an operational ecological classification (pelagic, demersal, benthic), is provided in Supplementary Table S2.
The analysis operates in a compositional framework because each region is represented by relative species contributions rather than absolute tonnage. k-Means clustering was applied separately to the industrial, artisanal, and total subsets using these regional proportional landing profiles. Candidate values of k were evaluated over k = 2, …, 9, or up to the number of regions minus one when applicable.
The number of clusters was selected using average silhouette as the primary criterion and within-cluster inertia as an elbow diagnostic. The silhouette measures the degree of cohesion and separation of the resulting partition, whereas the inertia curve indicates whether additional clusters produce diminishing reductions in within-cluster dispersion. The value of k that maximised the average silhouette was adopted (Rousseeuw PJ, Reference Rousseeuw1987), while the elbow curve was used as a complementary diagnostic. These diagnostics were reported to support transparency and reproducibility, not to imply that the selected partition is the only possible representation of the data. Complete diagnostic curves are shown in Figures 1 and 2
Diagnosis of k selection for artisanal landings, 2022–2024. The left panel shows the average silhouette for each candidate value of k, and the right panel shows the elbow diagnostic based on within-cluster inertia. Regional landings were first converted into proportional multispecies profiles, and k-means clustering was applied to these profiles. For the artisanal subset, the average silhouette reached its maximum at k = 7; therefore, k = 7 was adopted as the final solution, with the elbow curve used as a complementary diagnostic.
Diagnosis of k selection for industrial landings, 2022–2024. The left panel shows the average silhouette for each candidate value of k, and the right panel shows the elbow diagnostic based on within-cluster inertia. Regional landings were first converted into proportional multispecies profiles, and k-means clustering was applied to these profiles. For the industrial subset, the average silhouette reached its maximum at k = 2; therefore, k = 2 was adopted as the final solution, with the elbow curve used as a complementary diagnostic.
.
Under this procedure, k = 2 was selected for the industrial subset, k = 7 for the artisanal subset, and k = 2 for total landings. These choices define regional landing-composition types derived from administrative catch data for 2022–2024 (SERNAPESCA, 2022, 2023, 2024). The clusters should be interpreted as empirical summaries of reported landing composition within each subsector. They are not biological stocks, ecological assemblages, causal governance effects, or formal management units. Potential drivers of the observed similarities, including fleet behaviour, fishing effort, environmental conditions, market demand, institutional regulation, and species distributions, are not directly modelled here and should be evaluated in subsequent analyses.
Results
Industrial subsector
The industrial subset is described by two regional landing-composition clusters. The first cluster includes most administrative regions along continental Chile, from the northern and central coast to the southern fjord zone. Regions assigned to this cluster share a broadly similar relative landing profile, characterised by the joint presence of high-volume pelagic species and other industrially landed resources.
The second industrial cluster is restricted to the far south. Its composition is strongly influenced by the relative contribution of Dissostichus eleginoides, which differentiates it from the broader continental cluster. The industrial partition should therefore be interpreted as a descriptive contrast between a widespread mixed industrial landing profile and a southern profile with high relative dominance of D. eleginoides, rather than as evidence of a newly identified spatial, biological, regulatory, or management boundary.
Table 1 shows that the industrial two-cluster solution is largely driven by differences in relative resource dominance. In particular, the southern industrial cluster is strongly influenced by the relative contribution of D. eleginoides. This supports interpreting the industrial partition as a descriptive contrast in landing composition rather than as evidence of clearly separated multispecies assemblages.
Main resources characterising each industrial landing-composition cluster
Table 1 Long description
Mean proportions of each resource within regional industrial landings across 2022 to 2024 are reported for two landing-composition clusters, I1 and I2. Cluster I2 is composed solely of Dissostichus eleginoides, accounting for all landings in that cluster. Cluster I1 is more mixed, led by Trachurus murphyi at about 37 percent. In I1, the next largest shares are Dissostichus eleginoides at about 13 percent, Scomber japonicus at about 9 percent, and Anchovy at about 8 percent. Additional I1 components include Southern hake at about 6 percent, Nylon shrimp at about 5 percent, Hoki at about 4 percent, Pomfret (reineta) at about 3 percent, and an Others category at about 3 percent. Overall, I2 is highly specialized compared with the diversified composition of I1. These values summarize relative landing composition and should not be taken as evidence of biological stock structure, management boundaries, or cause-and-effect relationships.
Values are mean proportions of each resource within regional industrial landings for 2022–2024. The table describes relative landing composition and should not be interpreted as evidence of biological stock structure, governance units, or causal mechanisms.
Artisanal subsector
The artisanal subset requires seven regional landing-composition clusters to describe the 2022–2024 period. Unlike the industrial subset, no single cluster label dominates the full coastline. Instead, the artisanal solution separates administrative regions into several composition types, each defined by differences in the relative contribution of the main landed resources.
The descriptive implication is that recent Chilean artisanal landings cannot be summarised as a single homogeneous compositional profile. They are better represented as multiple regional landing-composition types distributed along different stretches of the coast. This result indicates greater regional heterogeneity in relative species composition within the artisanal subset than within the industrial subset.
Table 2 shows that the artisanal clusters are characterised by different dominant resources and secondary species combinations. The table therefore supports interpreting the artisanal subset as compositionally more heterogeneous across regions. This heterogeneity remains descriptive and should not be read as evidence of distinct local production systems without additional data on effort, gear, markets, institutions, and ecological conditions.
Main resources characterising each artisanal landing-composition cluster
Table 2 Long description
Mean proportions of individual resources within regional artisanal landings are listed for clusters A1 through A7 across 2022 to 2024. Several clusters are strongly dominated by one resource: A3 is mainly anchovy at about 0.724, A1 is mainly jumbo flying squid at about 0.674, A7 is mainly pomfret at about 0.575, and A5 is mainly Chilean sardine at about 0.527. A2 is more mixed but led by sea urchin at about 0.365 and red seaweed at about 0.240, with smaller shares for a red seaweed variant at about 0.120 and southern king crab at about 0.080. A4 is led by cochayuyo kelp at about 0.376, followed by Chilean hake at about 0.147 and kelp at about 0.112, with additional contributions from jumbo squid and Spanish mackerel near 0.078 and 0.072. A6 is kelp-focused at about 0.352, with notable secondary shares for other species at about 0.140, anchovy at about 0.115, kelp stipe at about 0.104, and jack mackerel at about 0.099. Across clusters, kelp or kelp-related categories recur as secondary components, while the other species category ranges from small to moderate, highest in A6 at about 0.140. Values are averages over multiple years and represent composition within each cluster, so they summarize typical mixes rather than totals or year-to-year variability.
Values are mean proportions of each resource within regional artisanal landings for 2022–2024.
Sectoral comparison
Taken together, the industrial and artisanal subsets differ in the degree of regional compositional heterogeneity represented by the clustering procedure. The industrial subset is summarised by a two-cluster partition, whereas the artisanal subset requires seven clusters to represent variation in relative species composition across administrative regions (see Figure 3).
Regional landing-composition cluster assignments for industrial and artisanal landings, 2022–2024. Panel A shows industrial clusters (I1–I2), and Panel B shows artisanal clusters (A1–A7). Points represent approximate administrative-region centroid locations, and colours indicate cluster membership within each subsector. The figure summarises regional landing-composition types rather than biological stocks, fishing grounds, governance mechanisms, or formal management units.
Figure 3 Long description
A) Title text at top reads, comma, Regional landing composition cluster assignments, comma, 2022 to 2024. Subheading reads, comma, A. Industrial landings. Tall, narrow map frame with a long north to south coastal outline and internal boundary lines. Circular point markers are placed along the coast from near the top to near the bottom, plus one marker inland. Legend at lower left reads, comma, Cluster, with entries, comma, I1 and I2. Markers use two distinct fill styles corresponding to I1 and I2. B) Subheading reads, comma, B. Artisanal landings. Same map frame, coastal outline and internal boundary lines as A. More circular point markers are placed along the coast from near the top to near the bottom, plus one marker inland. Legend at lower left reads, comma, Cluster, with entries, comma, A1, comma, A2, comma, A3, comma, A4, comma, A5, comma, A6, comma, A7. Markers use seven distinct fill styles corresponding to A1 to A7.
This sectoral contrast is quantitative rather than causal. It indicates that, under the same compositional workflow and the same cluster-selection criteria, the artisanal subset displays a more differentiated regional landing-composition structure than the industrial subset. The result should not be read as evidence that the two subsectors independently generate different spatial systems, since their organisation is already shaped in part by Chile’s regulatory and administrative framework.
Discussion
The main contribution of this study is to quantify regional variation in multispecies landing composition within Chile’s artisanal and industrial subsectors using a common compositional workflow. The analysis does not seek to demonstrate that the two subsectors are spatially or institutionally different. That distinction is already embedded in Chile’s fisheries system through differentiated access rules, reporting categories, management arrangements, and landing certification procedures (SERNAPESCA, 2022, 2023, 2024). In the artisanal subsector, territorial access arrangements such as AMERB have also been described as an important institutional feature of Chilean coastal fisheries (Gelcich et al., Reference Gelcich, Hughes, Olsson, Folke, Defeo, Fernández, Simon, Gunderson, Rodríguez-Sickert, Scheffer, Steneck and Castilla2010; Aburto et al., Reference Aburto, Gallardo, Stotz, Cerda, Mondaca-Schachermayer and Vera2013).
Within this context, the clustering results are best interpreted as regional landing-composition types. For the industrial subset, the two-cluster solution provides a compact summary of reported landings. One cluster groups most continental regions and is characterised by a mixed profile in which high-volume pelagic species coexist with demersal, crustacean, and slope resources reported by the industrial fleet. The other cluster is restricted to the far south and is strongly influenced by the relative dominance of D. eleginoides. This pattern suggests that the industrial partition mainly captures a dominant compositional contrast: a widespread mixed industrial profile versus a southern profile largely defined by the relative weight of D. eleginoides.
The artisanal subset shows a more differentiated regional composition pattern. The seven-cluster solution indicates that a single national artisanal profile is insufficient to summarise the relative species composition of reported artisanal landings during 2022–2024. Instead, the artisanal subset is better represented by several regional composition types, each defined by different dominant resources and secondary species combinations. This greater differentiation is consistent with the localised and territorially structured nature of Chilean artisanal fisheries, but the present analysis cannot determine the specific mechanisms behind the observed patterns. Explaining them would require additional information on fishing effort, gear, fisher organisations, markets, ecological conditions, and local institutions.
The contrast between the industrial and artisanal subsets should therefore be understood as a difference in compositional heterogeneity. Under the same preprocessing, proportional representation, and clustering criteria, industrial landings are summarised by a small number of broad regional composition types, whereas artisanal landings require a finer partition to represent variation in relative species composition across administrative regions. This asymmetry provides a quantitative description of reported landing profiles within subsectors that are already administratively and institutionally differentiated.
This distinction is important because regional landing compositions can be shaped by multiple, overlapping drivers. Similar reported species profiles may arise from ecological gradients, fishing effort, gear selectivity, market demand, access rights, reporting practices, or combinations of these factors. Conversely, differences in landing composition do not necessarily imply distinct biological assemblages or governance regimes. The clusters should therefore be treated as exploratory empirical summaries that help identify where further explanation is needed.
The value of the workflow lies in making regional landing composition comparable, transparent, and auditable. By converting species-by-region landing tables into regional proportional profiles and selecting the number of clusters with explicit diagnostics, the study provides a reproducible way to summarise multispecies administrative landings (SERNAPESCA, 2022, 2023, 2024). The resulting typologies support comparative description among regions and can guide future analyses that integrate landings with biological, environmental, economic, fleet, or institutional data.
Overall, the study provides an exploratory compositional baseline for Chilean fisheries landings during 2022–2024. Its main finding is that industrial and artisanal landings differ not only because they belong to institutionally distinct subsectors but also because their regional landing compositions exhibit different degrees of internal heterogeneity. Future work should evaluate the mechanisms underlying these patterns by incorporating fishing effort, gear, fleet behaviour, market conditions, environmental variables, and institutional information.
Generalisability to other fisheries
This pipeline is not specific to Chilean taxa or regulations. It requires only aggregated landings or catches by species and administrative unit for a defined time window, which many fisheries routinely report in statistical bulletins. The workflow can therefore be reproduced in other jurisdictions by building a species-by-area matrix, converting rows to proportional multispecies profiles, applying k-means clustering, and selecting the number of clusters with explicit diagnostics based on inertia and average silhouette.
The resulting typologies can support comparative description across regions and help identify where additional biological, environmental, fleet, market, or institutional data are needed before management interpretation is attempted.
Limitations and extensions
The regional coordinates shown in the maps are representative coastal points; they are not intended to reproduce full administrative polygons. A more spatially explicit rendering could incorporate official shapefiles of regional boundaries and port locations, which is technically feasible using the same tools applied here.
Repeating the exercise at the port level, rather than at the administrative-region level, would capture sub-regional structure that is currently averaged out. A sliding multi-year window (e.g. 5-year moving averages) would also make it possible to track shifts in cluster membership through time, rather than treating 2022–2024 as a single static snapshot.
A further extension would link clusters to gear type and post-landing destination (reduction to fishmeal and fish oil, fresh market, frozen product). That linkage would support a more explicit economic–ecological reading of each cluster, taking advantage of existing national reports that already distinguish between pelagic volume fisheries, demersal fisheries and deep-water southern fisheries.
Intraregional heterogeneity (e.g. differences between ports within the same administrative region) is not resolved by this approach.
Because the analysis is exploratory and based on a small number of administrative regions, future work should evaluate the stability of cluster assignments under alternative algorithms, including hierarchical clustering, and under different thresholds for retaining low-contribution species.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0025315426101775.
Acknowledgements
This manuscript was prepared from a standardised cleaning of SERNAPESCA landing tables and a fully reproducible Python/Colab workflow. Spatial figures were generated using Python geospatial libraries. Any minor discrepancies in regional place names or point coordinates do not affect cluster assignments, which are determined by multispecies composition and not by exact georeferencing. All data used are publicly reported regional landings by species and subsector, available from Chile’s National Fisheries and Aquaculture Service (SERNAPESCA) for 2022–2024. During the preparation of this manuscript, I used artificial intelligence (AI) tools (specifically, language models) to assist with minor improvements in grammar, phrasing, and clarity of expression. These tools were not used for data analysis, interpretation, or generation of scientific content. All intellectual contributions, interpretations, and conclusions are solely my own. The use of AI was limited to editorial support and aligns with the journal’s policy on responsible AI use.
Declaration of interest
The author declares no competing interests.
Data availability statement
The data analysed in this study are publicly available from the Servicio Nacional de Pesca y Acuicultura (SERNAPESCA), in its statistical yearbooks for 2022–2024: https://www.sernapesca.cl/informacion-utilidad/anuarios-estadisticos-de-pesca-y-acuicultura/. The study uses aggregated landings by species, subsector and administrative region. The analysis code is available upon request from the author. The Python scripts used for the analysis are available upon request from the author. Interested parties should contact the author directly; please allow for a reasonable response time.
Ethics and approach
This study analysed publicly available, aggregated fisheries statistics. It did not involve the recruitment of human participants or experimental procedures involving animals.
Funding statement
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
