Marine Protected Area Research and Global Governance Priorities: Thematic Shifts, Geographic Inequalities, and Science–Policy Alignment

Abstract

Marine Protected Areas (MPAs) are central to global ocean conservation, yet the latent thematic organization of three decades of MPA research has not been systematically characterized. This study applies BERTopic, a transformer-based neural topic modeling framework, to 10,735 peer-reviewed publications indexed in Scopus and Web of Science (1992–2025). Eight semantically coherent research themes were identified, and their structural relationships were quantified using the Relative Importance Factor (RIF) Index, a rank-based dominance measure derived from discrete power-law modeling. Marine Conservation Governance and Marine Biodiversity Patterns together account for 64.3% of the scientific output, forming the cognitive core of the field. Temporal analysis reveals that marine pollution, megafauna movement ecology, and Chinese coastal governance are the fastest-growing themes, yet they remain structurally peripheral, a phenomenon described here as accelerated marginality. Geographic analysis further identifies a pronounced epistemic governance gap, whereby countries hosting many of the world’s most valuable marine ecosystems contribute disproportionately little to the knowledge base guiding global conservation policy. Taken together, these findings carry a policy implication: achieving the objectives of the Kunming–Montreal Global Biodiversity Framework will require not only expanding protected areas, but also rebalancing the thematic and geographic architecture of the research that underpins them.

1. Introduction

Aquatic ecosystems play a critical role in sustaining global biodiversity and providing essential ecosystem services, including climate regulation, food security, and support for local and global economies [1,2]. Among the strategies developed to address mounting anthropogenic pressures on these systems, Marine Protected Areas (MPAs) stand out as the primary policy instrument for safeguarding marine species, habitats, and ecosystem services [3,4]. From their institutionalization at the 1992 Earth Summit in Rio de Janeiro and the Convention on Biological Diversity (CBD), through the 2002 Johannesburg Summit, the 2010 Aichi Biodiversity Targets, and SDG 14, to the 2022 Kunming–Montreal Global Biodiversity Framework and its 30 × 30 commitment to protect at least 30% of the world’s oceans by 2030 [5], successive policy milestones have catalyzed coordinated scientific production across disciplines and regions, leaving measurable imprints on the thematic priorities, geographic focus, and methodological approaches of MPA research [6].

Despite this sustained growth, the scientific literature on MPAs remains fragmented, geographically uneven, and in many cases too specialized to inform decision-makers effectively. Prior bibliometric studies have begun to map this landscape, analyzing current status and trends with a primary focus on conservation aspects [7], and examining the transition from conservation-centric goals toward broader commitments to global ocean sustainability [8]. However, these studies remain regionally or thematically constrained, and none has systematically examined the latent semantic structure of the entire field, that is, the hidden thematic architecture that organizes how researchers think, frame problems, and produce knowledge about MPAs across time and geography.

To address this gap, this study applies BERTopic [9], a state-of-the-art neural topic modeling approach based on transformer-based language representations, to a corpus of 10,735 peer-reviewed documents on MPAs published between 1992 and 2025. Unlike traditional approaches such as Latent Dirichlet Allocation (LDA) [10], BERTopic leverages contextual word embeddings to capture nuanced semantic relationships between documents, enabling more precise identification of thematic clusters and their evolution over time a capacity particularly suited to a field as multidisciplinary and rapidly evolving as MPA research. The study addresses the following central question: How has the thematic structure of MPA science evolved since 1992, and what latent topics, temporal dynamics, and geographic asymmetries characterize its cognitive architecture? Specifically, it identifies the field’s dominant, emerging, and declining research themes; quantifies their relative structural prominence using the Relative Importance Factor (RIF) Index [11]; maps their temporal trajectories against key international policy milestones; and examines geographic inequalities in thematic knowledge production providing an evidence-based foundation for science-policy alignment in the context of the 30 × 30 agenda.

2. Materials and Methods

2.1. Data Collection and Search Strategy

The bibliographic corpus was constructed following a retrieval protocol consistent with PRISMA reporting principles [12]. A systematic search was conducted across Web of Science (WoS) Core Collection and Scopus, querying the title, abstract, and author keywords fields using the term “marine protected area*” (Scopus: TITLE-ABS-KEY; WoS: TS), covering the period January 1992 to December 2025. The year 1992 was adopted as the starting point because the Earth Summit in Rio de Janeiro and the subsequent Convention on Biological Diversity (CBD) marked the first institutionalization of quantitative marine protection targets at the global level. Inclusion criteria restricted the corpus to peer-reviewed journal articles and reviews with non-empty title and abstract fields; book chapters, conference proceedings, editorials, and letters were excluded. Records from both databases were merged and deduplicated on the basis of DOI, title, and author fields. The complete PRISMA flow diagram is provided in Figure 1.

2.2. Descriptive Characterization of the Bibliographic Corpus

Descriptive bibliometric characterization of the corpus was conducted using the bibliometrix package [13] in R [14], computing annual publication counts, journal distribution, and geographic output by country. Geographic attribution followed the corresponding-author convention, which most reliably reflects primary research coordination in collaborative studies [15]. No citation-based indicators or network analyses were performed at this stage.

2.3. Transformer-Based Topic Modeling with BERTopic

Topic modeling was performed using BERTopic [9], which integrates transformer-based embeddings, dimensionality reduction, density-based clustering, and c-TF-IDF to identify semantically coherent topics from large corpora, outperforming traditional approaches such as LDA and NMF in semantic coherence for heterogeneous scientific corpora [16,17].

Each document was represented as a concatenated title-abstract pair and encoded into a high-dimensional semantic vector using SciBERT (allenai/scibert_scivocab_uncased), a transformer-based language model pre-trained on a large corpus of scientific publications [18] via the sentence-transformers library [19]; its domain-specific pre-training enables more accurate capture of specialized terminology relative to general-purpose models. Embeddings were reduced using UMAP [20] and clustered using HDBSCAN [21]; all hyperparameter configurations were selected through a systematic grid search and are fully documented in Supplementary Materials S1 and S2. Topic coherence was evaluated using the Cv metric [22] and validated through qualitative inspection by the research team.

To quantify the hierarchical organization of identified topics, the Relative Importance Factor (RIF) Index [11] was applied to the ranked topic-size distribution, computing pairwise dominance values and grouping results into five interpretive categories (Stable, Moderate, Significant, Critical, and Dominant). Temporal trends in annual topic prevalence were assessed by combining ordinary least squares regression with the Mann–Kendall test (alpha = 0.05), with key international policy milestones superimposed as temporal reference markers. Geographic topic profiles were constructed by cross-tabulating topic assignments with corresponding-author country (n ≥ 20 publications), using the World Bank income classification to characterize North–South asymmetries in thematic knowledge production.

3. Results and Discussion

3.1. Bibliographic Corpus Characterization

The systematic search and screening process yielded a final corpus of 10,735 peer-reviewed documents published across 1257 sources between 1992 and 2025 (

Table 1. General bibliometric indicators of the MPA scientific corpus (1992–2025).

Indicator	Value
Timespan	1992–2025
Sources (journals)	1257
Documents	10,735
Annual growth rate (%)	20.34
Mean document age (years)	8.76
Document type
—Original articles	10,081
—Review articles	654

), comprising 10,081 original research articles (93.9%) and 654 review articles (6.1%). Annual publication counts reveal a sustained upward trajectory over the three decades analyzed (Figure 2). Output remained modest during the 1990s, with fewer than 30 publications per year through 1999, before entering a phase of progressive expansion coinciding with the reinforcement of international marine protection commitments at the 2002 Johannesburg Summit. A second, more pronounced growth phase is observed from 2007 onward, with annual output surpassing 600 publications by 2017, a period bracketed by the adoption of the Aichi Biodiversity Targets (2010) and SDG 14 (2015). The corpus reached its highest annual output in 2025 (n = 900), with an overall annual growth rate of 20.34%, consistent with the hypothesis that successive global policy milestones catalyze scientific production in conservation-related fields.

The geographic distribution of scientific production reveals pronounced asymmetries in research output (Figure 3). The United States (n = 1975; 18.4%), Australia (n = 1097; 10.2%), and the United Kingdom (n = 924; 8.6%) together account for more than one-third of the global corpus. In contrast, large portions of Africa, the Pacific, and Southeast Asia, regions encompassing some of the world’s most ecologically significant marine ecosystems, exhibit minimal scientific output in the indexed literature. This geographic imbalance has direct implications for the representativeness of knowledge underpinning global MPA governance, a dimension examined in depth in Section 3.5.

3.2. The Latent Thematic Architecture of MPA: Topic Identification and Macro-Domain Structure

BERTopic identified eight semantically coherent thematic topics from the corpus of 10,735 documents, with only 13 records (0.12%) remaining unclassified as outliers, an exceptionally low noise rate that reflects both the thematic coherence of the MPA literature and the effectiveness of the SciBERT-UMAP-HDBSCAN pipeline. The complete topic inventory is presented in

Table 2. Thematic structure of MPA research (1992–2025) as identified by BERTopic transformer-based topic modeling. Topics are ordered by decreasing document prevalence. Keywords are ranked by c-TF-IDF score within each topic.

Topic	Thematic Label	N	%	Representative Keywords
T0	Marine Conservation Governance	4039	37.6	management, conservation, MPAs, fisheries, coastal, biodiversity, local, social, fishing
T1	Marine Biodiversity Patterns	2867	26.7	species, fish, sea, abundance, coral, distribution, habitat, diversity, reef, Mediterranean
T7	Coral Reef Fish Ecology	803	7.5	coral, fish, reef, species, fishing, biomass, abundance, effects, cover
T2	Fisheries and Spatial Management	946	8.8	fishing, fisheries, management, spatial, fish, reserves
T3	Shark and Megafauna Movement Ecology	877	8.2	foraging, habitat, sharks, acoustic, movements, turtles, conservation
T4	Genetic Connectivity and Larval Dispersal	577	5.4	connectivity, genetic, dispersal, larval, populations, larvae, diversity
T6	Coastal Ecosystem Governance in China	322	3.0	China, coastal, ecosystem, development, ecological, services, national
T5	Marine Pollution and Sediment Contamination	291	2.7	sediments, plastic, pollution, contamination, coastal, litter, debris
—	Unclassified (outliers)	13	0.1	—

. The distribution of scientific attention across the eight topics is markedly unequal: the two most prevalent topics collectively account for 64.3% of the corpus, while the remaining six divide the residual 35.7%. This concentration reflects a fundamental structural property of the field whose internal hierarchy is further quantified in Section 3.3. The intertopic distance map and hierarchical clustering (Figure 4 and Figure S1) reveal that the eight topics organize into three spatially coherent macro-domains, each grouping thematically related clusters and collectively structuring the remainder of the analysis.

Macro-domain I: Conservation Governance and Fisheries Management (49.4%). This macro-domain encompasses Topics 0, 2, and 6. Topic 0 (Marine Conservation Governance; n = 4039; 37.6%) constitutes the single largest thematic cluster and the gravitational center of the field. Its representative keywords, including management, conservation, MPAs, fisheries, coastal, biodiversity, local, social, and fishing, reveal a fundamentally integrative discourse that bridges ecological imperatives with institutional, social, and policy dimensions, signaling that MPA science is not, at its epistemic core, a purely biological enterprise [23]. Topic 6 (Coastal Ecosystem Governance in China; n = 322; 3.0%) merges with T0 at the shortest hierarchical linkage distance (d = 0.22; Figure S1), suggesting that China’s state-led conservation model is not epistemically peripheral but structurally embedded within the mainstream governance paradigm, a finding that anticipates the geographic asymmetries examined in Section 3.5. Topic 2 (Fisheries and Spatial Management; n = 946; 8.8%) completes this macro-domain, reflecting the persistent intersection of resource extraction and spatial conservation planning that has been central to MPA science since the Convention on Biological Diversity [6].

Macro-domain II: Marine Biodiversity and Emerging Environmental Pressures (29.4%). This macro-domain groups Topics 1 and 5. Topic 1 (Marine Biodiversity Patterns; n = 2867; 26.7%) constitutes the primary ecological knowledge base of MPA research, reflecting a broad comparative tradition of biodiversity assessment with a notable Mediterranean signature consistent with the long-standing European focus in MPA ecological research [7,8]. Topic 5 (Marine Pollution and Sediment Contamination; n = 291; 2.7%), while spatially proximate to T1, is structurally isolated in the hierarchical clustering solution (d = 0.75; Figure S1), confirming it as an autonomous emergent research front whose growth reflects a growing recognition that protected area effectiveness cannot be evaluated independently of the cumulative stressor landscape affecting marine ecosystems, an agenda whose expansion corresponds to the period following SDG 14 in 2015 [3].

Macro-domain III: Tropical Marine Ecology and Movement Science (21.1%). This macro-domain groups Topics 3, 4, and 7, collectively representing the specialized ecological and methodological science underpinning MPA design in tropical systems. Topic 7 (Coral Reef Fish Ecology; n = 803; 7.5%) captures the empirical literature on reef fish assemblages and trophic recovery within protected areas, consolidated by landmark global syntheses demonstrating substantially greater fish biomass and diversity in fully protected reefs [24]. Topic 3 (Shark and Megafauna Movement Ecology; n = 877; 8.2%) reflects the rapid expansion of acoustic telemetry and biologging methods, documenting movement patterns of apex predators and sea turtles whose home ranges routinely exceed the spatial extent of individual MPAs, with direct implications for transboundary governance. Topic 4 (Genetic Connectivity and Larval Dispersal; n = 577; 5.4%) constitutes the most methodologically specialized cluster, providing the quantitative basis for evidence-based MPA network design across broader seascape scales.

Taken together, the tripartite thematic architecture identified by BERTopic reveals a field organized around a dominant governance discourse, a robust ecological knowledge base, and a constellation of specialized methodological frontiers, encoding more than three decades of intellectual evolution, policy pressure, and disciplinary negotiation. The degree to which scientific attention is concentrated within this architecture and the structural implications of that concentration are quantified in the following section.

3.3. Thematic Dominance Structure of Marine Protected Areas Research

The ranked topic distribution was analyzed using a power-law framework and the Relative Importance Factor (RIF) Index [11]; bootstrap-based results are reported throughout, as they provided greater robustness to parameter uncertainty than alternative fitting procedures. The fitted power-law model yielded a scaling exponent of alpha = 1.289 (R² = 0.897), confirming a strongly hierarchical allocation of scientific attention (Figure 5). Conservation Governance (T0) emerged as the dominant topic with 4039 documents, followed by Biodiversity Patterns (T1) with 2867 documents; together they account for 64.3% of all topic assignments. Fisheries Management (T2), Shark Ecology (T3), and Coral Reef Fish Ecology (T7) formed a secondary tier with considerably lower frequencies, while Marine Pollution (T5), Chinese Coastal Governance (T6), and Genetic Connectivity (T4) occupy the most peripheral positions. Complete pairwise RIF values and dominance classifications are provided in Supplementary Table S4.

The RIF matrix reveals a highly asymmetric dominance structure centered on T0 (Figure 6). The dominance gap between T0 and T1 was classified as Critical (RIF = 8.5), whereas the gaps between T0 and T2, T3, and T7 reached Dominant levels, with RIF values of 30.5, 71.2, and 142.3, respectively. T1 occupied an intermediate bridging position, maintaining Critical or Dominant relationships with lower-ranked topics, including Fisheries Management (RIF = 3.6), Shark Ecology (RIF = 8.3), and Coral Reef Fish Ecology (RIF = 16.7). Among peripheral topics, dominance gaps were considerably smaller; the T3-T7 relationship, for instance, was classified as Moderate (RIF = 2.0). The RIF network corroborates this tripartite structure, with T0 forming a dominant governance core, T1 acting as an intermediate connector, and the remaining topics constituting a progressively weaker peripheral cluster (Figure 7). Critically, no single rank transition constitutes a discontinuous dominance break; structural peripherality instead accumulates gradually through compounding rank distances from the governance-ecology core, a property with direct implications for the emergence of new research agendas examined in Section 3.6.

3.4. Temporal Dynamics: Science Following Policy

The two-stage analytical procedure identified four topics exhibiting increasing trends, two classified as declining, and two exhibiting stable trajectories (

Table 3. Temporal trend analysis of MPA research themes (1992–2025). OLS β = linear regression slope (annual change in topic prevalence, ×10⁻³). R² = coefficient of determination. MK τ = Mann–Kendall tau statistic. Trend classification based on convergent evidence from both procedures at α = 0.05: Increasing/Declining = both tests significant and directionally concordant; Declining * = one test significant; Stable = no convergent evidence. Mean prevalence computed over years with ≥1 document assigned to the topic.

Topic	Thematic Label	Mean Prevalence (%)	OLS β (×10−3)	OLS R2	p (OLS)	MK τ	p (MK)	Trend
T0	Marine Conservation Governance	43.7	−976.98	0.297	0.001	−0.057	0.646	Declining *
T1	Marine Biodiversity Patterns	25.4	+593.94	0.411	<0.001	+0.439	0.001	Increasing
T2	Fisheries and Spatial Management	12.0	−95.95	0.025	0.373	−0.330	0.006	Declining *
T3	Shark and Megafauna Movement Ecology	7.3	+320.34	0.777	<0.001	+0.704	<0.001	Increasing
T4	Genetic Connectivity & Larval Dispersal	6.9	−48.46	0.003	0.745	+0.412	0.001	Stable
T5	Marine Pollution & Sediment Contam.	2.3	+124.04	0.475	<0.001	+0.631	<0.001	Increasing
T6	Coastal Ecosystem Gov. in China	2.7	+115.42	0.513	<0.001	+0.545	<0.001	Increasing
T7	Coral Reef Fish Ecology	10.7	−32.35	0.002	0.797	−0.121	0.319	Stable

; Figure 8). The most structurally consequential pattern concerns Topic 0 (Marine Conservation Governance), which despite constituting 37.6% of all classified documents exhibits a statistically significant negative linear slope (OLS: beta = −976.98 × 10⁻³, R² = 0.297, p = 0.001) not corroborated by the Mann–Kendall test (tau = −0.057, p = 0.646), yielding a Declining classification. This reflects a nuanced phenomenon: T0 is not losing documents in absolute terms, but its relative share of annual scientific production has been progressively diluted as more specialized research traditions consolidated their presence, signaling the maturation of the field toward a more pluralistic thematic structure. Topic 2 (Fisheries and Spatial Management) follows a similar pattern of relative contraction (MK: tau = −0.330, p = 0.006; OLS: beta = −95.95 × 10⁻³, p = 0.373; Declining), suggesting that the fisheries-MPA nexus is gradually being absorbed into the broader governance discourse or superseded by more ecologically specialized approaches to spatial conservation planning.

Among the increasing trajectories, Topic 3 (Shark and Megafauna Movement Ecology) exhibits the strongest and most statistically robust trend in the corpus (OLS: beta = +320.34 × 10⁻³, R² = 0.777, p < 0.001; MK: tau = +0.704, p < 0.001), representing an almost linear ascent from near-zero prevalence in the mid-1990s to sustained prominence in the post-2010 period. This inflection aligns closely with the adoption of the Aichi Biodiversity Targets in 2010, which created institutional demand for science capable of identifying where wide-ranging species require protection across spatial scales that transcend individual MPA boundaries [4,23]. Topic 1 (Marine Biodiversity Patterns) exhibits a similarly consistent increasing trajectory (OLS: beta = +593.94 × 10⁻³, R² = 0.411, p < 0.001; MK: tau = +0.439, p < 0.001), reflecting the sustained expansion of ecological monitoring and quantitative biodiversity assessment within MPAs [23].

The two most peripheral topics in the dominance hierarchy are simultaneously the two fastest-growing, a finding of particular structural significance. Topic 5 (Marine Pollution and Sediment Contamination) shows both a strong linear and monotonic increase (OLS: beta = +124.04 × 10⁻³, R² = 0.475, p < 0.001; MK: tau = +0.631, p < 0.001), with a trajectory that visibly accelerates following SDG 14 in 2015, when ocean health was formally embedded within the global sustainable development agenda [3,25]. Topic 6 (Coastal Ecosystem Governance in China) presents an equally robust increasing trend (OLS: beta = +115.42 × 10⁻³, R² = 0.513, p < 0.001; MK: tau = +0.545, p < 0.001), reaching its highest annual prevalence in 2022–2023, coinciding with China’s intensified domestic MPA expansion under the Kunming–Montreal Global Biodiversity Framework [5,23]. Topics 4 (Genetic Connectivity) and 7 (Coral Reef Fish Ecology) exhibit stable trajectories, reflecting not stagnation but consolidation as established methodological and ecological pillars of MPA science.

Taken together, these dynamics reveal a field structurally responsive to international governance signals: the post-Aichi surge in megafauna ecology, the post-SDG 14 acceleration of pollution research, and the post-30 × 30 consolidation of Chinese governance science collectively suggest that global biodiversity frameworks actively reshape scientific agendas by creating incentive structures that redirect research attention toward policy-relevant frontiers.

3.5. Geographic Asymmetries in MPA Knowledge Production: Thematic Specialization and Epistemic Inequalities

The geographic distribution of MPA scientific production is not merely unequal in volume; it is structurally differentiated in thematic content. Cross-tabulation of BERTopic assignments with corresponding-author country reveals that the 51 countries meeting the analytical threshold of n ≥ 20 publications exhibit markedly distinct research profiles, suggesting that what is studied about MPAs is systematically shaped by where that science is produced (Figure 9).

The most structurally consequential finding concerns the divergence in thematic profiles between the Global North and Global South. The United States (n = 1972), Australia (n = 1096), and the United Kingdom (n = 922), together accounting for 37.3% of the valid corpus, share a broadly similar thematic profile dominated by Topic 0 (Marine Conservation Governance: 36–47%) alongside substantial contributions to Topics 2 and 3, reflecting a shared tradition of applied, policy-oriented marine science driven by management effectiveness evaluation, spatial planning, and biotelemetry-based ecology [6,23]. High-income countries collectively allocate a significantly larger proportion of their research to Topic 2 (Fisheries and Spatial Management; 10.7% vs. 4.3%) and Topic 3 (Megafauna Movement Ecology; 9.1% vs. 5.3%), reflecting the methodological infrastructure that resource-intensive research paradigms require. By contrast, Global South countries devote a substantially higher proportion of their output to Topic 1 (Marine Biodiversity Patterns; 35.7% vs. 23.6%), Topic 5 (Marine Pollution; 4.5% vs. 2.0%), and Topic 6 (Coastal Ecosystem Governance in China; 8.0% vs. 1.4%). This divergence encodes a deeper asymmetry in the research questions that different scientific communities are positioned, resourced, and incentivized to pursue [26,27].

Several country-level specializations merit interpretive attention. Italy and Spain exhibit exceptionally high concentrations in Topic 1 (56.0% and 48.0%, respectively), reflecting the long-standing Mediterranean tradition of biodiversity assessment, mirrored in Latin America by Argentina (T1: 64.8%) and Cuba (T1: 66.7%), suggesting that MPA science in these regions is primarily anchored in ecological documentation rather than governance effectiveness research, with important implications for science-policy translation. China and Taiwan present the most distinctive national signature: both allocate the largest share of their publications to Topic 6 (33.5% and 43.1%, respectively), organized around state-led conservation models, ecological red-line zoning, and national marine park legislation. The near-absence of T6 contributions from non-Asian countries confirms that Chinese MPA governance science operates within a largely self-referential epistemic framework, structurally decoupled from the dominant Anglophone governance discourse. Brazil, the largest Global South contributor, allocates 11.7% of its publications to Topic 5, the highest proportion among all major producing nations, reflecting research imperatives driven by proximity to industrial and agricultural pollution sources that resonate with broader concerns about environmental justice in marine conservation [27].

The underrepresentation of African, Pacific, and Southeast Asian countries constitutes the most structurally significant gap for the 30 × 30 agenda. Tanzania (n = 41), Kenya (n = 37), Senegal (n = 26), and Fiji (n = 29) collectively produce fewer publications than a single mid-sized European research institution, yet their coastal and reef systems encompass a disproportionate share of global marine biodiversity. The Kunming–Montreal Global Biodiversity Framework’s commitment to equitable governance and benefit-sharing [23] demands not only expanded area coverage but a fundamental rebalancing of who produces the scientific knowledge that legitimizes and evaluates that coverage, a rebalancing that the current geographic distribution of MPA research has not achieved.

Taken together, the geographic analysis reveals that the global MPA knowledge system is epistemically stratified in its thematic priorities: the questions deemed most worthy of scientific investment, the methods deployed to answer them, and the governance challenges accorded greatest urgency all vary systematically with national income level and geopolitical positioning, a stratification that the aggregate dominance metrics of the RIF analysis alone cannot capture but that country-level topic profiles make structurally visible. The synthesis of these thematic, temporal, and geographic dimensions into an integrated science-policy framework is the subject of the following section.

3.6. Science-Policy Misalignment in the Global MPA Knowledge System

Three related but distinct forms of misalignment emerge from the integrated analysis, corresponding to three different axes of comparison: a thematic axis (science-policy gap: which topics concentrate epistemic authority versus which topics governance urgently needs—Section 3.3), a temporal axis (implementation gap: the lag between a policy milestone and its detectable effect on research output—Section 3.4), and a geographic axis (epistemic governance gap: where knowledge is produced versus where conservation value is concentrated—Section 3.5). These are not synonymous, though they compound one another: a topic can be simultaneously peripheral in dominance, slow to respond to policy, and geographically concentrated in a small set of countries. The distinctions among these three forms of misalignment, their respective axes of comparison, and the corresponding empirical evidence are summarized in

Table 4. Conceptual distinctions among the three forms of science–policy misalignment identified in the global MPA knowledge system.

Term	Axis of Comparison	Evidence in the Manuscript
Science-policy gap	Which topics concentrate epistemic authority vs. which topics governance urgently needs	Section 3.3 (RIF dominance)
Implementation gap	Lag between a policy milestone and its detectable effect on research output	Section 3.4 (temporal trends, Figure 8)
Epistemic governance gap	Where knowledge is produced vs. where conservation value is concentrated	Section 3.5 (geographic profiles, Figure 9)

.

The integrated analysis of thematic structure, dominance hierarchy, temporal dynamics, and geographic distribution converges on a finding that transcends any individual analytical dimension: the global MPA knowledge system exhibits a structural misalignment between the topics that concentrate epistemic authority and those most urgently required to meet the governance demands of the Kunming–Montreal Global Biodiversity Framework. This misalignment is systematically organized along axes of disciplinary inertia, methodological resource asymmetry, and geopolitical inequality in scientific production, constituting a science-policy gap at the level of knowledge architecture rather than merely at the level of individual research priorities.

The RIF analysis established that T0 and T1 together absorb 64.3% of the field’s documented scientific attention, forming a cognitive core from which all other topics are structurally subordinate. Rather than identifying a sharp boundary between central and peripheral topics, the dominance structure exhibits a progressive accumulation of thematic marginalization, whereby structural distance increases gradually across successive rank positions, creating a layered hierarchy of epistemic influence. Yet the topics whose growth rates are most pronounced, namely Shark and Megafauna Movement Ecology (R² = 0.777), Marine Pollution (R² = 0.475), and Coastal Ecosystem Governance in China (R² = 0.513), remain in the Dominant-tier periphery, separated from the epistemic core by structural gaps of 10- to 15-fold in relative prominence. The convergence of rapid growth and persistent marginalization defines the central paradox of MPA science at the threshold of the 30 × 30 era: the field is moving in the right directions, but not fast enough, and not from a knowledge base broad enough to generate the scientific density that transformative governance requires. The persistence of these peripheral yet rapidly expanding topics constitutes evidence of an implementation gap, a structural disconnect between the emerging conservation priorities articulated in international policy frameworks and the dominant scientific paradigms that continue to organize MPA governance decisions [28,29].

The temporal correspondence between policy milestones and research trajectories offers cautious evidence for science-policy responsiveness, but this responsiveness is structurally delayed and unevenly distributed. The lag between the adoption of a policy milestone and its translation into a detectable shift in research prevalence suggests that the mechanisms linking governance aspirations to scientific agendas operate through slow-moving institutional channels: funding reallocation, curriculum reform, journal scope expansion, and network restructuring. By the time a policy imperative generates sufficient research momentum to alter the dominance hierarchy measurably, the governance context it was designed to address may have already shifted [6,28,29]. The 30 × 30 target, adopted in 2022, will require scientific foundations in connectivity design, pollution management, equitable governance, and biodiversity monitoring that the current knowledge structure is not yet positioned to provide at the required scale or geographic breadth.

The geographic stratification documented in Section 3.5 introduces a further dimension of misalignment. The custodians of the world’s most ecologically irreplaceable marine systems in East Africa, the Pacific Islands, and Southeast Asia remain structurally marginal contributors to the knowledge base that governs those systems [26,27], generating what this study terms an epistemic governance gap: a mismatch between the geographic distribution of conservation value and the geographic distribution of conservation knowledge production. We label this an epistemic governance gap because its actionable component is epistemic: which knowledge is produced, and by whom. The geographic distribution of conservation value is treated here as an external, ecologically established input—determined by independent conservation-prioritization frameworks (e.g., IUCN Key Biodiversity Areas, CBD priority ecoregions) rather than by this study—against which the observed distribution of knowledge production is evaluated. The gap is thus epistemic in its remediable dimension, even though the benchmark against which it is measured is normative.

We do not claim that research conducted in the Global North is inherently incapable of engaging with Global South governance realities. Rather, the divergence documented in Section 3.5—where high-income countries’ output concentrates in resource-intensive methodological traditions (spatial planning, biotelemetry) while lower-income countries’ output concentrates in baseline biodiversity and pollution documentation—is more plausibly explained by differential access to long-term field presence, funding continuity, and institutional infrastructure than by any inherent property of where a researcher is based. It is this structural, resource-driven pattern—not an essentialist claim about capability—that we argue must be addressed for a ‘globally legitimate’ framework to be credible.

Several structural reorientations emerge as priorities for MPA science in the decade ahead. Deliberate investment in thematically peripheral frontiers, including pollution and stressor science, larval connectivity and network design, and socially grounded adaptive governance research, is necessary to reduce the dominance gaps that currently impede their integration into mainstream policy processes [25,30]. The epistemically distinct tradition of Chinese coastal governance science, currently self-referential and decoupled from the dominant Anglophone discourse, represents an underutilized reservoir of governance experience whose integration could meaningfully enrich the institutional frameworks available for evaluating the 30 × 30 commitment. The MPA knowledge system is not failing to produce science; it is producing science with extraordinary momentum and growing methodological sophistication. What it has not yet achieved is a systematic alignment between the structure of its knowledge production and the breadth of challenges that effective, equitable, and ecologically comprehensive marine protection demands.

3.7. Limitations and Future Research Directions

Several limitations must be acknowledged. The corpus is restricted to Web of Science and Scopus, which systematically underrepresent journals published outside high-income countries and research disseminated through regional channels; additionally, SciBERT was pre-trained predominantly on English-language literature, which may yield less precise semantic representations for non-English records, meaning that the geographic asymmetries identified in Section 3.5 may partially reflect structural inequalities in indexing coverage rather than purely differential research capacities [15]. The geographic attribution of publications to the corresponding author’s country does not capture the increasingly transnational character of contemporary MPA science, potentially overstating knowledge concentration in high-income nations. The BERTopic solution identified eight topics, a parsimonious architecture that may favor broad internal cohesion over finer-grained thematic discrimination; Topic 0 in particular may decompose into distinct sub-topics, such as adaptive co-management or marine spatial planning governance, whose separate dynamics would yield additional analytical insights. Finally, the RIF Index provides a cross-sectional measure of structural dominance; a fully longitudinal RIF analysis computing dominance structures separately for each policy milestone period would enable more precise quantification of how the knowledge hierarchy has shifted over time.

Future research should prioritize three directions: integration of grey literature from IUCN, CBD, and national MPA agencies, particularly in regions where peer-reviewed publication capacity is limited but practical conservation knowledge is rich [28]; construction of multilingual sub-corpora in Spanish, French, Portuguese, and Chinese to test whether North–South asymmetries are amplified or attenuated across linguistic research communities; and integration of citation network analysis with topic modeling to distinguish thematic prevalence from epistemic authority, a distinction the current analysis cannot fully resolve.

4. Conclusions

The empirical architecture assembled in this study converges on an interpretive claim that is simultaneously descriptive and normative: the global MPA knowledge system is not failing to produce science, but it is producing science whose internal organization remains structurally misaligned with the breadth and urgency of the conservation challenges it is called upon to address. The application of BERTopic to a corpus of 10,735 documents establishes a replicable framework for the scientometric analysis of complex, multidisciplinary conservation literatures, operating at the level of contextual semantic meaning rather than surface-level keyword frequency. The identification of eight coherent topics with a noise rate of 0.12% attests to both the internal cohesiveness of the MPA corpus and the discriminative capacity of the SciBERT-UMAP-HDBSCAN pipeline relative to traditional approaches.

The more substantively consequential contribution lies in the integration of BERTopic with the Relative Importance Factor Index, which for the first time transforms a qualitative typology of research themes into a statistically grounded, rank-based dominance topology. The RIF analysis reveals that structural peripherality accumulates gradually through compounding rank distances from the governance-ecology core, rather than through a single discontinuous break, implying that emerging research agendas face not a single institutional barrier but a sustained gradient of epistemic inertia. The convergent trajectories of rapidly growing yet persistently peripheral topics, most notably marine pollution science and Chinese coastal governance research, constitute the phenomenon designated here as accelerated marginality: both exhibit among the strongest temporal growth signals in the corpus, yet their absolute structural position in the dominance hierarchy has not been meaningfully altered over the study period. This decoupling illustrates a broader property of power-law scientometric systems: peripheral topics cannot close the gap with the dominant core simply by growing faster than it, because the core’s concentration of epistemic authority is self-reinforcing. Overcoming this asymmetry therefore requires deliberate structural intervention—such as prioritized funding, interdisciplinary training, and broadened journal scope—rather than organic growth alone.

This matters not because of who conducts the research per se, but because our own thematic profiling (Section 3.5, Figure 9) shows that what is studied differs systematically with where it is produced: high-income countries disproportionately produce spatial-management and biotelemetry-based research, while lower-income countries disproportionately produce baseline biodiversity and pollution research. Research agendas are therefore not geographically interchangeable—a pattern consistent with the broader literature on ‘parachute’ or ‘helicopter’ science in ecology, where externally led research tends to prioritize questions tractable from outside rather than the governance priorities of the region studied [31,32].

The geographic dimension deepens the normative stakes. The epistemic governance gap identified here, a mismatch between the geographic distribution of marine conservation value and the geographic distribution of conservation knowledge production, has direct implications for the credibility of globally ambitious targets such as 30 × 30. The custodians of the world’s most ecologically irreplaceable marine environments remain structurally marginal contributors to the knowledge base that governs those environments, a condition that the Kunming–Montreal commitment to equitable and inclusive marine governance formally rejects but that empirical evidence confirms remains unaddressed. The epistemically distinct tradition of Chinese coastal governance science, self-referential and largely decoupled from Anglophone discourse, represents an underutilized intellectual resource whose integration could substantially diversify the governance frameworks available for implementing the 30 × 30 agenda.

Taken together, these findings indicate that the effective implementation of ambitious marine conservation commitments will require not only expanded MPA coverage but a deliberate transformation of the epistemic architecture through which marine conservation knowledge is produced, validated, and translated into governance action. Beyond the specific case of MPAs, the combination of neural topic modeling and rank-based dominance analysis provides a general framework for investigating knowledge hierarchies in large scientific corpora, offering a transferable approach for examining how scientific attention is distributed, concentrated, and transformed across complex research domains.