The research, conducted by Yanjing Gao, Li Shen, Qiuhua Chen, and Hang Li, assembled a corpus of 45 core policy documents issued between 2013 and 2026, spanning national-level directives and policies for China’s first five officially established national parks: Sanjiangyuan, Giant Panda, Northeast China Tiger and Leopard, Hainan Tropical Rainforest, and Wuyishan. From this corpus, the team selected 13 representative policies across three tiers—central top-level design documents, laws and regulations, and master plans—and subjected them to a two-stage analysis that pairs machine learning with structured policy evaluation.

The first stage relied on BERTopic, a neural topic-modeling technique that uses transformer-based language models to detect latent themes in large text collections. The researchers generated sentence embeddings with a multilingual MiniLM model, then tuned clustering parameters through repeated UMAP-HDBSCAN runs, selecting the configuration that balanced topic coherence, diversity, outlier rates, and consistency across runs. The algorithm initially produced 16 topics, which the team manually consolidated into eight core themes after checking that the grouping was statistically defensible. A Monte Carlo permutation test based on 100,000 random reallocations confirmed that the eight-theme structure was significantly more semantically cohesive than chance would predict, with a p-value of 0.002.

Those eight themes map the intellectual terrain of China’s national park project. The most prominent, accounting for roughly 29 percent of the corpus, concerns governance, institutions, and coordinated spatial management—the machinery of who runs the parks and how. Flagship species conservation, featuring giant pandas, tigers, and leopards, came a close second at about 28 percent. Natural resource conservation and the regulation of development activities, ecotourism and community livelihoods, ecological restoration, fiscal support and ecological compensation, forest disaster early warning, and the regulation of illegal activities rounded out the picture. The trajectory is clear: policy has evolved from emergency rescue of individual endangered species toward comprehensive systems encompassing habitats, corridors, and whole ecosystems.

The second stage of the analysis is where the study makes its methodological mark. The Policy Modeling Consistency Index, or PMC-Index, is an established tool for scoring policy documents, but conventional applications use binary coding—a provision either exists or it does not. The researchers argue this flattens crucial distinctions between vague principles and operational rules. Their solution is a modular framework with a common core module applied to all documents and type-specific modules tailored to strategic documents, laws, and master plans, combined with a three-level ordinal coding scheme of 0, 0.5, and 1. A score of 0.5 signals partial institutionalization: responsible actors and measures are named, but key elements such as procedures, standards, resources, or feedback mechanisms are missing. A full score requires a complete chain linking actors, targets, measures, procedures, and monitoring.

The results are striking. The 13 documents achieved a mean standardized PMC score of 0.729, which the study’s rating scale classifies as good—evidence that China’s national park policy framework is now relatively well-developed. Scores were highest for legal authority and consistency, ecological conservation, and resource management, reflecting the system’s strengths in centralized coordination, spatial zoning, and species protection. The Master Plan for Wuyishan National Park earned the highest individual score at 0.847, the only document rated excellent, while the foundational 2017 Overall Plan for Establishing the National Park System scored lowest at 0.653.

But the deeper story lies in what scored poorly. Alignment with public values, stakeholder integration, cross-jurisdictional coordination, community empowerment, and monitoring-based adjustment all lagged behind. Notably, nearly half of all indicator observations—49.7 percent—received the intermediate 0.5 score, slightly more than received a full 1. This means the central challenge facing China’s national park policy has shifted: the question is no longer whether an institutional element exists on paper, but whether it is supported by quantitative standards, operational procedures, resource allocation, and feedback loops that make it real.

The study’s most consequential finding emerges when the two analytical stages are combined. By multiplying each theme’s prominence across the corpus by the institutional design strength of the documents addressing it, the researchers sorted all eight themes into four quadrants. Flagship species conservation and ecotourism-community livelihoods landed in the high-prominence, strong-design quadrant—the system’s showpieces. Ecological restoration and disaster management sat in the low-prominence, strong-design quadrant, reflecting detailed provisions concentrated in specialized documents rather than absence of design. But governance coordination and resource regulation, despite dominating policy attention, fell into the high-prominence, weak-design quadrant, with cross-jurisdictional procedures, central-local responsibility divisions, and dispute-resolution mechanisms remaining thin. Financing and legal accountability landed in the weakest quadrant, though the authors caution this partly reflects the dispersion of fiscal and compensation rules across specialized instruments outside the core sample.

This structural mismatch matters because national parks are increasingly asked to do something conventional protected-area governance was never designed for: adapt. As climate change shifts species ranges, vegetation zones, and disturbance regimes, fixed park boundaries may no longer contain the habitats that matter most. Four of China’s first five parks depend heavily on forest ecosystems, and several span provincial borders, meaning wildlife migration, wildfire, and pest outbreaks ignore the administrative lines on which governance is built. The study argues that effective conservation requires permanent cross-jurisdictional coordination mechanisms, harmonized monitoring standards, interoperable ecological information systems, joint law enforcement, and ecological compensation arrangements across provinces—precisely the areas where institutional design is weakest.

The community dimension is equally revealing. Ecotourism and livelihood themes score well, but the researchers note this strength flows mainly from compensation mechanisms, conservation employment, and project-based involvement rather than from formal community rights to participate in decision-making. Comparisons with Australia’s co-managed parks, such as Kakadu and Uluru-Kata Tjuta, suggest that even well-established co-governance arrangements do not guarantee communities an equal say. The study recommends that future reforms give communities genuine roles in resource monitoring, rule-making, impact assessment, and standard-setting, with clear rules for representative selection, terms of office, and how park authorities must respond to community input.

The authors’ prescriptions are refreshingly concrete. Rather than calling for new policy elements, they argue the priority is vertical translation: converting strategic objectives into measurable targets, formal inter-departmental collaboration procedures, stable benefit-sharing mechanisms, and adaptive management systems that link monitoring and evaluation to policy revisions and accountability. Each park also needs tailored institutional tools—threshold-based monitoring for Sanjiangyuan’s permafrost and grasslands, corridor cost-sharing among Sichuan, Shaanxi, and Gansu for giant pandas, cross-border coordination for Amur tigers, harmonized standards for Wuyishan’s tea-growing landscapes, and canopy restoration for the Hainan gibbon’s critically small population.

The study has limits the authors acknowledge. The PMC-Index measures the structural integrity of policy texts, not the ecological and social outcomes that follow implementation, and the sample excludes grassroots implementation documents and technical standards. Future work should pair policy scores with empirical data on species populations, land cover, visitor pressure, and enforcement performance. Still, the message is clear and globally resonant: in conservation, as in so much else, the gap between what governments say and what they operationalize is where the fate of ecosystems is decided. China’s national park system has built the skeleton; the next decade will determine whether it grows the muscle.

Subject of Research: Machine-learning and index-based evaluation of China's national park policy framework from 2013 to 2026

Article Title: Policy evaluation in China’s national park system: integrating BERTopic modeling with a modular PMC framework

Article References: Gao, Y., Shen, L., Chen, Q., & Li, H. (2026). Policy evaluation in China’s national park system: integrating BERTopic modeling with a modular PMC framework. Frontiers in Forests and Global Change, 9, Article 1950431. https://doi.org/10.3389/ffgc.2026.1950431

Image Credits: AI Generated

DOI: 10.3389/ffgc.2026.1950431

Keywords: national parks, China, BERTopic, PMC-Index, policy evaluation, biodiversity conservation, adaptive management, community participation, cross-jurisdictional coordination, climate change adaptation, ecological compensation, flagship species