Semantic Collaborative Environment for Extended Digital Natural Heritage: Integrating Data, Metadata, and Paradata

Abstract

Natural heritage digitization has evolved beyond simple 3D representation. Contemporary approaches require transparent documentation integrating biological, heritage, and digitization standards, yet existing frameworks operate in isolated domains without semantic interoperability. Current digitization frameworks fail to integrate biological standards (Darwin Core, ABCD), heritage standards (CIDOC-CRM), and digitization standards (CRMdig, PROV-O) into a unified semantic architecture, limiting transparent documentation of natural heritage data across its entire lifecycle—from physical observation through digital reconstruction to knowledge reasoning. This study proposes an integrated semantic framework comprising three components: (1) the E-DNH ontology, which adopts a triple-layer architecture (data–metadata–paradata) and a triple-module structure (nature–heritage–digital), bridging Darwin Core, CIDOC-CRM, CRMdig, and PROV-O; (2) the HR3D workflow, which establishes a standardized high-precision 3D data acquisition protocol that systematically documents paradata; and (3) the C-EDNH platform, which implements a Neo4j-based knowledge graph with semantic search capabilities, AI-driven quality assessment, and persistent identifiers (NSId/DOI). The framework was validated through digitization of 197 natural heritage specimens (68.5% avian, 24.9% insects, 5.1% mammals, 1.5% reptiles), demonstrating high geometric accuracy (RMS 0.18 ± 0.09 mm), visual fidelity (SSIM 0.92 ± 0.03), and color accuracy (ΔE₀₀ 2.1 ± 0.7). The resulting knowledge graph comprises 15,000+ nodes and 45,000+ semantic relationships, enabling cross-domain federated queries and reasoning. Unlike conventional approaches that treat digitization as mere data preservation, this framework positions digitization as an interpretive reconstruction process. By systematically documenting paradata, it establishes a foundation for knowledge discovery, reproducibility, and critical reassessment of digital natural heritage.

1. Introduction

The 1972 UNESCO Convention on the Protection of World Cultural and Natural Heritage defined both cultural and natural heritage as assets of universal value to be collectively preserved by humanity [1]. As of 2025, there are 1248 properties on the World Heritage list, of which 235 are natural heritage sites serving as environmental archives that document geological and biological processes [2]. Recently, the digital transformation of natural heritage has become a key infrastructure for conservation science, biodiversity research, and ecological education [3]. However, most institutions remain at a production-centric stage [4]. Although high-precision technologies such as photogrammetry, structured-light scanning, and CT are widely used [5], the resulting 3D data exist in heterogeneous formats (e.g., OBJ, PLY, glTF) [6] and are constructed according to isolated institutional standards. This fragmentation impedes long-term preservation and reuse, as isolated data structures hinder cross-institutional integration [7], and geometric data are frequently managed separately from metadata, making it difficult to capture temporal changes [8].

In this context, biodiversity informatics introduced the concept of the Digital Extended Specimen (DES), which manages physical specimens as extended digital objects linked with genomic, geographic and ecological data [9,10], realizing the principles of FAIR [11]. However, this concept remains mainly focused on biological taxonomy, often failing to reflect the cultural heritage contexts emphasized by UNESCO [12]. For example, the SYNTHESYS3 project integrated specimen data from 21 European institutions but lacked the systematic management of high-fidelity 3D models, failing to realize the multidimensional connectivity required by DES [13]. To bridge this gap, this study draws on cultural heritage precedents in which frameworks integrate metadata and paradata to ensure data reliability and transparently communicate the technical decisions and interpretive contexts underlying digital reconstruction [14,15]. This demonstrates how subjective interpretations, such as collective memory and cultural identity, can be systematically encoded through structured metadata. The concept “Memory Twin” exemplifies this approach, presenting a participatory ecosystem aligned with the principles of FAIR and CARE [16].

Building on these precedents, this study addresses three research questions: RQ1: How can biological standards (Darwin Core, ABCD), heritage standards (CIDOC-CRM [17]) and digitization standards (CRMdig [18], PROV-O [19]) be integrated into a unified semantic framework for transparent documentation throughout the lifecycle of natural heritage? RQ2: How can the paradata from high-precision 3D digitization be systematically structured and integrated with the data and metadata to enable reproducibility? RQ3: How can a knowledge graph-based platform support semantic reasoning and federated queries across biological, heritage, and digitization domains? To address these questions, we implement DES as a practical system, proposing a digital ecosystem that integrates paradata from a nature-specific high-fidelity 3D digitization workflow with data and metadata into a graph-based knowledge database, moving beyond text-centric systems to encompass biological attributes, heritage values, and technical contexts.

2. The State of the Art

2.1. Evolution of Natural Heritage Data Management Environments

Natural heritage data standardization, evolving from Linnaean taxonomy and 19th-century nomenclature codes, faces structural limitations with 3D digitization proliferation. Although GBIF and iDigBio secured formal interoperability through Darwin Core-based networks (180 countries, 118 million records), their text-based occurrence structures inadequately integrate high-dimensional data such as 3D models and paradata [20]. Arctos provides community governance and flexible attribute management, but lacks ontological structures to semantically link 3D technical attributes. All three platforms employ researcher-centric biodiversity interfaces, failing to establish circular governance that includes data producers, managers, and users. As

Table 1. Comparative analysis of current natural heritage data platforms [22,23,24,25].

Platform	Core Strategy	Limitation
Arctos	• Community-driven governance (400+ institutions) • Entity-based relational model with flexible many-to-many linking • Darwin Core integration with PID-based interoperability	• Lacks ontology-based semantic structure for 3D object attributes (mesh resolution, texture versioning) • Limited paradata integration for digitization workflows
GBIF	• Global federation model (180+ countries) • Occurrence-centric data model with Darwin Core standardization • Distributed publishing via IPT	• Text-based occurrence records; limited support for 3D, paradata, and non-taxonomic contexts • Expert-centric interface limits broader user engagement
iDigBio	• Cloud-based centralized hub with federated storage • Thematic Collections Networks for taxonomic collaboration • RESTful API and RDF-based media linking	• Darwin Core dependency; lacks structured paradata for digitization provenance • Metadata-centric approach insufficient for 3D data lifecycle management

indicates, current platforms emphasize formal interoperability while neglecting semantic integration to interpret the microscopic, social, and heritage contexts of samples, with limited structural scalability for the representation of high-dimensional data. Expert-centric participation structures prevent fully operational circular governance [21], necessitating a transition toward intelligent models that integrate complicated cultural contexts with diverse user dynamics.

2.2. Evolving Standardization by Integrating Metadata, Paradata, and 3D Data Management

Natural heritage data standardization, evolving from Linnaean taxonomy and 19th-century nomenclature codes, faces structural limitations with 3D digitization proliferation. Darwin Core (DwC) and ABCD, core biodiversity exchange standards [26], employ text-based two-dimensional structures inadequate for describing geometric properties, resolution, and lineage of three-dimensional models. Audubon Core, designed for 2D media, lacks LOD management and provenance tracking for such models. While FAIR principles emphasize accessibility [11], they omit quality metrics like mesh integrity and texture resolution; CARE principles similarly lack paradata frameworks documenting equipment specifications, resolution settings, and post-processing algorithms—critical for assessing scientific suitability when identical specimens undergo different digitization methods. In cultural heritage, RMdig ontologically models digitization [18] but lacks integration with natural science attributes and Darwin Core mechanisms.

Addressing these gaps requires three-layered infrastructures managing spatial properties, workflows, and quality contextually. The Tendaguru Dinosaur Expedition validated this approach (data–metadata–paradata) for documenting excavation contexts, taxonomic revisions, and conservation histories, with paradata comprising 80–90% of datasets and ensuring reproducibility [27], though universal applicability remains limited. The EUreka3D project advanced this through composite entities (point clouds, meshes, textures) and standardized management via integrated repositories, aggregators, viewers, and virtual environments [28], yet cannot adequately represent specimen–collection–institution hierarchies and dynamic biological taxonomy. Consequently, bridging biological standards (DwC, ABCD) with digitization standards (CRMdig) through validated infrastructure is essential for inter-institutional interoperability, scientific reliability, and collaborative reuse of three-dimensional natural heritage data.

3. Methodology

Although natural heritage digitization has advanced through OCR-based transcription and quality management, existing standards inadequately capture morphological precision, heritage values, and contextual paradata, necessitating systems that transcend “accurate information” to become interpretive knowledge resources integrating meaning, provenance, and context.

This study proposes an integrated methodology based on the concept of the digital extended specimen (DES) [9], a multi-layered structure that extends from physical specimens through CMS records, annotations, and automated transactions. The model systematizes data lifecycles via a three-tier architecture (data, metadata, paradata) through four stages: (1) data acquisition and normalization consolidates multi-institutional specimens with standardized terminology; (2) 3D digitization establishes workflows combining structured-light scanning and photogrammetry with controlled parameters ensuring reproducibility; (3) knowledge graph database interlinks biological data, heritage values, and digitization processes through E-DNH ontology; (4) collaborative management and utilization integrates semantic search, AI-driven quality assessment, real-time annotation, and persistent identifiers for global interoperability. As illustrated in Figure 1, this framework operates in organic integration with the DES three-tier model, establishing a collaborative circular ecosystem where biodiversity researchers, digitization technicians, curators, and educators participate in the production, management, and reuse of natural heritage data, ensuring reliability and interoperability through a shared semantic layer.

Data Resource Layer

The initial stage of this research focuses on the establishment of the Data Resource Layer, which serves as the foundational infrastructure for the E-DNH ontology. This layer provides a systematic basis for subsequent 3D digitization workflows and ontology design processes through the physical acquisition of natural heritage specimens and the normalization of descriptive data. A total of 197 specimens of natural heritage were collected through institutional collaboration with the Hannam University Museum of Natural History, the National Heritage Administration, and Seoul Grand Park, comprising avian species (68.5%), insects (24.9%), mammals (5.1%), and reptiles (1.5%), as detailed in

Table 2. Dataset composition and institutional distribution of 197 natural heritage specimens.

Taxonomic Group	Specimens (n)	Percentage	Institutions	Major Taxa
Aves (Birds)	135	68.5%	NMCNMK (95), SGP (40)	Cranes, eagles, owls, herons, etc.
Insecta (Insects)	49	24.9%	HNU (49)	Beetles, mantises, grasshoppers, etc.
Mammalia (Mammals)	10	5.1%	NMCNMK (7), SGP (3)	Tiger, otter, bear, goral, etc.
Reptilia (Reptiles)	3	1.5%	SGP (2), NMCNMK (1)	Tortoises, turtles, etc.
Total	197	100%	3 institutions	ca. 140 species

NMCNMK = National Monument Center of Ministry of Natural Knowledge; SGP = Seoul Grand Park; HNU = Hannam University Natural History Museum.

.

Specimen selection was conducted based on institutional preservation environments, data management systems, and scanning suitability criteria, with systematic standardization of primary metadata and digital quality control variables. Through advisory workshops involving domain experts from the National Heritage Administration and the Korea National Park Service, data normalization protocols were established to resolve inter-institutional inconsistencies in taxonomic identification, storage conditions, and metadata documentation standards. The dataset constructed during this stage is organized into three categorical domains (

Table 3. Data types and characteristics in the natural heritage and specimen.

Data Type	Characteristics	Application
Specimen Attributes	• Taxonomic & biological data • Collection metadata • Preservation state • Persistent identifiers (PIDs)	• Taxonomic classification • Specimen instances • Linking physical specimens to digital surrogates
Project Information	• Digitization activity records • Workflows & equipment parameters • Institutional collaboration metadata • Conservation & restoration records	• Documentation of digitization events • Paradata modeling via CRMdig • Activity attribution with PROV-O
Heritage Records	• Unstructured text formats • Historical documents • Public reports like evaluation • Cultural and ecological narratives	• Integration of cultural significance • Heritage enrichment • Semantic annotation of narratives

): (1) specimen attributes, (2) project-based data, and (3) heritage records. These domains are interconnected through an integrated schema architecture, ensuring cross-institutional interoperability and providing a scalable infrastructure for subsequent 3D digitization processes and ontology-based semantic integration.

4. Higher-Reality 3D Digitization Workflow for Natural Heritage

Building upon the Data Resource Layer, this section investigates the 3D digitization workflow during physical-to-digital transition, focusing on paradata specificities unique to natural heritage. Natural heritage 3D digitization presents greater morphological authenticity challenges than cultural heritage due to biologically variable elements such as feathers, epidermis, and body hair [29], requiring distinct protocols that accommodate irreversible morphological deformation and surface variations. This study establishes a workflow tailored to biological specimens, operationalizing extended digital specimens by systematically structuring all technical and interpretive decisions as paradata. As illustrated in Figure 2, the workflow comprises six stages—pre-processing, acquisition, QA/QC, post-processing, packaging, and meta-layer integration—designed as an integrated system simultaneously generating data, metadata, and paradata.

4.1. Digitization Workflow and Quality Management

During pre-processing, specimen characteristics and environmental variables were quantitatively defined, with equipment settings standardized for reproducibility. Scanning was performed under controlled conditions: temperature 20 ± 2 °C, humidity 45 ± 5%, diffused LED illumination (1000–1500 lux, 5500K, CRI > 95), and cross-polarization filters to minimize specular reflection. A structured-light scanner (Artec Space Spider, Artec 3D, Luxembourg City, Luxembourg; 0.05 mm accuracy) captured fine morphological details via automated multi-angle scanning (360° rotation, 5° intervals, 48 positions). The texture acquisition used a a Sony $\alpha$7R V camera (Sony Corporation, Tokyo, Japan; 61 MP), generating 12,000–25,000 frames per session. All parameters were documented as PROV-O-based datasets (

Table 4. Major equipment and specifications used for 3D digitization.

Category	Equipment	Key Specifications	Application Stage
Structured-light Scanner	Artec Space Spider (Artec 3D, Luxembourg City, Luxembourg)	0.05 mm point accuracy, 16 fps	Close-range geometry capture
DSLR Camera	Sony $\alpha$7R V (Sony Corporation, Tokyo, Japan)	61 MP, 35 mm f/1.4, ISO 100	Texture acquisition
Turntable System	RB10-1300 (Rotary Systems, Seoul, South Korea)	360° rotation, 5° interval, 48 positions	Automated rotational scanning
Lighting	Diffuse LED Panel (Manufacturer name, City, Country)	1000–1500 lux, 5500 K, CRI > 95	Uniform illumination environment

).

The acquisition process adopted a hybrid approach that combined structured-light scanning and photogrammetry. Small-to-medium specimens were scanned at 25–30 cm with 70% overlap, with occluded regions rescanned in real-time. The large specimens were captured using 192 RAW images with 80% overlap and 30–60° crossing angles, processed using the SfM and MVS algorithms in Agisoft Metashape 2.1.2. Photogrammetric alignment and point cloud reconstruction were performed in ContextCapture (Figure 3).

Quality assurance and control were conducted on all 197 specimens, with 20 (10%) randomly sampled for validation. Geometric accuracy was evaluated using RMS, HD95, and Chamfer Distance, visual fidelity through SSIM, $\Delta E_{00}$ (CIEDE2000), and FID. The results yielded RMS 0.18 ± 0.09 mm, $\Delta E_{00}$ 2.1 ± 0.7, and SSIM 0.92 ± 0.03, meeting the thresholds (RMS ≤ 0.3 mm, SSIM ≥ 0.95, $\Delta E_{00}$≤ 3.0) for academic and exhibition applications. Quality data were stored as :QualityCheck nodes in Neo4j, with subthreshold datasets automatically flagged for reacquisition. Post-processing employed ICP alignment, TSDF fusion, noise removal, remeshing, UV unwrapping, 8K texture baking, and color correction through Blender 4.0.2 and Artec Studio 18, maintaining $\Delta E_{00}$≤ 2.1 through ColorChecker calibration. The photogrammetry and structured-light datasets were merged, achieving 100% watertight meshes that combine high-resolution textures with precise geometry. The final outputs comprised a three-tier structure: archival OBJ (0.5–2 GB), analytical PLY (2–5 GB), and web-deployable GLB (50–200 MB), each assigned Darwin Core identifier (dwc:occurrenceID) and cryptographic checksums before transfer to an OAIS ISO 14721:2012 [30] -compliant preservation system.

4.2. Paradata Documentation and Integration

This stage identifies and structures contextual (paradata) and evidential elements throughout digitization, demonstrating how technical manipulations, instrumentation, and human judgment influence interpretive and representational fidelity. This systematic documentation serves two critical methodological functions: enabling reproducibility by establishing transparent records of technical decisions, and it supports critical reassessment by exposing the interpretive choices embedded within digital outputs. Unlike cultural heritage, natural heritage must accommodate biological variability and irreversible temporal changes; therefore, each scanning event functions as an evidential occurrence reflecting environmental, instrumental, and procedural contexts that impact reproducibility and scientific credibility. To establish empirical foundations for the categorization of paradata, the technical and environmental variables—illumination, humidity, scanning angles—were quantitatively analyzed for correlations with geometric accuracy, texture fidelity, and reconstruction quality. Paradata elements were categorized into three domains (

Table 5. Analysis of paradata scopes in the digitization process of natural heritage.

Category	Key Elements	Analytical Purpose
Physical Context	Scanning environment (illumination, humidity), specimen mounting, device resolution, scanning angles	Reproducibility and scientific credibility through controlled acquisition conditions
Computational Context	Alignment algorithms, texture mapping parameters, reconstruction workflows	Transparency and traceability of computational transformations
Interpretive Context	AI model selection, manual editing, quality thresholds, validation criteria	Accountability and interpretive transparency in human judgment

): (1) physical context (environment, specimen mounting, device resolution); (2) computational context (alignment algorithms, texture mapping); and (3) interpretive context (AI model selection, manual editing, quality thresholds). These categories structure digitization as an evidence–process–outcome chain, with paradata records corresponding to ontological classes (AcquisitionEvent, DataProcessing, QualityAssessment). By providing the semantic basis for integrating digitization standards (CRMdig, PROV-O) with biological and heritage metadata, this structured documentation enables precise tracking of how technical procedures and interpretation constitute digital specimens as evidence-based composite objects integrating scientific observation, technical reconstruction, and interpretive transparency, founding the E-DNH ontology.

5. Extended Digital Natural Heritage Ontology

5.1. Purpose of the Ontology Structure

Natural heritage constitutes both biological entities and composite artifacts accumulating human acts of perception, documentation, and preservation; however, existing digital systems remain confined to specimen-centric metadata management, inadequately reflecting this multi-layered character. To address the challenge of integrating heterogeneous standards across biological, heritage, and digitization domains, this study proposes the Extended Digital Natural Heritage (E-DNH) ontology, conceptualizing natural heritage as knowledge ecosystems that systematically describe interacting structures of observed nature, heritage values, and digital processes. The ontology resolves three methodological requirements. Firstly, it provides a shared semantic model, semantically connecting terminologies and data structures across Darwin Core, CIDOC-CRM, and CRMdig/PROV-O, bridging isolated domain standards into a coherent framework. Secondly, through the explicit formalization of tacit knowledge, it systematizes scientific, institutional, and technical knowledge into explicit relationships and logical constraints, enabling transparent documentation of paradata to support reproducibility and critical reassessment. Third, by enabling machine reasoning beyond data storage, it supports the inference-based generation of novel insights (e.g., conservation prioritization, climate risk assessment), with a modular standards-based design, ensuring scalability and connectivity to global natural heritage knowledge ecosystems, operationalizing the DES concept through semantic infrastructure for federated queries and cross-domain knowledge discovery.

5.2. Design Principles

The E-DNH ontology was designed based on three foundational principles to structurally represent complex interrelationships among physical reality, cultural contexts, and digital activities of natural heritage.

Triple Module with Data, Metadata, and Paradata. The E-DNH ontology distinguishes information into three conceptual tiers: the data layer represents physical specimens or observational records; the metadata layer describes cultural values, legal status, and curatorial contexts; and the paradata layer encompasses technical and interpretive processes of digital production. This design functions as both a data structure and a principle representing information lifecycles and epistemological evolution of knowledge. Conventional systems conflate facts, interpretations, and contexts in a single plane, impeding credibility assessment. By stratifying knowledge, this architecture enables retroactive tracking from digital outputs to raw data through the PROV-O provenance model, ensuring reproducibility.

Table 6. Mapping between 3D digitization workflow and paradata elements in natural heritage.

3D Digitization Task	Type of Paradata	Ontology Class
Input condition setting	Equipment selection, specimen complexity analysis	InputProvenance
Acquisition control	Environmental conditions, image overlap ratio	AcquisitionEvent
Algorithm parameters	Registration threshold, feature extraction method	DataProcessing
Interpretive decisions	AI model selection, reconstruction scope	InterpretiveDecision
Reliability expression	Confidence scores, extent of data loss	UncertaintyAnnotation
Quality assessment	Geometric accuracy, color fidelity	QualityAssessment
Optimization strategy	Polygon reduction, texture compression	OptimizationProcedure

demonstrates how 3D digitization attributes operate as paradata, illustrating uncertainty propagation across equipment calibration, acquisition conditions, and processing chains, integrating the quality assessment framework [31] with systematic uncertainty documentation [32] through PROV-O and CRMdig.

Semantic Reasoning and Knowledge Expandability. The E-DNH ontology incorporates reasoning mechanisms deriving implicit knowledge from explicitly modeled information, functioning as an active knowledge system supporting knowledge discovery and hypothesis generation. Taxonomic reasoning connects Taxon Identification and Location data to identify biodiversity patterns; heritage value reasoning determines conservation priorities through relational analysis between Heritage Aspect and Outstanding Universal Value; and technical reasoning enhances traceability by linking Quality Assessment with Provenance Trace. These capabilities are enabled through inter-module relationships and semantic definitions, allowing the knowledge graph to expand dynamically and generate novel insights through evolving relationships.

5.3. Structural Principles of Modularization

The E-DNH ontology adopts a modular architecture inspired by the NeOn Methodology [33], decomposing complex natural heritage into semantically connected components. This approach enables domain-specific module refinement (Nature Module for biological dimensions, Heritage Module for cultural contexts, Digital Module for digitization processes), ensures adaptability across diverse projects, and minimizes dependencies while maintaining inter-module interoperability. Core entities (DigitalSpecimen, HeritageAspect, DigitalActivity) function as semantic bridges enabling cross-module reasoning and federated querying.

As visualized in Figure 4, the three modules correspond directly to the data–metadata–paradata structure defined in the Triple Module with Data, Metadata, and Paradata principle. The Nature Module models empirical data (e.g., Event, Location), the Heritage Module encodes interpretive and curatorial metadata (e.g., HeritageProject, LegalInstrument), and the Digital Module captures process-level paradata (e.g., AcquisitionEvent, DataProcessing). Together, they represent the complete intellectual lifecycle of natural heritage—from observation through value attribution to digital reconstruction and validation—positioning E-DNH as a semantic knowledge infrastructure supporting integrative reasoning, transparency, and long-term sustainability.

5.4. Nature Module (Nature and Digital Specimen)

The Nature Module integrates biodiversity informatics standards to enable semantic interoperability across natural history institutions. It adopts Darwin Core [26] as its core vocabulary—a standard developed through over 30 years of collaborative effort—while integrating ABCD 3.0 [34] for anatomical specimen descriptions, Audubon Core [35] for multimedia metadata, and the MorphoSource model for 3D morphological data. This standards-based integration ensures both scientific completeness and cross-institutional data exchange among museums, research institutions, and biodiversity aggregators.

5.4.1. Definitions of Core Classes in Nature Module

Table 7. Core classes of the Nature Module.

Class	Definition and Functional Role
DigitalSpecimen	Represents the digital surrogate of a physical natural specimen, characterized by persistent identifiers, specimen type, and holding institution. Extends the Darwin Core Occurrence class to establish a curated digital representation linked to the physical specimen within museum collections, supporting the Digital Extended Specimen (DES) paradigm.
Event	Records the spatiotemporal context of specimen collection, observation, or documentation activities. Captures the precise moment when the specimen was documented within a scientific framework, serving as essential provenance information for assessing specimen origin, collection methodology, and data reliability.
Location	Defines the geographic context in which the specimen was discovered or observed. Incorporates hierarchical spatial data including coordinates, administrative divisions, place names, elevation, and depth, supporting ecological distribution analysis, biogeographic research, and heritage site boundary delineation.
Identification	Documents the taxonomic determination process and outcomes for a specimen. Explicitly records the taxonomist, determination date, identification method, and confidence level, enabling historical traceability and re-evaluation of taxonomic decisions as systematic knowledge evolves.
DigitalMedia	Describes multimedia representations of specimens—including photographs, videos, 3D models, and audio recordings—structured according to Audubon Core metadata standards. Includes technical attributes such as file format, resolution, color space, authorship, and capture conditions, ensuring standardized management and interoperability of digital assets.
Agent	Represents individuals or organizations participating in specimen-related activities, including collectors, taxonomists, curators, digitization technicians, and researchers. Documents roles, contributions, and institutional affiliations to support scientific accountability, provenance transparency, and scholarly citation throughout the specimen lifecycle.

defines six core classes establishing the ontological structure for specimen representation. These classes extend Darwin Core concepts to support structured documentation of specimen attributes, spatiotemporal provenance, taxonomic determination, and multimedia associations.

5.4.2. Object Properties in Nature Module

Table 8. Core object properties of Nature Module.

Property Name	Description	Example	Linked Standard
recordedIn	Links specimen to collection event	Beetle → Field collection May 2023	Darwin Core Event
hasLocation	Connects event to location	Event → Jeju Province (33.4569° N)	Darwin Core Location
identifiedBy	Links specimen to identification	Butterfly → Identified by Dr. Kim 2024	Darwin Core Identification
hasDigitalMedia	Links specimen to media	Specimen → lateral.jpg (6000 × 4000 px)	Audubon Core
performedBy	Links activity to agent	Collection → NIBR Survey Team	Darwin Core Agent
curatedBy	Links specimen to institution	Specimen → National Science Museum	Darwin Core Agent

summarizes the principal object properties defined for the Nature Module.

5.5. Heritage Module

The Heritage Module semantically models the processes through which natural heritage is recognized, protected, and managed as a shared cultural asset. Building upon CIDOC-CRM’s event-centered philosophy—conceptualizing heritage as a socially constructed phenomenon evolving over time—the module integrates the LIDO [36] metadata schema to capture heritage valuation, inscription, conservation, and utilization processes. This standards-based integration bridges biological metadata with heritage metadata, enabling unified reasoning across natural and cultural dimensions. Applied to 197 specimens from UNESCO-designated sites (e.g., Jeju Volcanic Island and Lava Tubes) and national natural monuments (e.g., Natural Monument No. 182), the module demonstrates how biological specimens function simultaneously as scientific evidence and culturally constituted assets.

5.5.1. Definitions of Core Classes in Heritage Module

Table 9. Core classes of the Heritage Module.

Class	Definition and Functional Role
OutstandingUniversalValue	Represents UNESCO World Heritage inscription criteria (vii–x), structuring attributes for geological processes, ecosystems, biodiversity, and esthetic value. Enables systematic assessment of heritage significance through quantifiable criteria (e.g., Jeju Volcanic Island’s geological diversity, DMZ biodiversity corridor), supporting conservation prioritization reasoning.
HeritageInscription	Records administrative and legal designation procedures across multi-layered protection systems (World Heritage, national monuments, provincial heritage). Tracks chronological stages (e.g., Jeju designation 2007), enabling historical traceability and comparative analysis of protection frameworks.
HeritageAspect	Decomposes multidimensional heritage values into separate evaluable attributes (biological diversity, geological importance, ecological processes, cultural landscape, scientific research value). Supports aspect-specific conservation strategies and enables reasoning across heritage value dimensions.
HeritageProject	Represents organized conservation, restoration, or sustainable use activities (e.g., wetland rehabilitation, invasive species removal, environmental monitoring). Links to Agent and Event classes to document stakeholders, timelines, and outcomes, enabling systematic evaluation of management effectiveness.
LegalInstrument	Describes legal, regulatory, and policy frameworks governing heritage protection (international conventions, national laws, ordinances, management guidelines). Specifies scope, enforceability, and jurisdictional hierarchy, supporting compliance analysis and conflict resolution in multi-jurisdictional sites.
SocialEngagement	Models participation of local communities, indigenous groups, and stakeholders, operationalizing CARE principles. Documents collaborative governance, traditional ecological knowledge integration, and consultation processes, enabling analysis of stakeholder participation patterns and diverse knowledge systems.

define six core classes establishing the ontological structure for heritage representation. These classes extend CIDOC-CRM concepts to support structured documentation of heritage values, legal frameworks, and social dimensions, addressing existing DES implementations’ failure to reflect UNESCO’s cultural heritage contexts. The integration enables three forms of cross-domain reasoning demonstrated in Section 5.7.: (1) conservation prioritization linking OutstandingUniversalValue with specimen condition assessments, (2) temporal analysis tracing HeritageInscription events against taxonomic revisions, and (3) stakeholder network analysis connecting SocialEngagement with institutional Agent roles, operationalizing UNESCO’s recognition criteria within a computable framework.

5.5.2. Object Properties in Heritage Module

Table 10. Core object properties of Heritage Module.

Property Name	Description	Example	Linked Standard
hasOutstandingValue	Links site to OUV	Jeju Volcanic Island → Criterion viii	CIDOC-CRM E18
inscribedAs	Connects value to inscription	OUV → UNESCO World Heritage 2007	LIDO recordWrap
hasAspect	Links value to aspect	Hallasan → Alpine vegetation aspect	LIDO objectWorkType
implementedThrough	Links value to project	Biodiversity → Restoration Project 2020–2025	CIDOC-CRM E7
regulatedBy	Connects to legal instrument	Designation → Cultural Heritage Act Art. 25	CIDOC-CRM E73
involvesStakeholder	Links project to stakeholders	Restoration → Jeju community council	CIDOC-CRM E39

summarizes the principal object properties defined for the Heritage Module.

5.6. Digital Module

The Digital Module structures all technical activities, decisions, and evaluations during three-dimensional digitization as paradata, directly addressing transparent documentation challenges. It integrates W3C PROV-O, which models genealogical relationships through the triad of Activity, Entity, and Agent, with CRMdig, the CIDOC-CRM extension, modeling digitization processes. CRMdig precisely represents Digitization Process (D2), Parameter Assignment (D13), and Digital Object (D1), providing a conceptual foundation for translating EU Eureka3D Quality Framework metrics into an ontological structure. Applied to 197 specimens digitized through robotic scanning, multi-sensor fusion, and AI-based texture reconstruction, the module operationalizes the HR3D workflow, formalizing technical decisions ensuring trustworthiness and reproducibility.

5.6.1. Definitions of Core Classes

Table 11. Core classes of the Digital Module.

Class	Definition and Functional Role
InputProvenance	Traces origins and generation context of raw digitization data. Structures input factors affecting data quality (scanner model, sensor configuration, environmental conditions, specimen preparation state), enabling systematic analysis of how acquisition context influences output fidelity.
AcquisitionEvent	Records the precise moment when digital data are captured from physical specimens. Specifies technical parameters (scan resolution, point-cloud density, capture angles, session duration), supporting reproducibility analysis and parameter optimization through correlation with quality outcomes.
DataProcessing	Documents computational procedures transforming raw data into 3D models. Records processing steps (point-cloud registration, mesh generation, noise reduction, hole filling) with algorithms, parameters, and software versions, enabling verification of processing decisions and replication of workflows.
QualityAssessment	Measures how faithfully digital outputs reproduce original specimen geometry, color, and texture. Performs quantitative evaluations based on Eureka3D quality indicators (geometric accuracy, color fidelity, texture clarity), enabling objective comparison across digitization methods and identification of quality thresholds.
UncertaintyAnnotation	Explicitly records confidence limits arising during digital reconstruction. Visually and semantically distinguishes reconstructed or AI-generated components (inferred surfaces, estimated colors, synthetic textures), quantifying uncertainty sources and degrees to support the critical interpretation of digital assets.
ProvenanceTrace	Establishes traceable networks, enabling backtracking from final digital objects to raw data and intermediate processing steps. Employs PROV-O properties wasDerivedFrom and wasGeneratedBy to construct complete genealogical chains, supporting authenticity verification and reproducibility validation.

define six core classes that establish the ontological structure for the representation of the digitization process. These classes extend PROV-O and CRMdig to support structured documentation of technical parameters, computational workflows, and quality metrics. The integration enables three forms of technical reasoning: (1) uncertainty propagation analysis tracing quality degradation through processing chains, (2) parameter optimization linking acquisition conditions with quality outcomes, and (3) provenance-based validation verifying digital asset authenticity, operationalizing digital specimens as interpretive reconstructions rather than neutral recordings.

5.6.2. Object Properties in Digital Module

Table 12. Core object properties of Digital Module.

Property Name	Description	Example	Linked Standard
usedInput	Links activity to input data	Alignment → Artec raw scan (0.1 mm)	PROV-O used
capturedBy	Connects data to acquisition	Raw images ← Robot scan May 2024	CRMdig D2
processedBy	Links data to processing	Point cloud → ICP registration	PROV-O wasGeneratedBy
assessedBy	Links output to QA	Fused model → RMS 0.23 mm	CRMdig D13
annotatesUncertainty	Links model to uncertainty	Restoration area → AI 68% confidence	CRMdig D1
tracedTo	Traces provenance chain	GLB ← Mesh ← Point Cloud ← Raw	PROV-O wasDerivedFrom

summarizes the principal object properties defined for the Digital Module.

5.7. Ontology-Based Knowledge Graph Implementation

To validate the practical utility of the E-DNH ontology for knowledge discovery beyond data retrieval, we constructed a knowledge graph comprising 1243 nodes and 3856 edges centered on 197 DigitalSpecimen nodes. Cross-module bridge relationships (hasOutstandingValue, digitizedBy, implementedThrough) connecting the Nature, Heritage, and Digital Modules enable semantic reasoning across biological, cultural, and technical dimensions. Graph analysis confirms the structural properties critical for knowledge navigation. DigitalSpecimen nodes function as network hubs (avg. degree 12.3), facilitating centralized access to multi-domain information; bridge edges demonstrate high betweenness centrality (0.42), whose removal fragments the graph into three isolated components, confirming their role as essential semantic connectors; small-world topology ($L=3.2$) enables efficient query traversal across domains; and high clustering (0.68) supports context-based subgraph extraction for domain-specific analyzes.

These structural properties operationalize three research hypotheses that examine how cross-domain integration reveals insights unattainable through isolated data systems, as illustrated in Figure 5. Hypothesis 1 (Biological–Heritage Integration): Taxonomic rarity correlates with heritage designation patterns, predicting conservation prioritization. Among 31 natural monument species, 90% are endangered (IUCN Red List), with the Japanese crane (Grus japonensis, NA0007) exemplifying how TaxonIdentification → HeritageAspect pathways reveal that designation decisions systematically integrate biological rarity with cultural significance, validated through the chi-square test (${\chi }^2=18.7$, $p<0.001$). This shows that heritage value is not arbitrarily assigned, but reflects scientifically grounded biodiversity priorities. Hypothesis 2 (Heritage–Digital Integration): Legal protection status predicts digitization quality specifications. Natural monuments receive 2× scan resolution (0.1 mm vs. 0.5 mm for general specimens) and achieve significantly stricter geometric accuracy (RMS 0.21 ± 0.08 mm vs. 0.42 ± 0.15 mm, $t=7.3$, $p<0.001$), demonstrating how HeritageInscription → QualityAssessment pathways directly translate legal mandates into quantifiable technical specifications. This validates that digitization is not technologically neutral but institutionally governed. Hypothesis 3 (Digital–Nature Integration): Institutional type determines output format differentiation. Educational, research, and governmental institutions produce systematically differentiated outputs through Agent → TechnicalSpecification pathways (ANOVA $F=12.4$, $p<0.001$), optimizing digitization workflows for diverse stakeholder needs (e.g., educational institutions prioritize web-compatible formats; research institutions prioritize high-resolution raw data).

These findings demonstrate that the E-DNH knowledge graph enables hypothesis-driven discovery by revealing causal relationships across domains that remain invisible in siloed data systems, directly addressing the research objective of establishing a unified semantic framework for transparent documentation and knowledge reasoning in natural heritage digitization.

6. Semantic Collaborative Environment for Natural Heritage

6.1. System Architecture Overview

The Collaborative Extended Digital Natural Heritage Platform (C-EDNH) is an intelligent data management framework that integrates the entire lifecycle of natural heritage data—collection, semantic enrichment, and utilization (Figure 6).

Extending the concept of the Digital Knowledge Ecosystem from the Basilica Iulia project to natural heritage, the architecture establishes a collaborative knowledge management environment where multidisciplinary researchers can explore, annotate, and share three-dimensional spatial information. By combining biological, heritage, and technical attributes with 3D digital twins, the platform transcends conventional metadata-centric systems to implement a knowledge graph-based Semantic Knowledge Loop, ensuring findability, accessibility, interoperability, and reusability throughout the data lifecycle. The architecture comprises three layers performing distinct but complementary functions within an integrated semantic framework.

Layer I—Data Acquisition and Digitization Layer

This layer encompasses the measurement-based digitization workflow, referred to as Higher Reality 3D Digitalize (HR3D). It converts physical specimens into digital representations while comprehensively documenting every processing step. Using an integrated workflow combining RealityCapture-based photogrammetry and robotic scanning systems, the platform standardizes high-precision measurement, registration, mesh generation, texture correction, and AI-assisted restoration. All outputs are recorded in standardized data formats and managed through persistent identifiers (NSId/DOI) to ensure long-term traceability and interoperability.

Layer II—Data Integration and Management Layer

This layer integrates and normalizes acquired 3D and textual data within an ontology-based framework. Designed in a Neo4j graph database, it semantically links biological, heritage, and technical attributes of natural heritage specimens. Following CIDOC-CRM, CRMdig, and Darwin Core standards, it defines relationships among data, metadata, and paradata in a triple-layer structure. Through E-DNH Tools, researchers can collaboratively annotate and validate heritage significance, legal protection status, and digitization quality. At this stage, digital representations are transformed into semantically enriched knowledge structures, enabling cross-domain reasoning and validation.

Layer III—Data Collaboration and Application Layer

This layer provides user-facing interfaces for knowledge discovery and collaborative annotation. It supports advanced semantic search, interactive 3D visualization, and community-driven knowledge curation, enabling researchers to query complex relationships (e.g., “Which endangered species in a given region have high-resolution 3D models?”) and contribute annotations that enrich the knowledge graph. The layered architecture supports bidirectional data flow, allowing Layer III annotations to be immediately reflected in the Layer II knowledge graph and guide Layer I scanning priorities. Through this integration, the platform unifies dispersed natural heritage data into a knowledge graph, implementing a collaborative digital ecosystem encompassing biological, heritage, and technical contexts.

6.2. Core Technological Components

This section presents the core technological components of the C-EDNH platform, corresponding to the layered architecture illustrated in Figure 7. These components form an integrated framework that supports 3D data processing, semantic knowledge structuring, and interoperability of international identifiers within a unified workflow.

HR3D(Higher reality 3D digitalization) reproduces natural heritage specimens with ultra-high precision by integrating Gaussian Splatting-based 3D reconstruction with AI-based inpainting based on the Latent Diffusion Model (LDM) (Figure 7a). It includes multi-view point cloud generation; surface-aware noise filtering (feathers, fur, scales); LOD generation; UV mapping; and integrated metadata management. The GS2Mesh architecture achieved significant improvements: the Chamfer Distance error reduced from 0.76 to 0.68 cm, file sizes decreased from 2400 to 310 MB, and the processing speed increased 830 times.

E-DNH tools provide semantic curation and digital asset workflow management (Figure 7b). Researchers collaboratively annotated heritage values, legal status, and biological classification, with annotations immediately reflected in the Neo4j knowledge graph for automatic semantic inference. Role-Based Access Control (RBAC) defines hierarchical permissions, while blockchain-based integrity logging ensures complete traceability.

The Hndle system-based PID infrastructure ensures the long-term traceability and international interoperability of digital specimens. All specimens receive an NSId (Natural Specimen Identifier) linked to DOI via the Handle protocol and synchronized with DiSSCo and GBIF APIs. The OpenSearch-based semantic engine processes multi-dimensional queries across taxonomic and spatiotemporal attributes, while the OAI-PMH protocol harvests metadata from Europeana, converting it to Darwin Core standards for federated semantic management.

7. Performance Evaluation and Validation

This section validates two research hypotheses addressing the practical utility of the proposed semantic framework. H1: The E-DNH ontology enables cross-domain reasoning that reveals causal relationships invisible in isolated data systems. H2: The platform achieves compliance with international FAIR data principles through semantic interoperability. Evaluation involves (1) ontology-based query testing (Section 7.1) demonstrating cross-module reasoning capabilities across Nature, Heritage, and Digital Modules, and (2) FAIR Data Maturity assessment (Section 7.2), quantifying data management maturity using RDA indicators. Together, these results confirm that the platform achieves both logical coherence in its knowledge structure and compliance with FAIR principles.

7.1. Validation of Cross-Domain Reasoning Capabilities

To test H1, we designed four query scenarios examining whether the E-DNH ontology can derive insights requiring integration across biological, heritage, and digitization domains. Using the Cypher query language in a Neo4j graph environment, we validated the ontology’s structural integrity and reasoning capability through representative cases demonstrating cross-modular inference.

7.1.1. Case 1: Causal Relationship Between Heritage Value and Digitization Quality

Research Question: Does heritage significance predict digitization quality specifications?

Query: “What heritage value does the red-crowned crane hold and what is the quality of its digital representation?”

This query integrates the Nature Module (species identification), Heritage Module (heritage value), and Digital Module (digitization quality) to test whether legal protection status causally determines technical specifications. The reasoning chain illustrated in Figure 8 connects biological identification (NA0007) → heritage value assessment → legal protection status → digitization process → quality verification.

Finding: The red-crowned crane, designated as Natural Monument No. 202 with high Outstanding Universal Value, received stricter digitization quality standards (geometric accuracy 0.23mm, completeness 97.8%), significantly exceeding general specimen standards (accuracy 0.42 ± 0.15mm, completeness 89.3 ± 4.2%). This validates the hypothesis that HeritageInscription directly translates into QualityAssessment parameters, demonstrating that digitization is institutionally governed rather than technologically neutral.

7.1.2. Case 2: Taxonomic Influence on Heritage Interpretation Patterns

Research Question: Does taxonomic classification predict heritage aspect emphasis patterns?

Query: “Which heritage aspects are emphasized among carnivorous mammals designated as Natural Monuments?”

This query tests whether biological taxonomy (Nature Module) systematically influences heritage interpretation (Heritage Module). The analysis examined 12 carnivorous mammal specimens across three families (Mustelidae, Felidae, Canidae).

Finding: While all carnivorous Natural Monuments emphasize “ecological significance,” family-level variation emerged: Mustelidae species (e.g., Eurasian otter) emphasize scientific research value in aquatic ecosystem studies (83% of cases), whereas Felidae species (e.g., Amur leopard) emphasize cultural symbolism and humanistic interpretation (75% of cases). Figure 9 demonstrates that taxonomic hierarchy serves as a structural determinant of heritage value semantics, revealing how biological classification informs cultural interpretation patterns—a relationship invisible in non-integrated systems.

7.1.3. Case 3: Multi-Criteria Prioritization in Digitization Workflows

Research Question: How do overlapping conservation statuses (IUCN + national designation) influence digitization priority decisions?

Query: “Which endangered species are held by Seoul Grand Park, and how are digitization priorities determined?”

This query examines whether institutions systematically integrate biological rarity (Nature Module) with legal protection (Heritage Module) to guide digitization workflows (Digital Module).

Finding: Seoul Grand Park implements a two-tier prioritization algorithm; the primary criterion follows IUCN categories (EN → VU → NT) and the secondary criterion elevates specimens with dual designation (IUCN + Natural Monument). Swan goose and red-crowned crane, possessing both EN status and Natural Monument designation, received highest priority (processed within 6 months vs. 18-month average). Figure 10 validates that HeritageInscription information operationally determines AcquisitionEvent sequencing, demonstrating evidence-based resource allocation in digitization projects.

7.1.4. Case 4: Project-Based Quality Assessment and Accountability

Research Question: Can heritage conservation project outcomes be quantified through digitization quality metrics?

Query: “What is the average quality of the digital specimens produced through the Crane Restoration Project?”

This query tests whether HeritageProject (Heritage Module) outcomes can be systematically evaluated through QualityAssessment (Digital Module), enabling an evidence-based assessment of conservation initiatives.

Finding: The Crane Restoration Project (2020–2025) generated 23 digital specimens with average geometric accuracy of 0.19 ± 0.05mm and texture fidelity (SSIM) of 0.94 ± 0.02, exceeding institutional benchmarks. Figure 11 demonstrates that the cross-module relationship between HeritageProject and DataProcessing enables translation of conservation project outcomes into quantifiable digital quality indicators, supporting an evidence-based evaluation of heritage initiatives.

Summary: These four cases validate H1 by demonstrating that the E-DNH ontology enables hypothesis-driven reasoning revealing causal relationships (heritage value → digitization quality); structural patterns (taxonomy → heritage interpretation); operational algorithms (conservation status → prioritization); and quantifiable outcomes (projects → quality metrics) that remain invisible in siloed data systems.

7.2. FAIR Data Maturity Evaluation

To test H2, we quantitatively assessed platform compliance with FAIR principles using the RDA FAIR Data Maturity Model (RDA-FDMM), which evaluates Findable, Accessible, Interoperable, and Reusable dimensions across 41 indicators. This evaluation examines whether the semantic framework achieves international data management standards. The assessment involved (1) mapping E-DNH data structures to RDA-FDMM indicators; (2) evaluating implementation levels (Essential, Important, Useful) across five maturity stages (Level 0: Not Applicable → Level 5: Fully Implemented); and (3) calculating weighted FAIRness scores. Figure 12 visualizes implementation status across all 41 indicators.

Results and Interpretation

The platform achieved a FAIR maturity score of 0.88 (87%), exceeding the project target (0.83) and significantly surpassing typical institutional repositories (0.65–0.75). As shown in Figure 12a, 37 of 41 indicators (90%) reached the fully implemented level, demonstrating systematic FAIR compliance.

Dimension-specific analysis reveals both strengths and areas for refinement. Findability achieved Level 3; while manual NSId assignment ensures 100% persistent identifier coverage, automated DOI integration (RDA-F1-01D, RDA-F1-02D) remains at the planning stage. Accessibility and Interoperability both attained Level 5, the highest maturity, through standardized API protocols (HTTP/S, OAuth2, OIDC) and robust integration with global infrastructures (GBIF, GenBank) via Darwin Core, CIDOC-CRM, CRMdig, and PROV-O alignment. Reusability achieved Level 3-4; technical mechanisms are fully operational, but license metadata policy (RDA-R1.2-01M) requires refinement for full COAR and OpenAIRE compliance.

These results validate Hypothesis H2, confirming that the semantic framework achieves FAIR-by-Design compliance through systematic integration rather than post hoc supplementation. High scores in Accessibility and Interoperability demonstrate technical effectiveness, while remaining gaps in Findability and Reusability concern governance policies (DOI automation, open licensing) rather than technical capabilities, addressable through administrative optimization. The radar visualization in Figure 12c illustrates balanced maturity across dimensions. Comparatively, the 0.88 score positions the platform among the top 10% globally, comparable to DiSSCo (0.86) and iDigBio (0.84), validating practical viability for institutional deployment.

8. Discussion

This study validated the proposed framework using 197 natural heritage specimens (140 bird, insect, mammal, and reptile species) collected through collaboration with Hannam University Museum of Natural History, National Research Institute of Cultural Heritage, and Seoul Grand Park. While the dataset demonstrated interoperability among heterogeneous institutional structures, taxonomic diversity remains limited, excluding herbarium sheets, geological samples, and microscope slides due to optical constraints. Future studies should develop adaptive protocols—polarization filtering, HDR-based texture correctio, hybrid illumination scanning and integrate WebGL-based 3D annotation APIs with automated RDF triple generation to enable interactive semantic annotation currently unrealized in PROV-O-based paradata documentation.

A critical limitation concerns cross-framework interoperability validation. Although the E-DNH Ontology successfully integrates standards, compatibility with GBIF IPT and BioCASe remains empirically untested. Future work must conduct federated query testing across institutional boundaries to validate semantic alignment consistency. Furthermore, this study does not sufficiently address the threshold of semantic alignment necessary for stable reasoning. Future research should investigate link-only integration modes, where heterogeneous metadata are connected via PIDs without strict prior alignment, allowing AI-based reasoning to learn latent correspondences while maintaining provenance-based accountability.

9. Conclusions

This study introduces an integrated framework redefining digital natural heritage as a collaborative and interpretive ecosystem. The Collaborative Extended Digital Natural Heritage Platform (C-EDNH), built on the E-DNH ontology and HR3D workflow, enables diverse participants to explore, annotate, and reinterpret natural heritage within an interactive 3D environment, transforming static data into a dynamic semantic network.

The framework achieves four key contributions. Firstly, it establishes a multilayered semantic integration model bridging Darwin Core (biological data), CIDOC-CRM (heritage value), and CRMdig (digitization process), enabling specimens to function as both scientific evidence and cultural heritage. Secondly, it operationalizes hyperreality as quantitative quality management (RMS 0.18 ± 0.09 mm; $\Delta E_{00}$ 2.1 ± 0.7), establishing measurable standards for reproducibility. Third, through PROV-O-based paradata documentation, it systematically structures technical decisions within shared prov/Activity frameworks, enabling transparent accountability and critical reassessment. Fourthly, it validates cross-domain reasoning capabilities through hypothesis-driven queries, demonstrating how heritage designation predicts digitization quality specifications and taxonomic classification influences heritage interpretation patterns—insights invisible in siloed data systems.

Future work will expand specimen diversity beyond current taxonomic constraints, enhance 3D semantic annotation capabilities, and validate global interoperability through integration with GBIF, DiSSCo, IIIF, and Linked Art infrastructures. Ultimately, this framework advances digital heritage management from isolated technical replication to an open, intelligent, and human-centered collaborative model, establishing a foundation for transparent and scientifically accountable digitization practices.