Application of 4D Technologies in Heritage: A Comprehensive Review

Abstract

Three-dimensional (3D) modelling has become essential in the heritage field, but static models cannot capture how heritages evolve over time. To this end, more and more studies have presented four-dimensional (4D, i.e., 3D + time) applications. However, there is no systematic overview. Hence, this study provides a comprehensive review of 4D technologies applied to heritage. Through a two-stage Web of Science retrieval, 56 studies (2000–2025) were analysed, covering 3D and heritage building information modelling (HBIM) construction, time incorporation, calibration, and interaction. This review explicitly focuses on how time is incorporated into survey-anchored 3D and HBIM models, and how calibration ensures the reliability of diachronic interpretation. The review distinguishes between survey-anchored 3D and HBIM pipelines, outlines diachronic modelling and calibration workflows, and summarises emerging extensions such as depth sensing, temporal lensing, and time traceback. Moreover, it also provides a mini-review of five-dimensional (5D) applications and their future directions inspired by the 4D overview. The specific contributions of this study are threefold: (i) it synthesises a complete and comparable workflow for integrating time into ordinary 3D models and HBIM; (ii) it formalises calibration principles and reportable metrics for trustworthy temporal reconstruction; and (iii) it clarifies how emerging 5D extensions build on 4D practice rather than replace it.

1. Introduction

Three-dimensional (3D) geometric modelling has now become foundational to heritage science [1,2]. Reality-based surveys, such as terrestrial laser scanning [3,4], mobile/airborne LiDAR [5,6], and image-based photogrammetry [6], now allow architects, conservators, and curators to document objects, buildings, and landscapes with metric fidelity, to diagnose condition and risk, and to communicate findings with wider publics [7,8]. Heritage building information modelling (HBIM) goes further by coupling surveyed geometry with materials, stratigraphy, and relationships [9,10]. Yet gaps persist. In particular, static 3D snapshots struggle to explain change—what altered, when, and where [1,11].

Four-dimensional (4D) technologies address these limits by adding an explicit fourth axis on top of survey-anchored 3D models. Diachronic models organise evidence into dated, source-labelled states within a single spatial frame [12,13], enabling trustworthy comparison across periods and audiences. While 4D applications have grown rapidly [14,15,16], no review synthesises the use of 4D technologies specifically for heritage. As detailed in Section 2, our search identified three review papers addressing 4D technology and heritage. A tourism-centred review of Australian georegions reports heritage contexts but privileges visitor experience rather than heritage modelling [17]. A scoping review of digital techniques in historic gardens surveys 3D and lists ‘4D documentation systems’ only as a future opportunity [18]. A systematic review of virtual restoration similarly projects ‘4D restoration’ as a next step [19]. Furthermore, five-dimensional (5D) applications are also quietly emerging [20,21]. As detailed in Section 2, our search identified only one review paper addressing 5D technology and heritage. This review focuses on the 5D diffraction imaging of functional materials, not heritage practice [22]. Therefore, taken together, these works confirm the momentum of 4D/5D technologies in heritage but also confirm the absence of a comprehensive, systematic review of 4D/5D applications in heritage.

Accordingly, this article offers the first systematic overview of the applications of 4D/5D technologies for heritage. Section 2 details a two-stage Web of Science retrieval that yields a focused corpus. Section 3 depicts ordinary 3D and HBIM pipelines and formalises calibration principles. Section 4 details how time is incorporated: (i) diachronic, survey-anchored scenes for ordinary 3D and (ii) in-model phasing for HBIM, with practical calibration checklists and cases. Note that this article does not compare survey-anchored 3D and HBIM approaches against each other, but synthesises them under a shared analytical frame—construction of geometry, incorporation of time, and calibration—to describe their respective workflows in a consistent manner. Section 5 broadens the fourth dimension beyond time, including depth sensing, temporal lensing, and time traceback. Section 6 synthesises limitations and pragmatic remedies. Section 7 provides a mini review of 5D technologies and further clarifies future directions inspired by the comprehensive review of 4D applications.

This review makes three contributions. Firstly, it synthesises a complete and comparable workflow for integrating time into both ordinary 3D models and HBIM. Secondly, it formalises calibration principles and reportable metrics to support trustworthy temporal reconstruction. Thirdly, it clarifies how emerging 5D extensions build on 4D practice rather than replace it.

2. Literature Retrieval

A two-stage retrieval in the Web of Science Core Collection (All Editions) was conducted to assemble a focused corpus on 4D/5D technologies applied to heritages, with publication years restricted to 1 January 2000 through 30 June 2025. The first round applied the identical Topic Search string (heritage AND “4D/5D” and variants) while limiting the Document Type to Review Article and returned four review papers without screening. The same search string was then used without the document-type restriction in the second round, yielding 190 records. Titles, abstracts, and conclusions (when necessary) were screened to determine relevance, retaining studies that explicitly addressed cultural heritage and used survey-anchored 3D geometry as the spatial basis.

Records were excluded when “4D/5D” referred to concepts unrelated to heritage modelling (e.g., 4D printing [23], 5D transportation frameworks [24], or dimensional feature vectors [25]), and studies in which 4D did not represent diachronic states of a heritage object were also removed. Following this selection process, 56 eligible articles were retained for full-text analysis. To avoid missing relevant literature, the same search string was cross-checked in Scopus during revision, and no additional studies meeting the above criteria were identified.

3. Three-Dimensional Models: Construction and Calibration

3.1. Ordinary 3D Models: Construction and Cases

3.1.1. Construction Methods of Ordinary 3D Models

Ordinary 3D models are digital representations of heritage objects, buildings, and sites, mainly focusing on geometry [26,27]. These models are created using laser scanning point clouds or image-based reconstructions and are typically meshed and textured to portray the current or historically reasoned states [28,29]. They do not encode building semantics, offering simple geometric shapes and surfaces that can be visualised, measured, and inspected [30]. These models serve as a bridge between expert analysis and public communication, enabling architects and conservators to interact with the geometry at various scales, while also providing the general public with photorealistic views based on survey data.

There are two primary methods for acquiring the data used in ordinary 3D models: active and passive systems.

Table 1. Acquisition methods for ordinary 3D models.

Method	Outputs	Notes
TLS and A/MLS	Dense point clouds	Highest geometric fidelity; occlusion and cost limits
Photogrammetry (SfM + DIM, incl. UAV)	Textured meshes, dense clouds	Rich texture; scale, lighting, and control requirements
Hybrid (scan + image)	Co-registered clouds/meshes	Complements strengths; alignment QA and uncertainty tags
Reverse modelling	Survey and archival inference	Reconstructs losses; flag hypothetical elements clearly

Note: UAV and QA represent unmanned aerial vehicle and quality assurance, respectively.

shows a concise comparison. Active systems, such as terrestrial laser scanning (TLS), measure surfaces using time-of-flight or phase measurements, producing precise point clouds of façades, interiors, and complex remains [18,31]. For larger areas, airborne or mobile laser scanning (A/MLS) can cover extensive terrains and streetscapes [32,33]. Passive methods, like photogrammetry, use multiple images processed through Structure-from-Motion (SfM) to create sparse scenes, followed by dense image matching (DIM) for detailed point clouds [34,35]. The data from both methods are often combined; laser scans provide stable geometry, while images contribute textures and fill gaps in hard-to-reach areas [36]. This combination ensures that all data is aligned within a single reference frame, ensuring accuracy.

When heritages are fragmented or lost, reverse modelling is used, combining surveyed data with documentary sources like historical maps or photographs [37,38]. In these cases, the modelled elements that cannot be physically measured are flagged as hypothetical, keeping a clear distinction between what has been surveyed and what is inferred [39]. This method allows a thorough and transparent representation of heritages, ensuring that interpretive assumptions are clearly noted.

3.1.2. Construction Cases of Ordinary 3D Models

In the Ávila medieval wall, laser scanning and photogrammetry were combined to create a cohesive model of gates, towers, and curtains [12,40], as shown in Figure 1a–c. The historic centre of Calw used a similar approach, merging laser scanning and aerial imagery to document street-level and roof structures [11,41,42,43,44]. At the Alphonse Raymond Factory in Montreal, a 3D model was created from contemporary photography, archival photos, and maps, with missing sections marked as hypothetical [45]. The San Giovanni in Conca in Milan used integrated laser scanning and imagery to capture fine architectural details [46]. Lastly, the Orígens Geopark in the Pyrenees utilised UAV-based photogrammetry for large-scale geological modelling, creating models for educational purposes while retaining the survey-based core [47], as shown in Figure 1d.

3.2. HBIM: Construction and Cases

3.2.1. Construction Methods of HBIM

Heritage building information modelling (HBIM) is a 3D model enhanced with detailed data that defines components, materials, and their relationships [48,49]. Unlike ordinary 3D models, which focus on basic geometry, HBIM integrates both geometric and non-geometric information [50,51]. This includes data about historical materials, construction techniques, and conditions, providing a more complete representation of the structure [48,52].

Constructing HBIM begins with an in-depth survey, especially for areas that are difficult to access. For example, UAV videogrammetry captures both exteriors and hard-to-reach interiors [53,54]. The resulting images are processed into a dense point cloud, which is refined by removing irrelevant data and adjusting the density using CloudCompare [55,56]. This processed data is then converted into an RCS-type file in Autodesk ReCap and imported into Autodesk Revit for modelling. In general, instead of using standard catalogue parts, custom families are created to represent architectural elements and materials, including any observed damage or decay [57]. These custom families help to ensure the model accurately reflects the building’s current state [57]. A concise HBIM workflow is listed in

Table 2. HBIM construction workflow.

Stage	Key Actions and Tools
Survey	UAV photogrammetry or TLS; capture exteriors and inaccessible interiors
Pre-processing	Noise removal, decimation, registration; CloudCompare
Point-cloud to BIM	ReCap to .rcs; import into Revit or ArchiCAD
Object modelling	Custom families; parametric profiles; materials/stratigraphy; damage mapping
Semantic enrichment	Attributes: chronology, techniques, pathologies; link docs

.

3.2.2. Construction Cases of HBIM

At the former Segrè Papermill (Tivoli), UAV videogrammetry captures both the building’s exterior and difficult-to-access interiors [16]. The point cloud is curated to focus on architectural details before being imported into Revit, where the model is created using custom families for both the building and machinery. At the Sanctuary of Hercules (Tivoli) in the same region, the same survey and modelling process is applied [16]. Differently, UAV imagery was used to document restricted areas. A large-scale HBIM including both heritages is shown in Figure 2.

3.3. Ordinary 3D Models and HBIM: Calibration and Cases

Across the studies, calibration approaches were classified using three recurring elements that appear regardless of scanning or modelling technique: (i) the geometric reference, (ii) the method used to achieve spatial fit, and (iii) the way uncertainty is communicated. Evaluation follows the same approach; studies are read in terms of whether they anchor geometry to a stable reference, explain how the fit is achieved, and make uncertainty explicit rather than implicit. This provides a consistent basis for comparing calibration workflows across heterogeneous cases without enforcing a shared toolset or accuracy benchmark.

3.3.1. Calibration Methods of 3D Models

Calibration for both ordinary 3D models and HBIM ensures alignment with survey data and a unified spatial reference [58]. A concise summary of the calibration workflow and targets is provided in

Table 3. Core calibration methods for ordinary 3D models and HBIM.

Method	Data/Inputs	Typical Outputs	Accuracy Goal
Helmert + ICP	Multi-sensor point clouds	Unified reference; refined alignment	≤5 cm; SD < 10 cm
TLS planning/targets	Scan stations; common targets	Clean registration	≤2–3 cm locally
Reverse-modelling	Historic drawings/photos; spatial invariants	Adjusted geometry	Case-dependent (~5–10 cm)
HBIM cloud-to-model	Revit + point cloud; CloudCompare	Deviation maps	−0.05 to +0.05 m; SD < 0.10 m

Note: SD represents standard deviation.

.

For ordinary 3D models, the process begins by merging different survey types, such as laser scans and image-based reconstructions, into a single reference system using a seven-parameter Helmert transformation [59,60,61]. This adjustment covers translations, rotations, and scale. If needed, iterative closest point (ICP) is used to refine the fit, especially when detailed 3D features are present [62]. TLS workflows further benefit from planning scan-station locations and using common targets [63,64], while façade geometry can be refined interactively using camera-matched images [11].

When historical data is used for reverse-modelling, adjustments are made based on spatial invariants like wall widths and distances, with historical drawings adapted to match extant structures [46,65,66]. Registered historic photographs provide additional validation by comparing viewpoints with the surveyed geometry [6,67,68].

For HBIM, the calibration process is divided into two phases. First, survey data is cleaned and filtered using tools like CloudCompare, then converted into formats suitable for parametric modelling in Autodesk Revit [69,70]. Second, cloud-to-model distance analysis is performed within Revit to assess deviations [69,70]. In theory, the cloud-to-model checks can ensure that the model accurately matches the point cloud, keeping the deviation within an acceptable range. Generally speaking, successful calibration is marked when deviations remain within −0.05 m to +0.05 m, with a SD below 0.10 m [71,72].

3.3.2. Calibration Cases of 3D Models

In Calw (Germany) [11,41,42,43,44], a multi-sensor workflow fuses TLS and image-based point clouds. Datasets are first brought to a common reference frame with a seven-parameter Helmert transform, then refined by ICP, improving façade and roof continuity for 3D modelling of the historical core area/urban core range.

In Ávila (Spain) [12,40], an A/MLS survey provided the Alcázar Gate point cloud. Eighteenth-century drawings were vectorized and non-uniformly adjusted to the extant fabric; relative positioning was used to reconstruct missing elements. Historic photographs informed assumptions (e.g., wall thickness). Signed difference evaluations of the MLS point cloud and the current state model were performed, and the following statistics were reported: mean −0.026 m and SD 0.121 m.

In Tivoli (Italy) [16], UAV videogrammetry produced dense point clouds that were cleaned in CloudCompare and linked to Revit for HBIM. Calibration relied on cloud-to-model distance analysis; most deviations were ≤0.05 m, with larger residuals around stressed vaults and similar zones.

4. Incorporation of Time as the Fourth Dimension into Three-Dimensional Heritages

4.1. Incorporating Time into Ordinary 3D Models: Approaches and Exhibition

4.1.1. Incorporation Approaches for Ordinary 3D Models

Time as the fourth dimension can be added to ordinary 3D models by creating distinct period-specific states that all share a common spatial baseline [50,73]. This baseline is derived from present-state models obtained through field methods, like TLS or A/MLS close-range photogrammetry, and UAV imagery (if necessary) [50,74]. These techniques ensure that all elements, such as façades, roofs, interiors, and landscapes, align within a single, unified reference [36]. Time is then represented by creating discrete scenarios tied to historical sources [75,76]. Rather than simulating continuous change, the approach presents clearly defined stages that users can switch between, compare, and analyse [36,76]. The procedures for diachronic 4D modelling from present-state baselines are summarised in

Table 4. Minimal protocol for diachronic 4D modelling from present-state baselines.

Step	Evidence	Action and Traceability
Baseline capture	TLS, A/MLS, UAV imagery	Georeference; full-envelope coverage; report accuracy.
Source gathering	Photos, plans, maps, texts	Log date/provenance; rate reliability.
Registration	Align sources to baseline	Control points; lens/scale correction; record residuals.
Reconstruction	Build period-state geometry	Separate modelled vs. measured; tag uncertainties.
Packaging and QA	Switchable scenes; view-matched photos	Versioned outputs; replicate viewpoints; label sources.
Principle	Transparency	Keep reconstructed geometry segregated; show source labels in each scene.

.

Historical photographs play a key role in conveying period-specific details [77]. When multiple photos exist, they provide various perspectives, enriching the model’s accuracy [78]. In the case of limited photographs, each one serves as a keyframe, helping represent a particular time period [78]. These photos are always interpreted in relation to the present-state model, not in isolation, ensuring consistency and enabling the replication of viewpoints for verification.

Historic plans, drawings, and maps provide another important input [21,79]. These documents are digitised and adjusted to fit the measured remains, particularly when distortions or scaling issues prevent direct use [80]. Crucially, reconstructed geometry is always kept separate from the observed fabric in the final outputs [36,81]. This transparency allows readers to distinguish between measured structures and those added based on historical research, maintaining credibility and providing clarity about the sources of uncertainty.

To ensure the accuracy of this process, the present-state baseline is treated as a critical foundation [82]. This requires comprehensive scanning and imaging to cover all parts of the building, including rooftops and hard-to-reach areas like narrow streets [83]. Once the baseline is established, period states are packaged as switchable scenarios within the same 3D environment.

4.1.2. Four-Dimensional Model Cases

The Ávila (Spain) case [12,40] combines a present-state model with historical plans and photographs. The model is captured through mobile laser scanning with a resolution of around 60 mm. Historical drawings are adjusted relative to the measured wall segments, and historical photos are used to check the viewpoints. A timeline interface allows users to explore different periods, with clear distinctions between observed remains and reconstructed additions.

The Alphonse Raymond Factory (Montreal, Canada) case [45] uses a survey-based approach to narrate the building’s evolution across the 20th century. The present-state model is complemented by archival photos, plans, and maps to create period states. The distinction between measured fabric and reconstructed elements is made clear, offering a transparent view of the building’s changes over time.

For the partial remains of the Basilica di San Giovanni (Conca, Milan, Italy) [46], a high-resolution present-state model is built using TLS and photographic techniques. Historical photos and texts guide the creation of earlier phases. The evolution of the basilica with time is presented in Figure 3. Reconstructed elements are kept distinct from what is still standing.

The Orígens Geopark, South-Central Pyrenees (Spain) case [47] pairs 3D models of key sites with reconstructions of past landscapes. Each reconstructed stage is tied to the same spatial baseline as the present-day model, ensuring visitors can see the changes over time while keeping the spatial baseline intact. This is a remarkable 4D work because it presents to us the changes that have occurred over nearly 600,000 years in the Orígens Geopark (see Figure 4).

The Historic Centre (Calw, Germany) case [11,41,42,43,44] uses TLS, close-range, and aerial imagery to create a complete georeferenced baseline. Limited archival resources result in a diachronic presentation restricted to areas where solid evidence is available, ensuring that the visualisation remains faithful to the evidence.

These cases follow the methods described in Section 4.1.1 (see

Table 5. Summary of 4D model cases and their diachronic setup.

Cases	Methods	Sources
Ávila	MLS (~60 mm)	Photos, plans; view replication; separate reconstructions.
Alphonse Raymond Factory	Surveyed model	Archival photos, plans, maps; transparent reconstructions.
S. Giovanni in Conca	TLS + photogrammetry	Texts, photos; earlier phases flagged.
Orígens Geopark	Georeferenced sites	Landscape stages tied to shared baseline.
Calw	TLS/aerial/close-range	Limited archives; visualise only evidenced areas.

), including (1) completing the present-state baseline, (2) adapting historical sources relative to measured remains, (3) keeping reconstructed geometry distinct from observed fabric, (4) using historical imagery to verify viewpoints, and (5) presenting period states as clearly defined, source-labelled scenes. This pipeline ensures a reliable and easy-to-follow diachronic interpretation that remains transparent and consistent across various sites.

4.1.3. Four-Dimensional Model Exhibition

Incorporating time into ordinary, survey-anchored 3D models serves a single purpose: to show change as a series of dated, source-labelled states while keeping one stable spatial frame. Period scenes switch within the same coordinates, so orientation, distance, and viewpoint remain constant even as gates, walls, churches, factories, or geoheritage outcrops appear differently across time. This makes three questions easy to answer—what changed, when, and where—without losing credibility.

Four-dimensional web viewers put the time dimension on top of a measured three-dimensional baseline in the browser [84]. In Ávila (see Figure 5), the Alcázar Gate and walls are shown as dated scenes controlled by a timeline; clicking hotspots opens period photographs and plans exactly at the gates or towers they describe, often from matched viewpoints [12,40]. In-scene labels keep the experience transparent, so visitors see at a glance what is surveyed today and what is reconstructed. Because everything stays in one georeferenced frame, people can compare periods without losing where they are.

Apps carry the same 4D packages to phones, tablets, kiosks, and museum rooms [85]. In Calw, the Calw VR app (see Figure 6a) uses a storyboarded menu and a time slider [41,44] so visitors move a façade or square through key dates with one gesture; audio narration and short texts appear at the right spots in the scene, while a map widget and “time buttons” jump between buildings and periods. A companion mobile route, Tracing Hermann Hesse in Calw (see Figure 6b), leads people along a city path tied to the writer’s childhood; along the route, textures and captions cue “past” versus “present”, helping non-experts follow the story on site [41,44]. SmartWall, for the Medieval Wall of Ávila, combines 360° panoramas, 3D/4D content, and a geospatial database in one web app configured for different audiences [40], as shown in Figure 6c; experts can drill into time states and sources, while students and tourists receive filtered views and guided virtual tours, the same packages being reused in local museums. All of these apps read the same georeferenced baseline and period bundles, so orientation and provenance stay consistent across devices [68].

4.2. Incorporating Time into Ordinary 3D Models: Calibration and Cases

4.2.1. Calibration Methods of Ordinary 3D Models with Time

This subsection defines calibration for time-labelled scenes built from ordinary 3D surveys (laser scanning and/or image-based reconstruction, no encoded semantics). One rule guides everything: use a single present-state baseline, make each historical scene obey spatial invariants measured on surviving fabric, show uncertainty inside the model, and re-check every slice against the same invariants [86,87]. A concise calibration checklist and reportable metrics are summarised in

Table 6. Calibration checklist and reportable metrics for time-labelled ordinary 3D scenes.

Step	Key Action Under Invariants	Reportable Metrics
Baseline co-registration	Helmert + ICP; station planning; targets	global/local RMS; control residuals
Document adaptation	Digitise drawings/maps; adapt to invariants	invariants values ± tolerances
Viewpoint registration	Align archival photos; check silhouettes/proportions	reprojection error; coverage notes
Uncertainty/meshing	Distinguish reconstructed vs. measured; multi-resolution	flagged extents; effective resolution
Iterative validation	Reapply invariants; record mismatches	per-slice residuals; excluded areas
Reporting	Summarise methods, sources, limits	RMS, resolution, invariants list

.

The initial phase involves baseline co-registration. Heterogeneous point clouds from TLS, A/MLS, or photos are fused with a seven-parameter Helmert transform and further refined with ICP, where surfaces allow [88,89]. Stations are planned and targets are used to stabilise local networks and limit drift; for façades, camera-matched rectification keeps local patches consistent with photographs [90]. Practical limits are stated up front; in Ávila, distance and coverage yielded an effective spatial resolution of about 60 mm at ~25 m [91].

The next step concerns document adaptation under invariants. Historic drawings, plans, and maps rarely scale uniformly. Therefore, they are digitised and adapted to the survey using spatial invariants observed on surviving fabric: stable wall widths, enclosure extents, repeatable distances, and angles [92]. Reverse-modelled parts can move only within ranges that those invariants permit. This preserves traceability to what remains on site and prevents geometric drift between time slices [87].

Subsequently, archival photographs are registered as viewpoints rather than loose illustrations. After drawings are adapted, the images are registered to the baseline so original vantages can be reproduced [19]. Those viewpoints are further used to check proportions, silhouettes, and visible extents for each period. If the archive is thin or the viewpoints are few, diachronic scenes are restricted to areas that the sources genuinely support [93,94].

A further essential step is making uncertainty legible. Geometry derived from documents is kept visually distinct from measured surfaces so that users can see the line between observation and historical reasoning [95]. Multi-resolution meshing is also used: dense sampling where preserved fabric provides the invariants that constrain adaptation, lighter sampling in a broader urban or landscape context. This protects fidelity at the anchors and keeps models efficient elsewhere [91].

The process continues with iterative validation and residual recording. After assembling each temporal slice, the same invariants are re-applied and the residuals are checked. When mismatches persist because of occlusions, ambiguous drafting, or limited photographic coverage, these mismatches are documented and the visual distinction is kept rather than forcing agreement [89]. The deliverable is a family of time-labelled states, each tied to named sources, all coherent with the same survey-anchored baseline.

The procedure concludes with targeted reporting. Effective resolution and registration RMS (local and global) are reported, the invariants used with measured values and tolerances are listed, and any deliberate limits on scene extent where evidence is weak are noted. These short disclosures are more useful than a single headline accuracy number and help subsequent conservation work [95].

4.2.2. Calibration Cases of Ordinary 3D Models with Time

In Ávila, Spain, the Alcázar Gate case [12,40] establishes an A/MLS baseline that frames all diachronic states. At ~25 m standoff, distance and coverage yield an effective spatial resolution of ~60 mm. Eighteenth-century drawings are vectorised and adapted under spatial invariants observable on the surviving masonry (e.g., arch spans, wall thickness, enclosure extents). Archival photographs are registered as viewpoints to check the silhouettes and proportions. Document-derived geometry remains visually distinct from measured surfaces. As shown in Figure 7, signed-difference comparisons between the current-state model and the MLS cloud are reported (mean −0.026 m, SD 0.121 m).

In Calw, Germany [11,41,42,43,44], the historic-centre workflow combines TLS for street-level capture with close-range and aerial imagery to generate dense image-matching clouds for roofs and courtyards. Heterogeneous point clouds are brought into a single frame by a seven-parameter similarity (Helmert) transform and are aligned by ICP wherever surface patches allow. Station planning mitigates drift in irregular street canyons, and façade rectification is kept consistent with photographs. Because archives are thin, the temporal set remains selective in time but spatially continuous across streets or blocks supported by sources, while reconstructed areas are explicitly flagged.

In Milan, Italy, a study of the Basilica di San Giovanni in Conca [46] documented extant remains with TLS and augmented the fine detail with triangulation-based scanning and calibrated DSLR imagery suited to a confined archaeological context. Registration relies on recognisable 3D features and ICP rather than targets. Multi-resolution meshing concentrates sampling where preserved fabric provides invariants and relaxes it in the surroundings. Document-derived elements remain visually distinct, and residual mismatches after re-applying invariants are recorded rather than cosmetically removed.

In Montreal, Canada, the Alphonse Raymond Factory [45] adopts a present-state survey as baseline and registers archival photographs, insurance maps, and drawings to that frame. Adaptation stays within ranges permitted by measured invariants; viewpoint-anchored checks support each phase. Where documentation is thin, the diachronic set is strictly curated to the record, and reconstructed elements are flagged and kept separate from measured surfaces.

4.3. Incorporating Time into HBIM: Approaches and Cases

4.3.1. Incorporation Approaches for HBIM

Time is modelled inside a single HBIM rather than split across multiple files [96]. Named historical phases are created in Autodesk Revit, and phase filters control what appears in each view [15]. Elements—walls, floors, machines—are assigned to the phases when they existed [97]. Switching phases changes visibility only; the coordinate system, dimensions, and view orientations stay fixed. Users can move through the site’s history without duplicating geometry and without losing a common measurement baseline [96,98]. Key steps and their rationale are summarised in

Table 7. HBIM time-integration workflow.

Component	Action	Outcome
Historical phases and filters	Define phases in Revit; apply filters	Time-sliced views in one model
Heritage families	Author parametric families for extant/reconstructed states	Object-level change captured
Point-cloud scaffold	Link consolidated survey cloud before authoring	Stable, shared coordinates across phases
Evidence links	Attach drawings/photos to elements	Transparent provenance in-model
Appearance overrides	Differentiate reconstructed vs. surviving parts	Readable visuals for experts/public
Historical phases and filters	Define phases in Revit; apply filters	Time-sliced views in one model
Heritage families	Author parametric families for extant/reconstructed states	Object-level change captured

.

Change is also described at the object level. Instead of generic libraries, teams author heritage-specific parametric families that follow real morphology, materials, and construction techniques [44]. In industrial archaeology, one family can hold two workable states in the same HBIM: the current, extant remains and a reconstructed last-operational configuration grounded in machine drawings, photographs, and other records [16,99]. Family instances are then tied to phases so the model can show the final production stage or an earlier arrangement while keeping a clear distinction in the model between what survives and what is reconstructed [100].

All views remain metrically anchored by a consolidated point cloud that is linked to Revit before phasing and family work begins. The cloud acts as a shared geometric scaffold [57]. Because authoring proceeds on that scaffold, phase-filtered views line up across time; distances and alignments are comparable, and nothing needs to be shifted or rescaled when the period changes [101,102]. This common baseline supports consistent measurement, annotation, and discussion across the diachronic views.

Evidence is kept with the geometry. Machine drawings, photos, and related records are linked to informative elements so that the basis for each reconstruction is available at the point of use when browsing phases. When visualising the last production stage, appearance overrides governed by phase filters make reconstructed equipment look different from surviving parts. The visual story stays readable for a general audience while the reasoning remains transparent for conservation practice [103].

In practice, the method keeps everything inside BIM. Phases and phase filters organise time, heritage families express object-level change, the point-cloud scaffold keeps all states in one coordinate frame, and in-model links tie each reconstructed element to its sources. Because time is handled as controlled visibility on top of one survey-anchored model, people can read “what changed, when, and where”, without moving to another file or losing the metric reference that underpins the survey [104].

4.3.2. Incorporation Cases

Two coordinated HBIM case studies at Tivoli apply the approach in the same order of operations [16]. The evolution of HBIM of the former Segrè Papermill and the Sanctuary of Hercules with time is shown in Figure 8.

At the former Segrè Papermill [16], a comprehensive point cloud is linked first to serve as the geometric base. A sequence of named phases reflects historically attested stages of the plant. Architectural fabric and industrial machines are attributed to those phases, so a single model can display the last production arrangement or an earlier configuration without altering coordinates. Families encode both the current remains and the last operational reconstruction grounded in technical documentation. In the final production phase, filters and appearance settings distinguish reconstructed machines from surviving components in the same space. Source material is linked to the elements it supports, so the justification for each reconstruction is visible while inspecting phase-filtered views. Because every view is tied to the same point-cloud scaffold, distances and alignments stay comparable when switching phases.

The Sanctuary of Hercules follows the same logic [16]. Ancient substructures and later industrial insertions are authored together in one HBIM and assigned to phases that reflect their historical placement. Phase switching shows how sanctuary volumes and industrial installations relate in different periods while the shared coordinate frame is preserved. The construction method remains constant: link the consolidated cloud, define phases and apply filters, author families that express change over time, and attach the records that support each element. Users can examine relationships across time confidently because all states rest on the same survey baseline.

Across both sites [16], work stays within the BIM environment from start to finish. Time slices are filtered views of a single HBIM, so attributes and relationships can be examined for any period without changing reference frames. Extant and reconstructed content are visually distinct yet metrically aligned. Documentation is available in-model at the point of need. This combination allows phase-aware exploration that is understandable to non-specialists and traceable for experts, while remaining faithful to the measured record.

4.4. Incorporating Time into HBIM: Calibration and Cases

4.4.1. Calibration Methods of HBIM with Time

Calibration in a time-enabled HBIM is organised around phases within a single model anchored to an immutable survey scaffold [105]. The scaffold is a consolidated, cleaned point cloud or mesh from laser scanning and photogrammetry, linked into Revit rather than imported into it so that coordinates, orientation, and scale remain fixed across all views [78,106]. Geometry is authored once and revealed by phase; nothing is duplicated or re-registered when time slices change [107]. This stable frame keeps calibration consistent across phases and supports direct comparison over time.

Families are heritage-specific and parametric. Where sources allow, two configurations occupy the same space: the current, cloud-observed state and a last-operational state reconstructed from records. Instances carry phase attributes and filters drive visibility. Extant, cloud-supported elements are eligible for quantitative checks, while reconstructed content stays visually distinct via overrides and remains traceable through in-model links to the evidence that justifies its form and placement [105,108].

Distance analysis runs inside Revit using Autodesk Point Layout (or an equivalent tool). Cloud-to-model residuals are computed only for extant elements visible in the active phase [109]. Acceptance typically targets a mean deviation between −0.05 m and +0.05 m, with a standard deviation below 0.10 m, testing geometric fidelity to the linked survey. Reconstructed items are documented but not forced through numeric thresholds [110].

Table 8. Calibration targets and evidence rules for time-enabled HBIM.

Aspect	What to Check	Tool or Source	Acceptance Rule
Scaffold	No re-registration across phases	Linked point cloud/mesh	Single immutable link; coordinates fixed
Extant elements	Cloud-to-model residuals	Autodesk Point Layout (or equivalent)	Mean −0.05 to +0.05 m; SD < 0.10 m
Reconstructed elements	Evidence linkage and visual override	Archives/drawings/photos; in-model links	Not numerically checked; source-traceable
Diagnostics	Clustered residuals and rationale logging	Residual heatmap; calibration log	Refine locally or record justified exception

summarises calibration targets and evidence rules.

Residual fields are read diagnostically. Clusters of medium or high error flag areas for local refinement. If reasonable adjustments cannot remove a discrepancy, the issue is recorded in a calibration log so later users can see what was checked, what changed, and why the remaining error was accepted. Because every phase references the same linked scaffold, deviations are directly comparable across time without new registrations, and observed changes remain interpretable within a single reference frame [111,112].

Acquisition limits do not alter the logic once the scaffold is consolidated. Where interiors are unsafe or inaccessible, curated UAV image blocks with forward and side overlap can provide additional photogrammetric coverage that feeds the same survey cloud. From that point, the workflow does not branch by space type; authoring remains scaffold-based, phases govern visibility, distance checks run on extant, visible elements, and acceptance thresholds stay consistent [113].

Calibration therefore operates at two coupled levels. At the model level, the scaffold fixes the coordinate system and prevents drift while time is read. At the element level, cloud-to-model distances validate extant objects, while reconstructions stay separate, styled, and source-linked. Housing both levels inside one HBIM ensures that calibration travels with the time reading and that error analysis is always available wherever the cloud observes the fabric shown in a given phase [110,112].

4.4.2. Calibration Cases of HBIM with Time

The former Segrè Papermill (Tivoli) is an example [16]. A consolidated cloud is linked into Revit as the fixed scaffold. Historically attested production stages are set as phases. Families are authored so that machines and fabric exist as informative objects with two configurations: extant remnants and last-operational equipment rebuilt from records. Instances are phased, and appearance overrides keep reconstructions distinct. Cloud-to-model distances are computed on extant architectural elements visible in the active phase. Acceptance uses the thresholds outlined above: mean −0.05 to +0.05 m and SD < 0.10 m. Residual clusters guide local refinement. This analysis is illustrated in Figure 9. Because the scaffold is linked once and does not change, the same reference and metrics apply at every stage.

The Sanctuary of Hercules (Tivoli) is another example [16]. The same method serves a site that combines ancient substructures with later industrial volumes. A curated cloud is linked as the single scaffold. Substructures and industrial inserts are authored as heritage-specific families and attributed to phases that reflect their historical placement. With phase filters active, stages are displayed inside the same HBIM without changing coordinates. Cloud-to-model distances are computed on extant components visible in the current phase, and the same acceptance thresholds apply. Hotspots prompt focused edits or, where justified, are recorded. Reconstructions remain visually distinct and source-linked, so their status is always legible.

5. Other Interesting Fourth Dimensions for Three-Dimensional Heritages

Depth sensing, temporal lensing, and time traceback represent a set of emerging 4D extensions. As most published applications are still exploratory, the sections below focus on their use cases rather than a quantitative comparison of technological maturity or adoption.

5.1. Depth

Adding depth information lets us inspect beneath the surface non-destructively [114]. Mercuri et al. [68] couple mid-wave infrared (MWIR) reflectography with pulsed thermography (PT). MWIR reflectograms, captured as an image set, support structure-from-motion to recover surface geometry. PT records transient temperature fields caused by subsurface heterogeneities (voids, repairs, layered media). After pixel-level co-registration using calibration targets, both data streams are fused on a single mesh so users can virtually “peel” the object and read layers through time. The same model runs in VR/MR for slicing and zooming. The method reorganises measured signals rather than inventing what lies inside; qualitative depth cues follow from thermal diffusion contrasts, while quantitative depth still requires assumptions on material properties and acquisition settings. The result is a depth-aware 3D model for study, communication, and conservation.

5.2. Temporal Lensing

Temporal lensing compares two time states within one registered 3D space. Fanini et al. [27] prepare a base scene Gb (current condition) and a target scene Gt (another date or hypothesis). A volumetric lens reveals Gt only inside a user-controlled region, while the rest remains Gb. Implemented with WebGL and exposed via WebXR, the interaction is simple: move the lens, adjust its radius/softness, and inspect where, how much, and in relation to what changes occurred—without swapping scenes or losing context. The technique scales from hand-held artefacts to architectural complexes and sites, supports guided tours and classrooms, and runs in standard browsers and headsets. For specialists, localised inspection aids documentation and discussion; for wider audiences, the in-place, side-by-side view makes complex histories legible.

5.3. Time Traceback

Time traceback in HBIM encodes a building’s evolution as explicit phases that can be played forward or backward. Mazzei et al. [96] define phases in the BIM environment and tag each element with when it appeared, was modified, sealed, or removed. Stepping backward progressively “disassembles” the model in time, exposing documentation gaps and focusing inquiry on unresolved intervals. Geometry, attributes, sources, and notes stay linked so hypotheses remain transparent and revisable. In practice, phase-based modelling captured, for example, the shrinkage and closure of arched openings and the staged evolution of structural parts in the De Simone factory. Parametric families suit repetitive, simple types (e.g., arches), whereas irregular or unique components are modelled non-parametrically to avoid brittle constraints. The result improves communication with curators and planners and guides interventions as new evidence emerges.

6. Limitations and Prospects of 4D Technologies in Heritage

Four-dimensional technologies let us see places not only as they are, but as they were. They join today’s surveys with dated evidence, so that change can be read in space. The payoffs are clear for documentation, analysis, and outreach. Still, three limits recur: thin or unreliable records for past states, calibration that gets harder as time slices multiply, and tools that remain demanding for general users.

6.1. Scarcity and Inaccuracy of Historical Data for Temporal Reconstruction

Most sites do not have a full, consistent record for every period. Photos may be few, oblique, or cropped. Plans and maps can be distorted or drawn to unknown scales. Text sources often resist safe conversion to geometry. Present-day scans and image-based models provide a solid baseline, but adding time reopens gaps. The safe rule is simple: separate what was measured on site from what was reconstructed from documents, and mark that difference inside the model. Four actions help. Firstly, keep source-to-geometry links at the element level so anyone can check what supports a given wall, span, or machine. Secondly, adapt historic drawings to the survey using stable spatial invariants, such as wall thicknesses, spans, distances, and angles, instead of stretching the whole sheet. Thirdly, limit reconstructions to areas that the record truly supports, and mark unsourced zones as schematic or unknown. Fourthly, grow the record through routine collaboration between historians, archaeologists, surveyors, and architects; small finds tied back to the baseline can close stubborn gaps.

6.2. Calibration and Error Propagation in Temporal Models

Aligning many datasets to one coordinate frame is demanding; performing this across time raises the stakes. Each period must obey the present-state scaffold or small local errors will spread. A reliable pattern helps control drift: fuse laser and image-based clouds into one frame, author geometry on that scaffold, then run cloud-to-model distance checks per phase on extant elements. Where projects publish acceptance bands—for example, mean deviation within ±0.05 m with SD < 0.10 m—results are easier to trust and compare. Problems grow when old drawings are forced to fit, camera poses are weak, or edits made for one period quietly break another. Two habits keep models honest: keeping one linked point-cloud scaffold under all views so the frame never moves, and logging residuals instead of hiding them. Useful aids include simple distance-field dashboards inside the modelling software, viewpoint-based photo registration to test silhouettes, and short calibration notes that list the invariants used and the residuals found. The goal is a traceable calibration story, not a single headline number.

6.3. Accessibility and Usability for Non-Specialists

Rich models can still feel opaque. If time is buried in menus, or reconstructed parts look the same as measured fabric, people cannot tell what they are seeing or why it matters. Interfaces should do more of the explaining [115]. Three changes pay off fast. Firstly, the addition of a clear timeline with named phases and a one-click way to switch periods without moving the camera enables what changed, when, and where to stay in view. Secondly, reconstructed elements can be given a distinct appearance and on-object source links can be attached so that evidence is one click away. Thirdly, the visit can be guided with hotspot navigation and short captions; the performance can be kept high with multi-resolution packaging—dense where evidence demands close reading, lighter in context [116]. Immersive options can add presence, but a browser viewer with good defaults reaches more people.

6.4. Interoperability and Long-Term Preservation

A further challenge concerns the limited adoption of shared data standards and interoperability frameworks across 4D heritage workflows. While Industry Foundation Classes (IFC) and the CIDOC Conceptual Reference Model (CIDOC CRM) offer complementary routes for geometry-centric and provenance-centric exchange, neither is consistently implemented in the reviewed studies, and mappings between them remain largely manual. This gap affects long-term preservation and multidisciplinary reuse: diachronic models may be technically complete yet difficult to operate outside their native platforms. Strengthening alignment with open, standardised schemas, rather than tool-specific formats, represents a practical direction for future development.

7. Five-Dimensional Technologies: A Mini Review and Future Directions

7.1. Mini Review

A 5D model can be framed as one formed of extensions built on reality-based 3D and 4D workflows rather than a new geometry. Doulamis et al. [81] introduce 5D modelling as a “3D geometry model + time + levels of detail (similar to depth information, see Section 5.1)” on a single georeferenced baseline, aimed at producing spatiotemporal, predictive 3D maps for conservation. The emphasis is placed on selective modelling and CityGML-style structures so that very large cultural landscapes remain navigable while still allowing local high-fidelity inspection. The practical promise is that once the baseline is stable, dated states and risk indicators can be added and compared without re-registering or duplicating models, keeping site-wide readings coherent.

Li [21] presents 5D GIS as a map-centred environment where the third dimension augments 2D GIS, the fourth dimension records time, and the fifth collects intangible layers such as oral histories and community memory. The intent is to let users query “what was here at a given time” and “whose stories attach to this place” inside one interface that stays locked to measured geography. The contribution is not a new capture method but a way to organise spatial, temporal, and narrative data together so that change in use and meaning can be read in place.

Chiabrando et al. [105] also retrace practice “from point clouds to user-oriented HBIM”, treating survey-derived point clouds as the unavoidable knowledge base for authoring object-level historic building models. Here, 5D appears in the widely used management sense of “3D plus time and cost”. The paper underscores interoperability, traceability from geometry back to consolidated surveys, and a clear visual and semantic separation between extant fabric and reconstructed parts, so non-experts can read the model while professionals can manage the works and condition.

Rizvi’c et al. [20] further show “time travel” experiences built from present-state scans combined with carefully sourced reconstructions, delivered in VR and AR (see Figure 10). Period scenes switch inside one spatial frame; plans, photos, and texts are pinned to locations; and the user compares “what changed, when, and where” without losing bearings. The additional “dimension” here is the curated storyline and the explicit link from each reconstructed element to its documentary support, still riding on survey-true scenes.

7.2. Future Directions

Consensus holds that 3D denotes geometry and 4D adds explicit, dated time states on the same spatial baseline. By contrast, the meaning of 5D remains unsettled and competing usages coexist. In current practice, the “fifth dimension” is variously equated with cost tracking, with levels of detail (similar to depth information, see Section 5.1), or with AR/VR delivery. A dialectical reading clarifies the boundary conditions. Cost belongs to HBIM’s attribute space and can be joined by many other management details, including maintenance or retrofit records; if any such attribute were promoted to a new dimension, the term would lose discrimination. AR/VR similarly functions as a communication interface layered over 3D/4D content; as shown in Section 4.1.3, App Calw VR changes how users engage but not the ontological axes of the model. Consequently, a possible definition treats 5D as a detail layer grounded in reality-based surveys: depth, stratigraphy, and diagnostic signals that extend surfaces into measurable layers while staying georeferenced, queryable, and provenance-linked. Such a definition separates 5D from project metadata and presentation technology, supports interoperability with open schemas, and remains legible to non-specialists through clear visual encodings of certainty and source. Notably, the final determination of the fifth dimension still needs further discussion after extensive research.

Implications for calibration follow directly from 4D practice in Section 4.2 and Section 4.4. A single, immutable survey scaffold must anchor all phases and fifth-layer views, avoiding any re-registration between time slices. Quantitative checks on extant fabric should report cloud-to-model residuals within declared bands (for example, mean −0.05 to +0.05 m and SD < 0.10 m), with document adaptation constrained by spatial invariants and archival photographs registered as viewpoints. If 5D encodes depth or diagnostics, additional metrics are required, including pixel- or voxel-to-mesh co-registration error, signal-to-noise, inversion assumptions, and layer-thickness sensitivity. Each release should include a concise calibration digest covering baseline accuracy, per-phase residuals, fifth-layer metrics, excluded zones, and evidence links.

8. Limitations of This Work

8.1. Scope and Methodological Limitations

This review is qualitative and workflow-focused rather than bibliometric or quantitative. The two-stage retrieval provides transparency in inclusion and exclusion principles, but it does not claim to exhaustively measure publication volume, citation structure, or geographic output. Because the aim is to synthesise methodologies rather than to evaluate global production patterns, no formal assessment of dataset objectivity or completeness was conducted beyond cross-checking the search string in Web of Science and Scopus. Future work combining 4D heritage modelling with bibliometric or scientometric analysis could complement this synthesis by providing publication-level statistics.

8.2. Regional and Disciplinary Representation

Heritage digitisation practices vary significantly across regions, disciplines, and funding infrastructures; the reviewed corpus inevitably reflects where 4D/5D modelling is documented in peer-reviewed journals indexed by Web of Science and Scopus. Contributions from archaeology, conservation science, architecture, and computer graphics are well represented, whereas projects reported mainly through grey literature, national repositories, or non-indexed venues may be under-represented. Expanding future reviews to include multilingual databases and institutional archives may help counterbalance this publication-driven bias.

9. Conclusions

Drawing on 56 articles about the 3D geometry of heritages (2000–2025), this review consolidates how 4D technologies add dated states to survey-anchored 3D and how HBIM encodes time within one coordinate frame. The synthesis spans construction, incorporation, calibration, and exhibition, translating recurring limits into actionable workflows and highlighting calibration as a necessary condition for trustworthy diachronic reconstruction. The main findings are drawn as follows:

(a)

For ordinary 3D models, a pipeline is identified, which couples TLS, A/MLS, and photogrammetry with reverse modelling where needed, then incorporates time as discrete, source-labelled states on the shared baseline. Exhibition uses switchable scenes, matched-viewpoint photos, and clear separation of measured and reconstructed parts. Cases at Ávila, Calw, the Alphonse Raymond Factory, San Giovanni in Conca, and Orígens Geopark demonstrate scalable, evidence-bounded diachronic reading.

(b)

For HBIM, phasing and filters inside a single, point-cloud-linked model are established, with heritage-specific families encoding extant and reconstructed states while coordinates remain fixed. Evidence is linked at the element level, and appearance overrides keep hypotheses legible. Calibration runs as cloud-to-model checks only on extant, phase-visible elements to ensure that phase comparison is metrically reliable. Coordinated cases at the former Segrè Papermill and the Sanctuary of Hercules show reproducible phase switching without geometry duplication.

(c)

Other delivery and dimensional extensions of 4D technologies are summarised: depth via MWIR reflectography fused with pulsed thermography on a single mesh, temporal lensing for in-place local comparison, and HBIM time-traceback for reversible phase playback. It should be noted that there is controversy over the definition of the fifth dimension in 5D technology—levels of detail, intangible documentation, cost, or AR/VR tool—yet the reviewed evidence suggests that these 5D directions develop on top of rather than replace 4D practice.