SDT4Solar: A Spatial Digital Twin Framework for Scalable Rooftop PV Planning in Urban Environments

Highlights

• Integrating higher spatiotemporal resolution in solar energy modelling reveals significant differences in annual rooftop solar energy assessment results.
• The Spatial Digital Twins (SDT) framework serves as a basis to streamline dynamic modelling and systemic energy potential analysis for future solar energy planning.

What are the main findings?

• A proof-of-concept SDT framework (SDT4Solar) designed for solar energy assessments has been developed to present an integrated approach that enables dynamic solar potential analysis.
• Results from the prototype deployment of SDT4Solar demonstrate that annual rooftop PV energy generation estimates are less than half of those reported in previous studies that used lower spatiotemporal resolution datasets.

What is the implication of the main finding?

• The SDT4Solar prototype demonstrates the feasibility of using dynamic solar components in energy planning and the potential for integrating datasets, enabling more accurate estimations of solar potential to support more effective rooftop energy planning and future deployment.
• The results from this study highlight the importance of considering the spatiotemporal component in solar modelling and the utility of unified databases to improve the accuracy of potential energy assessments in solar energy planning.

Abstract

To sustainably power future urban communities, cities require advanced solar energy planning tools that overcome the limitations of traditional approaches, such as data fragmentation and siloed decision-making. SDTs present a transformative opportunity by enabling precision urban modelling, integrated simulations, and iterative decision support. However, their application in solar energy planning remains underexplored. This study introduces SDT4Solar, a novel SDT-based framework designed to integrate city-scale rooftop solar planning through 3D building semantisation, solar modelling, and a unified geospatial database. By leveraging advanced spatial modelling and Internet of Things (IoT) technologies, SDT4Solar facilitates high-resolution 3D solar potential simulations, improving the accuracy and equity of solar infrastructure deployment. We demonstrate the framework through a proof-of-concept implementation in Ballarat East, Victoria, Australia, structured in four key stages: (a) spatial representation of the urban built environment, (b) integration of multi-source datasets into a unified geospatial database, (c) rooftop solar potential modelling using 3D simulation tools, and (d) dynamic visualization and analysis in a testbed environment. Results highlight SDT4Solar’s effectiveness in enabling data-driven, spatially explicit decision-making for rooftop PV deployment. This work advances the role of SDTs in urban energy transitions, demonstrating their potential to optimise efficiency in solar infrastructure planning.

1. Introduction

Rooftop solar planning has advanced significantly in recent years [1], driven by the commitment to climate change mitigation and urban energy resilience. There have been major efforts to promote solar energy in urban planning [2]; however, the delayed integration into the process has often led to unsatisfactory results [3]. More recently, there has been a growing recognition of the need to conduct optimal planning of rooftop PVs due to the issues of cost-effectiveness and grid-friendliness [4]. Contemporary solar planning approaches are increasingly leveraging geospatial tools and techniques to enhance the reliability and efficiency of the PV energy potential estimations. Readily available geospatial tools such as ESRI’s Solar Analyst Tools and GRASS r.sun enable a convenient computation of solar irradiation in a streamlined software. However, these tools often rely on 2D or 2.5D geospatial data input, which simplifies the representation of urban features, especially the geometries and orientations of rooftops. Furthermore, these tools often omit key important atmospheric parameters that influence solar irradiation and energy potential estimates [5]. Web-based platforms such as PVGIS, RETScreen, SAM, and Helioscope provide interactive interfaces for rapid estimations and feasibility analysis of PV performance at the building scale. Although these platforms incorporate high-temporal solar irradiation data, their limited capability to integrate complex spatial context in urban environments makes them insufficient for large-scale rooftop solar planning. In this regard, it is not only the temporal reliability of solar radiation values data that should be considered, but also the spatial granularity and contextual accuracy needed to minimise estimation errors.

Studies have developed approaches to address these overlooked factors. These include algorithms and techniques such as the Roof-Solar-Max algorithm that considered detailed rooftop shape, orientation, and size to optimize PV panel placements at a regional level [6]; SNU Solar tool that calculates irradiance and electricity generation using local weather data, spatial components, and PV system parameters to provide temporal results of solar electricity estimates [7]; and Ener3DMap-SolarWeb roofs that automatically computes the photovoltaic potential of rooftops using integrated dataset and solar radiation models into one web-based tool [8]. However, despite the advancements in computational algorithms and the incorporation of detailed physical parameters, the integration of datasets for solar potential assessments under dynamic real-world conditions remains a significant challenge. To inform future improvements in rooftop solar modelling and energy potential estimation, it is essential to identify and address persistent limitations in current solar assessment methods and evaluate their application in real-world contexts.

1.1. Persistent Challenges in Solar Rooftop Spatial Planning

1.1.1. Limitations of Conventional Urban Modelling Approaches

Conventional urban modelling techniques often simplify urban geometries, failing to capture the complexities of the built environment [6]. Topographic factors, particularly slope, elevation, orientation, and size, play a significant role in influencing surface solar radiation [9,10,11] and thus must be accounted for effectively in energy planning.

Numerous studies have demonstrated how these simplifications introduce errors. For instance, Al-Quraan et al. [12] and Schallenberg-Rodríguez [13] exemplified percentage errors of 2.7% to about 18% in energy estimation when the ideal tilt on PV systems has not been applied or panels have been mounted horizontally. Similarly, Pitt and Michaud [14] noted a 15% discrepancy between GIS-based solar modelling outputs and average solar insolation values from NREL, attributed to unrecognised angles of sloped roofs. These assumptions, often necessary due to data limitations, often misrepresent shading conditions, roof form, and irradiance. A list of studies that illustrate these limitations is presented in

Table 1. Examples of studies highlighting limitations in urban modelling techniques for rooftop solar assessments.

Reference	Assumption and Rationale	Impact
[9,15]	Shading is not considered.	Overestimates surface solar irradiation values.
	Relied on broad classifications with roof slopes.	Misleading solar potential, especially on sub-optimal roofs.
[16]	Use of both manually measured and default building heights due to limited data access.	Misrepresentation of suitable roofs.
	Excluded slope and shading on the assumption that airport roofs are flat with minimal vegetation.	Fail to capture granular solar potential variability.
[17]	Only slopes less than 45° were considered as steeper slopes were assumed to receive less irradiation.	Misleading solar potential, especially on sub-optimal roofs.
	Excluded shading due to the remote characteristic of the study area, with low-rise and shade-free.	Fail to capture granular solar potential variability.
[18]	Omitted both slope and shading with the assumption that uniformity in building design and height would eliminate shading effects.	Fail to capture granular solar potential variability.
[19]	Adopted a uniform building height value of 30ft to align with the local building codes	Fail to capture granular solar potential variability.
	Shadowing was excluded due to software limitations.	Overestimates surface solar irradiation values.
	Air temperature substitutes for the effect of shading.	Inaccurate solar potential estimation, as air temperature is unrelated to solar access.
[20,21,22,23]	Used shading coefficient	Either overestimates or underestimates surface solar irradiation values.

.

1.1.2. Fragmentation of Solar Energy Assessment-Related Datasets

Beyond the limited precision in urban modelling, rooftop solar spatial planning is further hindered by data fragmentation, which contributes to the simplification of urban features, reduction in scope, and inaccuracies in solar assessments [24]. Improving data availability and data standardisation are critical steps towards overcoming data fragmentation, thereby enhancing the applicability and reliability of urban energy systems [25].

In the study of Chow et al. [26], it has been acknowledged that the urban environment consists of complex objects that present significant challenges to solar assessment methods. Given the multidimensional nature of solar energy potential assessments [21], past research used a variety of methodologies, varying in sophistication and requiring different levels of expertise [11,27,28].

In the context of determining geographical and physical potential, datasets often exist in different formats and are managed in various systems, posing a challenge in maintaining data integration, interoperability, and consistency in spatial assessments. The study by Sakti et al. [29] identified several datasets commonly used to model optimal areas for PV deployment and accurately assess solar potential. It also emphasises the interconnectivity of these datasets and illustrates the application within the approach. This not only highlights the diversity of required datasets but also underscores the need for better strategies to manage, integrate, and utilise datasets to support a robust solar energy assessment process.

1.1.3. Inadequate Integration of Dynamic Solar Modelling Techniques

The trend of spatial solar modelling approaches has undergone transformation through interdisciplinary innovations [1,27,30,31,32]. Despite these advancements, methods still lack continuous granular solar modelling capabilities essential for city-level rooftop assessment. For instance, Cuesta-Fernández et al. [23], Cieślak and Eźlakowski [33], Baghani [34], Jo et al. [35], Kouhestani et al. [36], and Mishra et al. [37] has used geospatial data to estimate irradiance using GIS tools; yet this method often produces static maps that require effort to fully transform the results into interactive visualisations. Furthermore, GIS tools are generally derived from prediction models that might be limited in representing real-world solar conditions. Studies such as Dehwah et al. [38], Massano et al. [39], McIntyre [40], and Mikovits et al. [41] have attempted to model solar irradiance using forecasted meteorological data to achieve high temporal resolution; however, these modelling approaches often do not offer the necessary spatial granularity. Research by Adjiski et al. [42] and Margolis et al. [43] has addressed this challenge by adopting a hybrid approach that combines LiDAR-based datasets with TMY data to enhance the solar assessment methodology. Similarly, An et al. [44] and Boulahia at al. [45] used a 3D environmental modelling software that utilises synthetic climatic data to identify solar potential. Chow et al. [26] adopted a stimulation tool along with LiDAR datasets to prove that solar systems can alleviate carbon emissions. Thebault et al. [46] used the concept of 3D GIS irradiation modelling to account for roof schematics, local meteorological data, and shading.

Although the hybrid solar modelling approach is a promising solution to enable dynamic solar modelling, it faces challenges related to computational complexity due to varying data resolution [27,30,32]. Identifying an approach that facilitates dynamic modelling while maintaining spatial compatibility within datasets would significantly increase the value of solar energy planning.

1.1.4. Insufficient Support for Collaborative Decision-Making

Contemporary approaches have often presented planning insights from a static map, where solar energy assessment results have been limitedly showcased and explored. While this is often a good way to deliver insights, it does not allow further support in decision-making, limiting its value in creating real-world impact. For example, Quirós et al. [47] aimed to present the solar potential of the building to provide options for using renewable energies in the study area that could be helpful in climate change mitigation. Similarly, Huang et al. [48] presented the distribution and yearly average of solar radiation for suitable roof planes through a map. In the study by Yang et al. [49], potential areas for rooftop PV deployments and the corresponding installed capacity were presented, simulating three scenarios using maps. Ayodele et al. [50] mapped the total roof area, maximum installable capacity, and annual energy output of each local government area in Ibadan to ease the optimal planning and development of solar rooftop PV. Notably, the study by Alvarado et al. [51] made a significant shift by demonstrating strong stakeholder interest in interactive solar planning through a platform, emphasising the need for an inclusive platform and the value of user engagement. This transition from static maps to interactive platforms represents a crucial advancement in solar energy planning, enabling inclusive and collaborative decision-making.

While several studies have provided a more accurate approach to solar modelling and assessments [52,53,54], few have addressed the issue of data fragmentation within the modelling and assessment process. This study tries to address this gap by presenting a new approach that directly tackles this challenge, improving the applicability, reliability, and practicality of solar assessments in supporting stakeholder decision-making.

1.2. Spatial Digital Twins (SDT) as a New Paradigm for Spatial Solar Modelling and Planning

One possibility for resolving gaps in data fragmentation and for demonstrating the potential of incorporating realistic and dynamic data, advanced spatial solar modelling, and coupling of GIS technologies within a single system is through an SDT.

SDT is a special type of digital twin that represents a significant advancement over traditional digital twins. It embeds spatial characteristics, particularly geographic location, scale, and elevation, that enhance the contextualization of real-world entities and processes [55]. While both SDTs and conventional DTs aim to model, monitor, analyse, and simulate, SDTs offer deeper capabilities in analysing spatial relationships within real-world environments. It enables the sharing and combining of data within a geospatial network [56], adding a critical spatial dimension to the understanding of urban systems.

SDTs has proven its capabilities in several fields of study [55] and have displayed its capabilities by (1) simulating at various levels and scale by user-controlled scenario input, reducing manual workflows, (2) providing a 3D realistic environment, allowing for early-stage insight to reduce uncertainties and minimise risks, (3) predicting futuristic insights, presenting timely response and cost-efficient decisions, and (4) advancing querying capabilities supporting responsive and data-driven planning. SDTs also harness innovative technologies, as presented in Figure 1, that support urban environment modelling, facilitate the integration and analysis of diverse data, and dynamically deliver actionable insights.

This capability of SDTs is valuable in urban planning and renewable energy assessment, where spatial heterogeneity and temporal fluctuations play a crucial role in the assessment and planning process. Beyond the conventional thematic layering of data, SDTs combine the aspect of time to identify real-time phenomena and deliver spatiotemporal insights accordingly, such as the responsive flood digital twin by Turner and Sun [57].

In solar potential modelling, these tools and capabilities play a transformative role in providing fit-for-purpose feedback. Numerous studies [1,58,59,60,61,62] have articulated this potential; however, few advance beyond conceptual models into practical SDT applications. It has been noted by Teofilo and Sun [1] that there had been no apparent direct application of solar energy planning utilising digital twin technologies. Furthermore, existing implementations remain limited in scope. For instance, Bocullo et al. [63] applied digital twins to district-level renewable energy retrofits, demonstrating financial and self-sufficiency benefits, yet models lacked granular solar irradiation analytics. Similarly, Shirinyan and Petrova-Antonova [54] employed a 3D CAD-based model to simulate solar radiation for small-scale deployments, showcasing the relevance of digital twin concepts for solar analysis, yet lacking broader scalability and integration capabilities. These limitations present a timely opportunity to develop a foundational SDT tailored for solar energy applications to support sustainable energy management.

To address this gap, we propose a conceptual framework called Spatial Digital Twin for Solar Energy Planning (SDT4Solar), as illustrated in Figure 2, that draws from the shared values noted in [64,65,66], i.e., interoperability, security, public good, scalability, trustworthiness, data ethics, and real-world synchronisation.

This paper aims to establish the foundational approach for SDT4Solar, a Spatial Digital Twin-based framework for integrated rooftop photovoltaic (PV) potential modelling and planning. It presents a proof-of-concept to demonstrate how SDT4Solar can support adaptive, data-driven energy planning and decision-making in complex urban environments. The key contributions of this paper are (i) the development of a ‘living system’ that enables continuous update leveraging a cyber-physical paradigm to enhance broader urban energy system; (ii) the integration of multi-modal datasets with digital twin technologies to support data-informed solar energy planning; (iii) the presentation of a dynamic modelling approach that utilises real-time solar observation sensor data to manage for roof solar energy generation at an urban scale.

To further evaluate the potential value of the SDT Framework towards urban sustainability and examine its practical implementation, we conducted a case study to assess the feasibility of the framework. In addition, this study included a comparison between rooftop solar PV energy production generated using the SDT4Solar approach and that reported in a previous study by Kalyan and Sun [17] that employed a conventional spatial modelling method. Our objectives are the following:

• Develop the SDT4Solar technical framework by defining key parameters, methods, and workflows for integrating geospatial databases, 3D city models, and solar modelling techniques within a Spatial Digital Twin environment.
• Establish an integrated urban energy database that enables seamless data flow, supports semantic interoperability, and manages multi-source datasets including 3D building geometries, land use, and meteorological data.
• Model rooftop suitability, solar irradiance, and PV energy potential dynamically across urban scales to support realistic and scalable rooftop PV deployment scenarios.
• Design and implement a web-based interactive SDT4Solar prototype, providing a user-friendly interface and decision-support functionalities for urban planners, policymakers, and stakeholders engaged in solar energy planning.

2. Materials and Methods

2.1. Methodological Framework

The SDT4Solar concept introduces a foundational methodological framework (as seen in Figure 3) that models and integrates static urban datasets with live solar radiation measurements to support dynamic urban rooftop PV planning.

The core of this framework involves (1) digital characterisation of the urban environment through digital surface model (DSM), digital building and vegetation model (DBM/DVM); (2) integration of datasets into a database; (3) rooftop solar potential modelling involving a live solar radiation sensor; and (4) a testbed for insight visualisation and exploration.

2.2. Urban Spatial Representation

The creation of urban environment layers, specifically the DSM, DTM, DBM/DVM, and DEM, was through the ‘LAS Point Statistics as Raster’ tool. The DSM raster layer was generated as a base layer to provide a realistic representation of the urban surface in the development of the SDT4Solar prototype. The DBM/DVM, along with the DEM, were utilised to represent the built environment in three dimensions, supporting the analysis of shadowing and solar exposure. The DTM, on the other hand, served as a positional reference for the terrain and building elevations during the generation of 3D building models.

Suitable roof areas were identified from the generated building raster layer using the ‘LAS Point Statistics as Raster’ tool. This raster was then converted into polygon features using the ‘Raster to Polygon’ tool to delimit rooftop boundaries. To ensure geometric continuity, the ‘Fill Gaps’ and ‘Dissolve’ tools were applied to remove unenclosed features and overlapping polygons. The ‘Regularise Building Footprint’ tool was applied to refine the building features and define individual suitable roof areas. Finally, roof points located within the refined building features were extracted from the original point cloud layer for use in modelling solar radiation.

The construction of the 3D building models was done using the ‘LAS Dataset to Raster’ tool. The DTM raster layer was first generated, followed by the creation of the DBM/DVM raster layers, to accurately capture building structures along with vegetation above the terrain. The ‘LAS Building Multipatch’ tool was used to generate building models. Given the quality of the input data, models can be built up to a level of detail (LoD) 2 [67]. Point clouds have been used to generate LoD2 Models due to their capacity to accurately capture roof geometries, such as roof orientation, slope, and morphology, which is essential for irradiance modelling. Although higher LoDs offer finer architectural detail, the increased data density and computational demand may not yield proportional benefits for large-scale urban planning applications. The progression of creating LoD2 models is illustrated in Figure 4.

Rooftop orientation is a key factor in determining optimal locations for rooftop PVs, which is defined by both aspect and slope. Aspect refers to the direction in which the roof faces, while slope refers to the degree of roof steepness. The ‘Aspect’ and ‘Slope’ analysis tools were applied to the DBM to determine the slope and aspect of each roof point. The result of the analysis was incorporated into the roof points.

2.3. Integration of Datasets into a Central Database

The datasets were restructured according to the CityGML schema [67] to ensure consistency for seamless integration into the 3DCityDB environment. The ‘Building’ and ‘BuildingPart’ specifications, which are subclasses of ‘_AbstractBuilding’, have been adopted to standardise the representation of spatial data. This study utilised the LoD2 data structure, as presented by Gröger et al. [67], and the Feature Manipulation Engine software was employed to convert the models into CityGML format, aligning the attributes, particularly the roof type, height, and geometry. This methodology is consistent with the approach taken by Diakité et al. [56]. The PostGIS-enabled PostgreSQL database was extended with 3DCityDB to enhance the management of 3D data. It served as the backend environment where 3D building models and other solar-related datasets are stored, managed, and retrieved.

2.4. 3D Solar Rooftop Potential Modelling

Central to this study is the demonstration of a fundamental approach to dynamically model solar radiation, estimate potential electricity generation, and assess potential carbon emission offset on a granular rooftop level. This approach involved solar observation data from a live IoT sensor with a calibration formula that adjusts the solar radiation values of each roof point based on both slope and shadowing effects, resulting in a more realistic estimate of solar irradiation.

2.4.1. Calculation of Slope Effect

The slope reduction factor (SlopeRed_Loss) refers to the effect of roof inclination on the amount of solar radiation received by the roof surface. Since the angle of the roof affects how the sunlight is incident on the PV surface [68], this factor is used to adjust the estimated solar radiation accordingly. To determine this factor, PV mounting simulations were conducted using the PVWatts^® Calculator tool [69]. Following the methodology of Jacobson and Jadhav [70] and Teofilo et al. [16], the optimal tilt angle for the study area was identified by testing the PV panel output at various inclinations and selecting the tilt angle that resulted in the highest energy generation. The angle that yields the highest PV generation is identified as the optimal slope. For roof points with slopes that deviated from the identified optimal tilt angle, the associated percentage loss in solar radiation was calculated and used to adjust solar radiation at each point accordingly.

2.4.2. Determination of Shadowing Loss

In line with the objective of achieving a realistic and dynamic representation of the urban environment, a more granular analysis of rooftop shadowing has been conducted. To achieve this, the traditional planar roof shadowing representation was replaced with a point-based spatial approach, similar to the multi-resolution shadowing method employed by Lukac et al. [71]. Roof surfaces were modelled using a grid of equally spaced points to account for localised shading effects at a finer resolution. This method enables the identification of partially shaded roofs and allows for proportional reductions in estimated solar generation, in contrast to planar-based methods that often disregard solar output entirely in the presence of any shading.

Shadow impacts on the roof were calculated by accounting for obstructions from adjacent buildings and vegetation on all levels. A dynamic shadow analysis was conducted at 15-min intervals throughout the year to capture temporal variations in shading. The analysis began with the calculation of the sun’s azimuth and altitude to determine the sunlight availability. Subsequently, a Hillshade analysis was applied within these solar windows to identify the presence and extent of roof shadowing. The resulting shadowing impact is applied to each roof point, allowing the precise quantification of localised solar energy losses due to spatio-temporal shading over time.

2.4.3. Calculation of Urban-Adjusted Solar Radiation, Annual Rooftop PV Energy, and Carbon Offset Potential

Solar observation sensor, such as that described in [72], often records data at a single location to represent the current condition of an area. While prior studies [40,73] have validated that ground-measured solar radiation is generally reliable within a 30 km radius, applying uniform values across a study area can introduce significant inaccuracies in solar electricity estimates. This often results in misleading outcomes in solar energy planning, particularly where urban form introduces complex, localised variations.

To overcome this limitation, a point-based spatial adjustment method was implemented. This uses rooftop-level parameters and shadowing factors to spatially calibrate the ground-based solar radiation measurements. The adjusted radiation values were applied to each rooftop point across the study area, enabling not only a realistic but also a spatially sensitive estimation of solar rooftop PV generation.

The adjustment process begins by assigning the solar radiation sensor observation data to each roof point, followed by the application of Equation (1). This was done to account for both environmental and technical factors unique to each roof point. In this way, the spatiotemporal variations in roof characteristics and shading were accounted for, eliminating the potential errors that occur when using constant values across spatially diverse building structures.

$${Point}_{SolRad}= {SolRad}_{Sensor}\times \left({\displaystyle \frac{\left({DaylightInt}_{Freq }-{ShadowedInt}_{Freq} \right)}{{DaylightInt}_{Freq }}}\right)\times (1-{SlopeRed}_{Loss})$$

(1)

where Point_SolRad is the adjusted solar radiation in W/m² of roof point, SolRad_Sensor is the average hourly solar radiation value from the sensor in W/m², ShadowedInt_Freq is the total number of hours during daylight that are shadowed, DaylightInt_Freq is the total number of daylight hours, and SlopeRed_Loss accounts for roof slope reductions. Once the adjusted solar radiation has been determined, Equations (2) and (3) have been applied to calculate the potential PV electricity (Point_ElecPot) for each roof point and roof power generation (Roof_Power), respectively. Both are measured in W, as follows:

$${Point}_{ElecPot}= {Point}_{SolRad} \times PR \times r$$

(2)

$${Roof}_{Power}= \left({\displaystyle \frac{\sum {Point}_{ElecPot}}{N}}\right)\times A$$

(3)

where PR is the performance ratio and r is the PV module efficiency. Then, ∑Point_ElecPot is the sum of potential electricity generation at each rooftop, N is the total number of points used in the calculation within each roof, and A is the total suitable area. Finally, the rooftop energy generation (Roof_Energy) was calculated using Equation (4) and is expressed in kWh, as follows:

$${Roof}_{Energy}= {\displaystyle \frac{ {Roof}_{Power} \times {DaylightInt}_{Freq }}{1000}}$$

(4)

The calculation of carbon emissions offset (t CO₂-e) was performed using Equation (5), based on the formula outlined by DCCEEW [74].

$${t CO}_2-e={\displaystyle \frac{ {(Roof}_{Energy}\times \left(EF2+EF3 \right))}{1000}}$$

(5)

where EF2 is the scope 2 emission factor and EF3 is the scope 3 emission factor from the Australian National Greenhouse Accounts Factors 2024 [74].

2.5. Development of SDT4Solar Prototype Web-Based Interface

The prototype front-end interface was developed using CesiumJS (version 1.117.0.) The interface is conceptually designed, containing 3 main components: (a) the control panels, (b) the map display area, and (c) the query capabilities.

The control panels feature buttons for the adjustment of settings and layers based on user needs. Most of these panels utilise prebuilt toggle buttons from CesiumJS, which include tools for basemap or terrain selection, home navigation, scene selection, and search functions. The map display component provides users with access to and interaction with the simulated results or query results based on user inputs. A basemap and terrain were incorporated to enhance a realistic 3D representation. The query capabilities enable the retrieval of information about specific buildings, including the estimated solar radiation, potential PV energy output, and associated carbon offsets. Queries involve fetching data from the IoT sensor, calculating solar radiation, energy potential, and potential carbon emissions, and then allowing users to select individual building models within the 3D environment to view the corresponding attributes.

2.6. Case Study Implementation

The SDT4Solar framework was applied to the Ballarat East locality, as seen in Figure 5a. The locality is a part of the City of Ballarat in Australia, which aims to become the first regional Australian city to be powered by clean energy sources [75]. This makes Ballarat East a strategic case study for implementing the framework to support the stakeholders in achieving their clean energy goals.

In addition, the city has a unique regional urban landscape, characterised predominantly by residential neighbourhoods [76], positioning itself in achieving its clean energy targets with the help of residential solar. This characteristic makes the city an ideal case for examining rooftop solar potential. Moreover, with the projected population growth [77], there is an urgent need for strategic rooftop PV planning and on-monitoring to support sustainable solutions that can meet future energy demands.

To support the implementation, this study leveraged a 15-min interval solar observation sensor from the City of Ballarat [72] (see Figure 5b), which provided high-temporal-resolution radiation to dynamically model solar radiation. Solar radiation values were derived from the API endpoint https://data.ballarat.vic.gov.au/api/explore/v2.1/ (accessed on 31 October 2024). Ground measurement readings have been disclosed by Cros et al. [73] to be valid within a spatial extent of 0.5°, or approximately 60 km. The sensor is within the valid range, therefore supporting its applicability. This sensor formed the dynamic component of the analysis, enabling solar potential estimates to be adjusted based on real urban conditions. The ground-based solar radiation sensor was chosen in this study as a prototype component for atmospheric conditions, as its readings reflect actual solar conditions in the area. For instance, during periods of cloud cover, the pyranometer’s radiation measurement records lower irradiations, capturing real-time fluctuations in solar radiation. This enables the model to incorporate actual atmospheric conditions without requiring coarse-resolution meteorological datasets.

The static components of this study consisted of high-resolution 10 cm point cloud data from ELVIS [78], which was used to model the terrain, building, and surface. This dataset was used to obtain a higher level of accuracy in the determination of semantic content for roof structure and surface variation, essential for assessing rooftop PV potential. High-resolution 30 cm aerial imagery and Ballarat East boundary polygons were also sourced from DataVic [79] to support rooftop structure, shading analysis, and study area delineation. The spatial resolution of the imagery significantly exceeds the 0.6–1 m range mentioned by Li et al. [80], which can provide site-specific mapping to support urban area management. Additionally, the Statistical Area Level 1 (SA1) dataset from the Australian Bureau of Statistics [81] has been incorporated to support the derivation of planning insights within the prototype SDT4Solar.

3. Results

3.1. SDT4Solar Technical Deployment

3.1.1. Modelling of the Urban Environment

The framework allowed the modelling of the urban environment, most importantly, the building features. The modelling approach allowed the creation of hip, flat, and gable roof types. Figure 6 compares the building models generated in ArcGIS Pro for this study with their real-world counterparts, as viewed in Google Earth Street View.

3.1.2. Integrated Database

Due to the presence of an integrated database, not only is importing and exporting of datasets accomplished, but also querying of dataset information. From a 3DCityDB query, a report on its current structure and operational status can be determined, confirming the existence of entities in the database.

3.1.3. Dynamic 3D Solar Potential Modelling

Figure 7 presents the result of the rooftop shadowing analysis, illustrating the spatio-temporal shading effects on sample roofs. The sample illustration focuses on June 21, the shortest day of the year in the Southern Hemisphere, when shading impacts are typically most significant. The figure displays results in different time intervals and the corresponding roof shading effects. Lighter points represent sunlit rooftop points, whereas darker points reflect areas affected by shadowing.

The 30-min interval results display the progression of the solar exposure of rooftops as the sun rises during the daylight period. The hourly interval results illustrate a more apparent effect of shading within the rooftop as the solar position changes throughout the day. The daily shading analysis reveals more subtle yet important shading changes in relation to the Earth’s Orbit. The monthly variation gives a broader seasonal trend in rooftop shading.

Subsequently, the distribution of the estimated solar radiation potential was derived. Figure 8 illustrates the aggregated urban-adjusted solar radiation values throughout the year with the corresponding rooftop area. The colour intensity indicates the level of roof solar irradiation, with darker features representing higher potential at each specific time interval.

With ground-measured solar radiation processed at 15-min intervals, solar energy output estimates can be generated and aggregated across different temporal scales. Figure 9 illustrates the comparison between monthly aggregates and the annual aggregated value. The figure demonstrates that annual estimates do not reliably represent monthly variations and are limited in accurately capturing seasonal dynamics. It reveals that only in March and September the monthly values align closely with the annual average. In contrast, the deviation is most significant during June and December, indicating that relying solely on annual aggregates in these periods could significantly misrepresent actual solar energy potential.

3.1.4. Prototype SDT4Solar Web Application Interface

The SDT4Solar concept led to the development of a testbed featuring an interactive environment that enables users to engage with a prototype decision-making interface for solar energy planning. The interface includes an introductory description window and the SDT4Solar environment. The introductory description window outlined key information such as the purpose of SDT4Solar and its functionality, and the expected outcomes (see Figure 10a). The SDT4Solar environment serves as the core component of the interface, enabling users to interact with spatial data, perform data queries, and simulate solar potential (see Figure 10b).

The SDT4Solar environment has a direct link with a live sensor and backend database. The functionality of the SDT4Solar environment includes a zoom in/out that allows users to explore on different levels. At a building level, the interface allowed a query capability that allows users to explore individual building models with their semantic information through mouse hovering or on-click interactions. At the local scale, users are required to input PV specifications to generate solar potential analysis results and filter building attributes based on energy output and suitability criteria.

3.2. Ballarat East Operational Implementation Insights

3.2.1. Urban Environment Model, Roof Schematics, and Building 3D Model

The implementation results show that a total of 2475 features were identified as suitable rooftops, in accordance with Part J9 of the National Construction Code [82] and the minimum surface area required to power a 1.5 kW solar PV system [83,84]. Rooftop elements such as roof valleys, roof ridges, and access or maintenance space that is typically around 2–3 feet around installations, were also excluded, resulting in a final suitable roof area of 515,443.65 m² (see Figure 11).

Additionally, the building heights across the 3D modelled buildings ranged up to 15.38 m, with an average height of 3.55 m, reflecting the low to mid-rise building character within the area.

Findings from the roof characterisation indicated that west-facing roofs are the most common roof orientation in Ballarat East, with 112, 518.28 sq. m. in total area, followed by east-facing roofs with 107,418.98 sq. m., then south-facing roofs at 114,393.60 sq. m. Although not considered the ideal orientation for PV installations in the southern hemisphere [70,85], south-facing roofs have the largest area in Ballarat East, totalling 102,297.13 m².

The slope analysis in Ballarat East shows that a steep grade, ranging from 18° to 27°, is the most common roof slope, covering a total area of 222,350.81 m². This is followed by roofs with a slope of 3° to 11°, covering 103,019.75 m², and then roofs with a slope of 11° to 18°, with a total area of 61,116.83 m².

3.2.2. Ballarat East SDT4Solar Integrated Database

A database query of the Ballarat East 3DCityDB reported the existence of 9984 entities for both ‘APPEAR_TO_SURFACE_DATA’ and ‘SURFACE_DATA’, along with 4992 entities each for ‘BUILDING’ and ‘CITYOBJECT’. Additionally, the database query shows the 355,745 entities under ‘SURFACE_GEOMETRY’, and all other entity types returned a count of zero.

3.2.3. Ballarat East Rooftop Solar Potential and Environmental Impact

This study found that rooftops with a tilt angle of 30° to 31° facing north yielded the most optimal energy production. In contrast, deviations from this optimal range resulted in varying degrees of energy loss. Roof slopes that are less than 3° and between 59° and 72° were associated with an estimated 10% reduction in energy. Slopes ranging from 3° to 11° and 50° to 59° resulted in about 5% loss, while slopes between 11° to 18° and 42° to 50° led to approximately 2% energy loss. A smaller loss of around 1% was observed for slopes ranging from 18° to 27° and from 33° to 42°. Close to optimal performance, with only 0.72% loss, is expected for roofs having a tilt of 27° to 30° and 31° to 33°. However, steep slopes between 72° and 90° were associated with a significant 20% drop in energy production output. Additionally, the shadowing analysis reveals that shading occurs on average 3762 times a year, at 30-min intervals.

As a result, the calibrated solar radiation indicated that the average monthly solar observation sensor is 385.47 kW/m², representing a correction of nearly 35% from the original raw sensor reading of 597.30 kW/m². The highest discrepancy is observed in December with approximately 226 kW/m², while the lowest discrepancy is found in September, at about 39 kW/m². Figure 12 presents the monthly comparison of the raw solar radiation observations and the urban-adjusted sensor solar radiation values (kW/m²).

Overall, a total energy potential of 15,171.108 MWh or 15.17 GWh can be generated from the identified suitable roofs, with an average annual energy production of 6.12 MWh. The spatial distribution of the roof solar energy estimate is shown in Figure 13.

3.3. Comparison of SDT4Solar and Conventional Spatial Assessment Results

With the corrected solar radiation values, suitable area, and conservative estimates for both efficiency ratio and performance ratio of 15% and 86%, respectively, for the calculation of solar energy potential, the average rooftop in the locality potentially generates around 6.13 MWh of energy, which is almost 14 MWh lesser than what Kalyan and Sun [17] had reported for average energy production of residential buildings in the city of Ballarat. This difference is largely attributed to the shading effects that were not accounted for in the broader city-level estimate, the limited spatiotemporal analysis of solar radiation, and the omission of energy losses due to roof slope.

4. Discussion

4.1. Key Contributions

This paper focuses on developing SDT4Solar as a proof-of-concept, aligning with the recommendations from recent literature [1,58,60,62] on the use of digital twins. To the best of our knowledge, this represents the first application of a solar observation sensor in the context of urban rooftop PV energy planning.

A key contribution is the introduction of a framework that supports continuous solar energy potential and carbon emissions offset through dynamic solar modelling and enhanced semantic city information models. Specifically, the framework enabled the realistic modelling of the urban environment to enhance the contextualisation of the urban setting for solar energy planning. As seen in Figure 6, rooftop building features, particularly individual roof schematics, have been modelled closely to their real-world counterparts. Building features were also modelled and enriched through a standardisation process that integrated semantic information to improve understanding for each feature involved in solar modelling. Disparate solar energy-related datasets were also unified into a geodatabase, enabling efficient data management. The unification of data facilitated easy data retrieval, which is beneficial for future development, adaptations, and scaling efforts. A dynamic alternative to conventional solar potential modelling through the integration of a solar radiation IoT sensor was also implemented. Through this, a cyber-physical system of solar assessment is enabled, overcoming the limitations of traditional solar energy assessments and static digital solar maps.

4.2. Comparison with Existing Approaches

This proposed SDT approach differs from existing DT framework and implementations, such as Clausen et al. [86], where DT have been used to provide a realistic simulation of PV implementations to assess the effects of environmental factors using game engines, or Polo et al. [87] that utilised the concept of DT to model dynamic shading in PVs to increase the accuracy of building integrated PV predictions. However, both applications lacked real-world synchronisation and did not provide real-time insights, limiting their applicability for adaptive planning or decision-making.

The framework allowed the real-time data integration, providing an accurate representation of the current solar and urban conditions along with its potential PV energy and carbon emission offset estimates. This sets it apart from the conventional solar modelling and assessment frameworks provided by researchers such as [26,52,71,88,89,90,91] or [53,92], which primarily relied on static meteorological datasets or utilised typical meteorological values or the use of spatial solar modelling tools, respectively.

4.3. Technical Implementation

Integrated database provided opportunity to both import and export semantic 3D models and other energy datasets through the assistance of software and tools, including FME (workbench 2024.0.1), PostGIS (3.4.1-enabled PostreSQL 16), 3DCityDB (4.3.0) and CityGML (3.0). Similar to Diakité et al. [56] and Turner and Sun [57], a manual re-uploading approach using a third-party REST API was employed due to the size of the data. Unlike Diakité et al. [56], where the semantic information of the datasets was more structured and enriched, this study focused primarily on adding essential roof schematic details as a prototype deployment feature, rather than a fully enriched semantic model.

This study also calibrated solar data from an IoT sensor, enabling the representation of the current urban conditions while still accounting for shadowing impacts and roof schematics for individual buildings, to eliminate overestimations or percentage error from a spatial perspective. Setting this study apart from Kalyan and Sun [17], which had previously examined modelling and planning the solar potential of the study area.

The testbed SDT4Solar generated insights into the spatial variation of rooftop solar radiation, spatial patterns of solar electricity potential, and the spatial distribution of carbon offsets through the visualisation of location-specific and time-sensitive solar potential across the case study area. Unlike previous studies [51,93,94,95] that presented web-interfaces that relied on static mapping services, this study leveraged CesiumJS to deliver a fully interactive and rich 3D interface, thereby setting the stage for the completion of a continuous feedback loop for iterative solar planning, as seen in Figure 2.

Aligning with the DT principles, the SDT4Solar testbed can act as a decision-making support tool due to more accessible and quicker solar potential analysis. In addition, the prototype promotes transparency in both data source and usage, reinforcing data ethics and accountability. The custom functions within the prototype allowed the retrieval and display of key parameters of solar planning, fostering trust among users and data providers. This promotes a collaborative environment for spatial planning interventions. Furthermore, with most datasets managed through a centralised database and solar radiation data sourced directly from the Local Government server, the system ensures data security and safeguards the integrity of information.

4.4. Significance of SDT4Solar in Ballarat East

Based on the results of the implementation, the application of the proof-of-concept SDT4Solar framework to Ballarat East offered key insights to the locality’s current potential, aligning with its long-term environmental vision, and opportunities for future development.

From the identified buildings (as shown in Figure 11), the average suitable roof face area is 32.81 m², indicating that more PV stringing or separate PV deployment within roofs is required, which leads to increased costs and energy generation losses. These findings underscore the benefit of incorporating larger roof planes on optimal aspects for future building designs to support future-proofing renewable energy efforts.

An estimated total power generation loss of at least 2.39% is expected across suboptimal roofs if PV systems are mounted parallel to existing slopes. With the notable number of roofs with suboptimal roof slopes, these findings highlight the importance of roof slope in estimating energy potential and system capacity. Additionally, it highlights how PV mounting and placement can impact the long-term performance of PV systems.

Furthermore, the shadowing analysis of this study confirms the significant spatial impact of surrounding vegetation on shading-related energy losses, as seen in Figure 14. This underscores the practical importance of including detailed shadow analysis, especially for peak sunlight hours, to maximise PV generation. Consequently, strategic identification and placement of future PV systems are essential to avoid underperforming or dormant panels on roofs. From a planning perspective, these insights can inform building setbacks to maximise solar access.

Comparing the current carbon budget of 14.7 million t CO₂-e for the whole city of Ballarat and the total estimated annual carbon emission offset of 13,047.134 t CO₂-e implies that Ballarat would need more than 1000 Ballarat East locality equivalents to neutralise the carbon budget. This technical potential reflects the in situ analysed renewable energy potential for actual urban conditions and constraints.

4.5. Role of SDT in City-Scale Solar Energy Planning

The rapid increase in big data accessibility and technological advancements paved the way for making SDTs feasible for city-scale solar energy planning. Through this, the adoption of city-scale SDTs in solar planning introduces a shift from static and fragmented solar assessments to a dynamic and integrated modelling system. Unlike most geospatial-based approaches, which rely on simplified representations of the urban environment, as mentioned in Section 1.1, SDTs enable the visualisation and rich representation of the physical world through the added semantic layer within 3D urban models. This enhances stakeholder understanding by providing greater contextual awareness when utilised in an interactive environment.

Additionally, the persistent challenges in data harmonisation and the time-consuming nature of dataset integration in solar energy modelling and estimation were also effectively addressed through the adaptation of SDTs through an integrated database with 3DCityDB, and data standardisation through CityGML. The integration of data allowed more efficient processing using consistent datasets, thereby reducing analysis times, supporting cyclic analysis workflows, and enhancing scalability for broader urban applications.

Unlike the conventional approach of utilising historical datasets or estimated solar irradiance values, SDTs can account for temporal variations, particularly shading and solar radiation fluctuations, in the context of solar energy planning. This capability helps reduce estimation errors and improve the reliability of solar potential estimations.

The developed methodology in this study enabled the real-time modelling of solar energy at a minimum spatial resolution of 10 cm and temporal resolution of 15 min, accounting for shadowing and real 3D urban conditions. This represents a comparative advancement for GIS-based methodologies reviewed by Massano et al. [39] for the estimation of renewable energy system potentials. In addition, the approach overcomes the limitations of coarse spatial resolution inherent in ground-measured solar radiation data, allowing for its application at the rooftop level.

Furthermore, SDTs enhance stakeholder engagement by adding interactivity to solar information, fostering participatory planning and feedback, as presented in Figure 2, and emphasised by Alvarado et al. [51]. While this study presents a proof of concept, it successfully addresses many of the foundational principles of digital twins as presented by Bolton et al. [64] and helps support initiatives towards renewable energy transitions [75] through data-driven and scalable solutions.

4.6. Limitations and Future Directions

This study presents a feasible approach for solar energy planning; however, barriers still exist in this proof-of-concept study. The primary limitation of the technical framework lies in the substantial computational resources required to process LiDAR point clouds. Although LiDAR provides detailed 3D representations of the urban environment, its processing demands computation resources, especially when scaling the application to larger areas.

To further increase the real-time functionality of the SDT4Solar system, future developments should focus on integrating edge computing along with the cloud-based geospatial infrastructure. This hybrid approach would reduce the latency of the data analysis, creating a more robust system that delivers timely decision-making insights. Additionally, expanding the current system to incorporate other energy-related datasets, particularly existing individual residential PV energy generation data and household energy consumption, can enhance the accuracy and contextual relevance of solar potential estimates. Extending SDT4Solar into existing energy systems would enable a more holistic perspective of the overall energy supply and demand dynamics, thereby facilitating data-driven energy planning. While substantial time and effort are required to initially build and set up an SDT, this can be offset in the long term by the reduced planning errors and increased resiliency to future scenarios.

With the focus of this study on investigating the applicability of dynamic solar potential modelling within a digital twin framework, only a prototype SDT testbed was developed. As a result, the complete cyclic loop of the SDT4Solar concept (as seen in Figure 2) has not been fully implemented. To realise the full potential of the SDT4Solar approach, future deployment should aim to add planning features and functionalities. Future implementations of this study could be expanded by integrating IoT-enabled distributed energy grid data, further enhancing urban solar management. This would facilitate seamless analysis of solar energy potential, aligning closely with the goals of sustainable energy management.

Future work can also be scaled on a larger scale with automated simulation of digital surface models as new data becomes available to increase the precision of the identification of suitable building roofs as the urban environment changes.

5. Conclusions

To sustainably power future urban communities, advanced solar energy planning tools are crucial for overcoming the limitations of traditional methods. This study presents SDT4Solar, an SDT-enabled framework that integrates city-scale rooftop solar planning through 3D building sematisation, solar modelling, and a geospatial database.

By integrating dynamic modelling with advanced spatial analysis and IoT, SDT4Solar facilitates 3D rooftop PV simulations that enhance both the accuracy and equity of solar deployment. Applied to Ballarat East Area in Victoria, Australia, the framework demonstrated its practical significance by estimating an annual rooftop generation of 15.17 GWh and an associated carbon emission offset of 13,047.134 t CO₂-e across a total rooftop area of 18,294.0625 m². These spatially explicit insights and quantifiable results provide valuable guidance for local government strategies aimed at achieving net-zero ambitions.

The strength of SDT4Solar lies in its modular technology architecture and interoperability, enabling scalability from neighbourhood energy planning to metropolitan-scale renewable deployment. Furthermore, the framework provides a replicable and data-driven workflow specifically designed to accommodate future enhancements and integration into broader urban energy systems. This ensures its continued relevance in addressing dynamic and complex challenges in urban energy sustainability.

However, the framework faces limitations including the high computational demands of processing multi-source datasets, which affect responsiveness and real-time capabilities. Moreover, while the SDT4Solar advances rooftop solar energy planning through dynamic solar radiation modelling and geospatial integration, it currently lacks incorporation of socio-economic and regulatory factors that critically shape solar PV adoption. Future research should expand the framework to include these dimensions, which are critical for supporting inclusive and equitable solar planning strategies within diverse urban environments.

Overall, this proof-of-concept validates the feasibility and potential of leveraging SDTs for urban solar energy planning. It advances the roles of digital twins in supporting data-driven and spatially explicit decision-making that can accelerate sustainable urban energy transitions.