The main goal of this SLR is to identify the MDE techniques used for smart cities, the benefits of using MDE, and the particular challenges encountered in doing so. The following research questions are formulated to achieve this goal.

The population in this study is the intersection of the domains of MDE and smart cities. Research studies published in peer-reviewed journals, conferences, and book chapters from January 2019 to August 2024 are considered. Fig. 1 depicts the research methodology used to identify primary studies. The rest of this section provides descriptions of the phases of the study protocol. The results of the study are covered in Section 3.

We searched four scientific databases, including the Sciencedirect, ACM Digital Library, IEEE Xplore Digital Library, and SpringerLink, to find relevant articles for our systematic review. The search strategy was formulated after conducting a pilot search using different combinations of the terms "model-driven engineering", "MDE", "smart city", and "smart cities". The search was restricted to title, abstract, and keywords. Also, other terms such as "urban" and "intelligent" were used, but they gave us unrelated papers on smart cities. Table 1 presents the exact strings used in each database along with the frequency of the retrieved papers.

Establishing exclusion and inclusion criteria is a crucial phase in the research selection procedure. A study was excluded if it:

We include all studies that address aspects of MDE and its relationship to the smart city. The study selection was done in multiple steps (see Fig. 1):

3. In the initial stage of the research selection process, we reviewed the abstracts and titles of 223 publications, discarding those unrelated to MDE and smart cities. The specified exclusion criteria were applied to categorize each paper as relevant, non-related, or unclear. This stage concluded with 4 duplicated studies, 20 relevant documents, 92 unrelated, and 127 unclear studies.

4. The second step involved examining the unclear studies in full to assess their relevance to MDE applications in the smart city domain. For each paper, we began by reviewing the introduction and conclusion sections, and then examined other sections as needed for decision-making. This full-text review resulted in 74 papers identified as relevant.

5. The last step of study selection was to read the full text of the 94 papers. We had a total of 42 primary studies for our SLR.

Model-Driven Engineering (MDE) tools are pivotal in addressing smart city challenges through domain-specific abstractions and automation, as highlighted in a review of the application of emerging technologies in [17]. The SI4IoT methodology proposed in [10] employs a domain-specific language (DSL) and model-to-text transformations to integrate heterogeneous IoT systems across platforms like Arduino and Node-Red, validated through performance tests using Gatling. InterSCity, introduced in [18], was an open-source microservices platform designed for scalable smart city applications, tested via large-scale simulations with OpenStreetMap and Sim-Diasca. Industrial IoT (IIoT) interoperability is handled in [19] by proposing a Model-Driven Architecture (MDA) framework, which reduces development effort for heterogeneous devices, demonstrated through a smart city case study. A study reported in [20] simplifies IoT application development with AutoIoT, a user-driven framework that generates server-side code from JSON models, evaluated through an experiment involving 54 developers. Despite these advancements, tools FaultFlow, detailed in [21], which automates failure analysis through SysML-to-STPN transformations, reveal limitations in scalability testing, as seen in its Pollution Monitor System case study. The MIKADO framework in [22] addresses KPI assessment challenges using metamodel-based DSLs, validated through a demonstration case and scalability studies. Research reported in [23] introduces an MDE framework for Web of Things (WoT), enabling protocol interoperability and application heterogeneity, validated through an LED control case study on an ESP8266 IoT device. An integrated architecture combining MD-DSM and smart city platforms to facilitate application development across domains is proposed in [24]. A DSL for FIWARE-based IoT simulations, validated through case studies in smart building and agricultural IoT environments, is provided in [25]. Table 2 presents the MDE tools along with their types and applications in the domain of smart cities.

Several studies, like [26], discussed that security and privacy solutions in smart cities emphasize model-based analysis and user-centric consent management. Research presented in [27] proposes SoSSecML, a modelling language for secure Systems-of-Systems (SoS), combined with Multi-Agent Systems (MAS) to predict cascading attacks in smart buildings, validated through a case study of Adelaide University's energy systems during peak consumption hours. The "privacy paradox" in IoT with the Informed Consent Management Engine (ICME), addressed [28], which provides fine-grained visibility into data collection practices, was tested via a Node-RED implementation and interviews with 15 industry experts. Formalization of safety and security verification for IoT applications using rewriting logic (Maude) is presented in [29], transforming ThingML specifications into analyzable models, as demonstrated in the PingPong IoT case study. While [30] systematically reviews 803 studies on model-based security testing in IoT, highlighting gaps in empirical validation, [21] presents a FaultFlow tool that automates failure analysis for IoT systems through SysML-to-STPN transformations, tested on a Pollution Monitor System. These approaches predominantly rely on simulations and case

studies, with limited integration into broader MDE workflows. MDA is used in [31] to ensure Role-Based Access Control (RBAC) in IoT devices, validated through a smart door lock case study, which grants or denies access based on sensor inputs. Fig. 3 depicts the distribution of the validation methods used in studies discussing the security and privacy aspects of MDE in the domain of smart cities.

Scalability and interoperability solutions focus on distributed architectures and semantic integration. InterSCity platform leverages microservices and simulations (OpenStreetMap, InterSCSimulator) to validate scalability for smart city applications [18]. A model-based fleet deployment framework for the IoT-edge-cloud continuum, tested in collaboration with a smart healthcare provider to automate software orchestration across edge gateways, is proposed in [18]. A study reported in [11] integrates smart traffic fog nodes and homes using a model-driven approach, validated through a realworld case study. Meanwhile, [33] systematically reviews existing Fog modelling languages to identify gaps in standardization and interoperability for edge-fog-cloud systems. Optimization of Quality of Service (QoS) in vehicular fog computing (VFC) through dynamic resource provisioning, using a multi-objective optimization model to balance latency and resource utilization, is presented in [34]. Semantic interoperability is addressed in [35] through knowledge base overlays to unify heterogeneous IoT data streams, and in [36] via the Distributed Data Interoperability Layer (DDIL) for smart grids. Despite progress, standardization remains a challenge, particularly for edge-fog-cloud deployments. Table 3 presents the proposed MDE-based scalability solutions along with validation methods in the domain of smart cities.

Digital Twins (DTs) and 3D modelling are increasingly applied for predictive analytics and system resilience. In [37], authors propose an Urban Intelligence, a modular DT framework which combines multidisciplinary city models (infrastructure, user behaviour) with numerical optimization, targeting urban planning challenges. Enhancement of cyber-physical system resilience through the integration of digital twins with failure models, as validated in smart city and industrial IoT scenarios, is discussed in [38]. In contrast, [39] analyzes the transition from 3D modelling to DTs in smart manufacturing and transportation, highlighting limitations in real-time synchronization and applicability across domains. "Digital Dice", introduced in [39], is a W3C-compliant virtual representation of IoT devices using microservices, demonstrated in a smart building case study. While these frameworks show promise in predictive maintenance (e.g., [38]) and urban planning (e.g., [37]), adoption in domains like waste management remains limited. A framework for synchronizing is introduced [41], city spatial models with graph-based analyzable models, enabling safe model-driven development of composite systems in physical spaces. Synchronization is enabled in [42] between CityGML models and analysable models, ensuring consistency in physical space modelling for composite systems. Fig. 4 depicts the distribution of the DT applications by domains.

Strategic alignment bridges technical implementations with city management goals through standardized frameworks and KPIs. Design principles for Smart City Enterprise Architectures (SCEA) are derived from a review of the literature and validated through case studies of two cities [43]. ICT criteria, as incorporated in [44], were used in smart city rankings through a Multiple-Attribute Decision Making (MADM) approach, tested on New York, Seoul, and Santander. A study reported in [45] automates dashboard development with Cities-Board, transforming KPI lists into interactive visualizations for urban data. In [67], authors explore model-driven optimization for 6G network planning, although limited to single-objective functions. These efforts highlight the need for domain-specific KPIs (e.g., MADM in [44]) and automated governance tools (e.g., Cities-Board in [45]) to align technical deployments with strategic objectives. A model-driven approach, as outlined in [46], offers a technology-agnostic solution for transforming KPIs into interactive dashboards, thereby reducing manual effort and errors in dashboard setup. Real-time IoT applications for connected vehicles are targeted in [47], utilizing communication protocols such as Wi-Fi, Bluetooth, and 5G to enhance the Quality of Experience (QoE). Table 4 presents strategic alignment approaches along with their use cases proposed in the domain of smart cities.

Testing and verification ensure system reliability through formal methods and empirical validation. In [48], authors analysed 803 studies on Robotic and Autonomous Systems, identifying temporal logic and state machines as widely used modelling approaches, and highlighting gaps in industrial applicability. Research work discussed in [29] formalizes the semantics of ThingML using rewriting logic (Maude), enabling automated verification of IoT applications, as demonstrated in the PingPong distributed services case study. STAS for runtime generation of component-based applications, proposed in [49] and validated through mashup user interfaces, demonstrating the importance of dynamic testing in MDE. A hardware-in-the-loop (HIL) testbed for edge-based cyber-physical systems, simulating environmental scenarios to evaluate dependability in smart city mobility, is developed [50]. In [51], the presented approach ensures functional correctness in WoT applications through model checking, extending the Mozilla WebThings platform. A CI/CD metamodel to standardize IoT architectures is highlighted in [52], though empirical validation remains limited. While simulations dominate validation (e.g., [18]), tools FaultFlow [21] demonstrate the potential of automated failure analysis, albeit with scalability constraints. "Ref. [53]" proposes MLM to streamline the development of QESs, demonstrating its applicability through running examples. Fig. 5 depicts the distribution of the verification methods in testing studies, with simulations as the most widely used verification method.

The six themes identified in Section 3 highlight both the transformative potential and the persistent challenges of applying MDE in the domain of smart cities.

Tools and Techniques: The growing use of DSLs and model-to-text transformations, as seen in tools like SI4IoT and AutoIoT, demonstrates the adaptability of MDE in addressing the heterogeneity and complexity of smart city systems. These tools allow developers to abstract away low-level implementation details, enabling a stronger focus on high-level design and faster prototyping. However, scalability remains a critical issue. For example, while InterSCity shows promising performance in large-scale simulations, its real-world effectiveness across diverse urban settings-each with unique infrastructural, socio-economic, and regulatory challenges-has not yet been thoroughly evaluated. This gap emphasizes the need for tools that are not only technically robust but also capable of adapting to varying contextual demands.

Security, Privacy, Scalability, and Interoperability: The increasing reliance on IoT devices in smart cities has amplified concerns around data security and user privacy. Model-based approaches like SoSSecML and ICME offer innovative solutions by embedding security and privacy considerations into the design phase. However, these approaches often rely on simulations or limited case studies, raising questions about their scalability and effectiveness in real-world deployments. For instance, while ICME provides fine-grained visibility into data collection practices, its reliance on expert interviews for validation limits its generalizability. Moreover, the dynamic nature of cyber threats necessitates continuous updates to these models, which is often overlooked in current implementations. The distributed nature of smart city systems necessitates scalable and interoperable solutions. Frameworks like InterSCity and DDIL address these challenges through distributed architectures and semantic integration. However, the lack of standardization, particularly in edge-fog-cloud deployments, remains a significant barrier. For example, while [33] identifies gaps in Fog modelling languages, there is a pressing need for standardized frameworks that can seamlessly integrate edge, fog, and cloud resources. This lack of standardization not only hampers interoperability but also increases the complexity of system integration and maintenance.

Digital Twins and Emerging Technologies: DTs represent a paradigm shift in how smart city systems are designed, monitored, and optimized. Frameworks like Urban Intelligence and Digital Dice demonstrate the potential of DTs in urban planning and smart building management. However, their adoption in other domains, such as waste management and transportation, is still limited. Additionally, real-time synchronization across heterogeneous systems remains a challenge, as highlighted by [39]. This limitation is particularly critical in applications requiring real-time decisionmaking, such as traffic management and emergency response.

Strategic Alignment and Enterprise Architecture: Aligning technical implementations with city management goals is crucial for the success of smart city projects. Frameworks like SCEA and tools like Cities-Board provide valuable insights into this alignment. However, the development of domain-specific KPIs and automated governance tools is still in its early stages. For instance, while [44] incorporates ICT criteria into smart city rankings, there is a need for more comprehensive frameworks that can evaluate the socio-economic impact of smart city initiatives. This gap underscores the importance of developing metrics that capture not only technical performance but also social equity, environmental sustainability, and economic viability.

Testing and Verification: Ensuring the reliability of smart city systems through rigorous testing and verification is essential. Formal methods, such as rewriting logic and model checking, are widely used for this purpose. However, the reliance on simulations for validation, as seen in [18] and [50], limits the generalizability of these methods. Tools like

FaultFlow demonstrate the potential for automated failure analysis, but their scalability in large-scale deployments remains unproven. This limitation highlights the need for more robust testing frameworks that can handle the complexity and scale of real-world smart city systems.

The findings provide clear answers to the research questions posed in this study:

RQ1: What MDE techniques have been used for smart cities? MDE techniques applied in smart cities include DSLs, model-to-text transformations, and formal modelling approaches. Tools like SI4IoT and AutoIoT leverage DSLs to integrate heterogeneous IoT systems, while frameworks such as InterSCity use microservices for scalable smart city applications. Additionally, formal methods like rewriting logic and model checking are employed for system verification and validation. These techniques enable developers to abstract low-level implementation details, focus on high-level design, and accelerate prototyping. However, their adoption varies across domains, with some areas like urban planning and smart building management showing more progress than others, such as waste management and transportation.

RQ2: What are the potential advantages of using MDE techniques to address smart city challenges? MDE techniques offer several advantages in addressing smart city challenges. They enable the abstraction of complex system details, allowing developers to focus on high-level design and rapid prototyping. Tools like SI4IoT and InterSCity demonstrate how MDE can streamline the integration of heterogeneous IoT systems and improve scalability. Furthermore, MDE supports the embedding of security and privacy considerations into the design phase, as seen in frameworks like SoSSecML and ICME. By providing modular and hierarchical modelling capabilities, MDE techniques also facilitate the development of scalable and interoperable solutions that are critical for the distributed nature of smart city systems.

RQ3: What are the challenges of adopting MDE techniques for smart cities? Despite their advantages, adopting MDE techniques for smart cities presents several challenges. Scalability remains a significant issue, as many tools, such as InterSCity, perform well in simulations but lack real-world validation across diverse urban contexts. The lack of standardization, particularly in edge-fog-cloud deployments, hampers interoperability and increases system integration complexity. Security and privacy concerns are also prominent, especially in IoT deployments, where dynamic cyber threats require continuous model updates. Additionally, the reliance on simulations for validation, as seen in tools like SoSSecML and FaultFlow, limits the generalizability of these methods. Finally, integrating MDE with legacy systems and emerging technologies like 5G and blockchain remains a work in progress, highlighting the need for further research and development.

It is also important to highlight that our findings have significant implications for various stakeholders involved in smart city development. Researchers should focus on addressing the scalability and interoperability challenges of MDE tools and techniques. More empirical studies and real-world validations are needed to ensure the practical applicability of these tools. Additionally, exploring the integration of MDE with emerging technologies, such as 5G and blockchain, presents a promising avenue for future research. Practitioners can leverage existing MDE tools, such as SI4IoT and InterSCity, to streamline the development of smart city applications. However, they must be mindful of the scalability and interoperability limitations of these tools. Adopting standardized frameworks and governance tools can help align technical deployments with strategic city management goals. Policymakers should prioritize the development of standardized frameworks and governance tools to ensure the successful implementation of smart city projects. Investments in Digital Twins and predictive analytics can enhance urban planning and system resilience. Additionally, fostering collaboration between academia, industry, and government can accelerate the adoption of MDE in smart cities.

There can be different threats to the validity of study results. First, while the research provides a comprehensive review of MDE techniques for smart cities, it is limited to studies published between January 2019 and August 2024. This timeframe, though extensive, may exclude earlier foundational works that could offer additional insights into the evolution of MDE in smart cities. As a result, the findings may not fully capture the historical development and longterm trends in the field.

Second, the scope of the literature review is primarily limited to peer-reviewed peer reviewed journals, conferences, and book chapters, which may exclude valuable insights from industry reports, grey literature, and practical case studies. This could introduce a bias toward academic perspectives, potentially overlooking practical challenges and solutions identified in real-world implementations.

Third, the analysis is restricted to studies published in English, potentially overlooking contributions from non-Englishspeaking regions. This may limit the generalizability of the findings, as smart city initiatives in non-English-speaking countries may face unique challenges and adopt different approaches.

Finally, the field of smart cities is rapidly evolving, with new tools, frameworks, and technologies emerging continuously. While this study captures developments up to August 2024, some findings may become outdated as the field progresses. These validity threats highlight the need for future research to broaden the temporal scope, include diverse linguistic and regional perspectives, and account for the fast-paced evolution of smart city technologies.