Impact statements
This research provides a critical roadmap for the future of water management by outlining a 30-year evolution of digital twin (DT) systems, intelligent virtual replicas of our natural and engineered water systems. As climate change intensifies the frequency of extreme flooding and water scarcity, traditional static modeling tools are no longer sufficient to ensure community safety. By introducing a structured five-level framework, this study guides the transition from basic digital maps to sophisticated, autonomous, artificial intelligence (AI)-driven systems capable of monitoring and managing water resources in real time.
The primary impact of this work lies in its role as a strategic blueprint for scientists, urban planners and policymakers worldwide. Rather than treating DT as isolated technical experiments, this research redefines them as sociotechnical infrastructures built for the public good. By advancing toward higher levels of autonomy, disaster response teams can utilize AI that not only predicts floods with improving lead time and accuracy but also coordinates critical infrastructure such as dams and drainage systems to mitigate damage before it occurs. Furthermore, the integration of immersive 3D environments and conversational AI makes complex hydrological data intuitive for nonexperts, fostering transparent and participatory governance where citizens and officials can codesign climate adaptation strategies together.
Ultimately, this research offers a pathway for interconnected water management across political and ecological boundaries using open standards and federated learning. It transforms how society interacts with the environment by turning complex data into a transparent, actionable tool for building a more resilient, equitable and water-secure world.
Highlights
-
• Maps the 30-year technological evolution of hydrological digital twins from physical models to autonomous and interconnected systems.
-
• Proposes a dual-trajectory vision advancing vertical artificial intelligence (AI) autonomy and horizontal cross-domain federation.
-
• Presents a layered implementation framework and capability matrix linking enabling technologies to functional maturity.
-
• Outlines a research agenda prioritizing agentic AI, interconnected systems and ethical water governance.
Introduction
Digital twin (DT) technology has emerged as a transformative paradigm across multiple domains, enabling the creation of dynamic, real-time digital replicas of physical systems for enhanced monitoring, simulation and decision-making (Qian et al., Reference Qian, Liu, Ripley, Qian, Liang and Yu2022). The conceptual foundation was originally introduced by Michael Grieves in 2002 as the Conceptual Ideal for Product Lifecycle Management (Grieves, Reference Grieves2005). This framework matured when the specific DT terminology was formally adopted within the National Aeronautics and Space Administration (NASA) Technology Roadmap (NASA Roadmap, 2010) and subsequently adapted to model spacecraft systems (Glaessgen and Stargel, Reference Glaessgen and Stargel2012). A DT is formally defined as being composed of a physical entity, the virtual model and the continuous bidirectional connection between them (Grieves and Vickers, Reference Grieves and Vickers2016; El Saddik, Reference El Saddik2018). DTs have evolved far beyond their manufacturing and aerospace origins to play pivotal roles in healthcare, transportation, energy, smart cities and environmental sciences (Barricelli et al., Reference Barricelli, Casiraghi and Fogli2019; Tao et al., Reference Tao, Zhang, Liu and Nee2019; Hämäläinen, Reference Hämäläinen2021). Within hydrology, the adoption of DTs presents a critical opportunity to transform the monitoring, understanding and management of water systems amid growing uncertainty and climate-induced variability (Rigon et al., Reference Rigon, Formetta, Bancheri, Tubini, D’Amato, David and Massari2022; Henriksen et al., Reference Henriksen, Schneider, Koch, Ondracek, Troldborg, Seidenfaden, Kragh, Bøgh and Stisen2023).
Unlike traditional physically based models that rely on static, historical datasets, or Internet of Things (IoT) sensor networks that merely collect one-way data feeds, a hydrological DT is distinguished by its continuous bidirectional synchronization. It acts as a living system where real-time physical data constantly update the virtual model, and the virtual model’s predictive analytics inform automated or human-in-the-loop decisions in the physical world. Although recent literature has proposed structural blueprints for these systems, such as the multilayer framework by Liu et al. (Reference Liu, Guo, Wang and Peng2018), the Earth DT layers for environmental applications (Al-Obeidat et al., Reference Al-Obeidat, AlTaee, Abbes, Hamlich, Dornaika, Ordonez, Bellatreche and Moutachaouik2024; Brocca et al., Reference Brocca, Barbetta, Camici, Ciabatta, Dari, Filippucci, Massari, Modanesi, Tarpanelli, Bonaccorsi, Mosaffa, Wagner, Vreugdenhil, Quast, Alfieri, Gabellani, Avanzi, Rains, Miralles, Mantovani, Briese, Domeneghetti, Jacob, Castelli, Camps-Valls, Volden and Fernandez2024), and specifically the six-dimensional architectural blueprint for river basin DTs introduced by Pal et al. (Reference Pal, Marttila, Ala-Aho, Lotsari, Ronkanen, Gonzales-Inca, Croghan, Korppoo, Kämäri, Rooijen, Blåfield, Silander, Baubekova, Bhattacharjee, Haghighi, Uvo, Kaartinen, Rasti, Klöve and Alho2025), this position article addresses a different critical gap. Rather than focusing solely on system architecture, we propose a five-stage technological evolution model that maps the historical and future trajectory of DT capabilities, illustrating how they transition from static replicas to autonomous, federated systems.
Traditionally, hydrological modeling has relied on static or semidynamic tools such as physics-based simulations, geographical information systems (GIS) platforms and scenario-based forecasting to address flood modeling, watershed planning and groundwater analysis (Singh, Reference Singh2016, Reference Singh2018; Krajewski et al., Reference Krajewski, Ghimire, Demir and Mantilla2021). However, as hydrological events grow in frequency, intensity and complexity due to climate change, urbanization and land use, these conventional approaches and tools lack the capacity for real-time responsiveness, integration of heterogeneous datasets, interoperability and adaptive behavior under dynamic or extreme conditions (Henriksen et al., Reference Henriksen, Schneider, Koch, Ondracek, Troldborg, Seidenfaden, Kragh, Bøgh and Stisen2023; Ge and Qin, Reference Ge and Qin2025). Consequently, there is an urgent need for intelligent modeling frameworks capable of integrating real-time data from sensor networks, satellite observations and cloud platforms with advanced simulation capabilities, immersive visualization and artificial intelligence (AI)-driven analytics (Park and You, Reference Park and You2023; Li and Demir, Reference Li and Demir2024; Manocha et al., Reference Manocha, Sood and Bhatia2024).
The timing for a hydrological DT revolution could not be more critical. Over the past 5 years, there have been significant breakthroughs in AI, immersive web technologies, edge computing and the metaverse. AI approaches, including deep learning (DL), generative modeling and reinforcement learning, have matured to the point that predictive accuracy in hydrology has improved significantly (Sit et al., Reference Sit, Demiray, Xiang, Ewing, Sermet and Demir2020; Sajja et al., Reference Sajja, Xiong, Mermer, Sermet and Demir2025a). Specifically, this enhanced precision is achieved through the integration of physics-informed machine learning (ML), which embeds governing physical laws, such as acoustic scattering or hydrodynamic flow, directly into neural networks to ensure that predictions remain physically valid even with sparse data (Nie et al., Reference Nie, Yu, Wang and Liu2025). Furthermore, the use of ensemble ML models allows for robust, high-resolution temporal forecasting, whereas explainable AI techniques, such as Shapley additive explanations analysis, provide critical model interpretation, unmasking the black box of complex algorithms for decision-makers; what was once limited by sparse data and rigid models is now attainable with unprecedented precision (Kreuzer et al., Reference Kreuzer, Papapetrou and Zdravkovic2024; Mermer et al., Reference Mermer, Zhang and Demir2025). These advanced AI tools provide powerful, transparent mechanisms for predictive modeling, anomaly detection and decision support in complex water systems without constant manual calibration (Demiray et al., Reference Demiray, Sit, Mermer and Demir2024; Abdulameer et al., Reference Abdulameer, Al-Khafaji, Al-Awadi, Al Maimuri, Al-Shammari and Al-Dujaili2025).
Concurrently, immersive visualization tools (augmented reality [AR]/virtual reality [VR]/extended reality [XR]), gaming engines and web-based collaborative platforms have reshaped participatory environmental modeling and stakeholder engagement (Chandramouli et al., Reference Chandramouli, Zhou, Moreland, Teegavarapu and Chandramouli2022; Oyshi et al., Reference Oyshi, Maleska, Schanze, Bormann, Dachselt and Gumhold2022). The rise of the metaverse, conversational AI and Web 3.0 ecosystems indicates that hydrological DTs must also evolve, becoming more user-centric, adaptable and platform-interoperable. Meanwhile, scalable cloud architectures and IoT infrastructures continue to support real-time synchronization between digital and physical hydrological systems (Minerva et al., Reference Minerva, Lee and Crespi2020; Kaynak et al., Reference Kaynak, Mermer, Sermet and Demir2025). The convergence of these technological advancements marks a pivotal moment: DTs must transition from passive replicas to active, autonomous systems that can support real-time, cross-domain governance and long-term water sustainability strategies.
In this position article, we argue that the trajectory of DT technology in hydrology must expand beyond traditional frameworks to embrace intelligent, autonomous and federated capabilities. We propose that DTs must evolve along two simultaneous trajectories: (1) vertically, toward higher levels of intelligence, autonomy and adaptability, and (2) horizontally, toward greater integration and federation across domains such as meteorology, infrastructure, policy and ecology. Our proposed DT technological evolution model maps this progression, correlating enabling technologies with functional capabilities, including virtualization, real-time synchronization, advanced modeling, AI-based analytics and autonomous operation.
Crucially, we advocate for viewing hydrological DTs not only as technical systems but also as sociotechnical infrastructures designed for participatory governance, transparency and ethical AI integration. Key design considerations must include data governance, system transparency, interoperability and equitable access. We envision a future in which DTs of rivers, aquifers and watersheds are comanaged by AI agents, public institutions, scientists and community stakeholders working together to enhance climate resilience, optimize water resource management and foster inclusive governance.
This study presents a narrative synthesis and perspective by drawing upon a curated selection of seminal technological milestones and recent advances in hydrological modeling to synthesize the evolving landscape of DT applications. Building on this synthesis, we propose a five-stage technological evolution model for hydrological DTs that provides a conceptual roadmap for the field’s progression from static digital replicas to autonomous, federated intelligent systems.
This article is organized as follows: The “Background on DT in hydrology” section provides background on DTs in hydrology, defining their conceptual components and detailing their current applications. The “Technological evolution of DT” section presents the proposed five-level DT evolution model. The “Key considerations and recommendations” section highlights key considerations and offers targeted recommendations related to data interoperability, computational complexity and ethical collaboration. Finally, the “Conclusion and future directions” section concludes the article by summarizing the key findings and outlining future directions for intelligent, federated and ethically grounded DT ecosystems in hydrology.
Background on DT in hydrology
Definition and conceptual framework
Although the introduction established the general DT concept, within the specific context of hydrology, a DT is defined as an intelligent, dynamic and interactive digital representation of natural and engineered water systems, such as river basins, reservoirs, groundwater aquifers, urban drainage and watershed-scale ecosystems that continuously mirrors and evolves alongside its physical counterpart using real-time data, analytics and simulation tools (Qian et al., Reference Qian, Liu, Ripley, Qian, Liang and Yu2022; Rigon et al., Reference Rigon, Formetta, Bancheri, Tubini, D’Amato, David and Massari2022; Crespi et al., Reference Crespi, Drobot and Minerva2023; Henriksen et al., Reference Henriksen, Schneider, Koch, Ondracek, Troldborg, Seidenfaden, Kragh, Bøgh and Stisen2023; Park and You, Reference Park and You2023; Shao et al., Reference Shao, Zheng, Xu, Gu, Qin and Hu2025; Bigdeli, Reference Bigdeli2026).
Unlike traditional hydrological models that rely on static data and predefined assumptions, a DT is characterized by continuous, bidirectional synchronization between the physical environment and the digital model. For example, during extreme weather events, rather than simply issuing a passive forecast, a DT dynamically assimilates real-time sensor data to predict flood inundation while simultaneously testing mitigation scenarios and potentially interfacing directly with physical infrastructures to optimize automated responses, such as adjusting dam releases or urban drainage systems (Park and You, Reference Park and You2023; Suquet et al., Reference Suquet, Nguyen, Ricci, Piacentini, Bonassies, Fatras, Lavergne, Brunato, Gaudissart, Guzzonatto, Froidevaux, Guiot, Valladeau, Poisson, Huang, Bretar, Kettig and Blanchet2023; Ge and Qin, Reference Ge and Qin2025).
Furthermore, although traditional hydrological modeling has long relied on scenario testing, DTs transform this process into a live capability. What-if scenarios are generated and continuously updated in real time as the baseline conditions of the physical environment change, providing decision-makers with an adaptable platform to visualize, monitor and refine sustainable management strategies during unfolding environmental events (Brahmbhatt et al., Reference Brahmbhatt, Maheshwari and Gudi2023). To achieve this, DTs integrate real-time measurements from distributed sensor networks, crowdsourced citizen monitoring, remote sensing platforms and hydrological databases (Sermet et al., Reference Sermet, Villanueva, Sit and Demir2020a).
To achieve the dynamic bidirectional capabilities that separate a hydrological DT from traditional modeling, these systems rely on a structured architectural framework. Four technological and foundational categories define the system architecture and functionality: data and synchronization, visualization and communication, modeling and analysis, and governance and operation (Figure 1).
Conceptual components of DT framework in hydrology.
Data and synchronization: Reliable, high-resolution and continuous data streams are the backbone of any DT. In hydrology, data are acquired from a combination of in situ sensors, satellite observations, remote sensing platforms and historical records, including precipitation, streamflow, water quality, soil moisture and groundwater levels. Recent advancements in the assimilation of Sentinel-based Leaf Area Index for modeling surface-ground water interactions in irrigation districts exemplify how high-resolution remote sensing can seamlessly update and constrain digital models (Zafarmomen et al., Reference Zafarmomen, Alizadeh, Bayat, Ehtiat and Moradkhani2024). IoT-enabled hydrological sensor networks and telemetry systems provide up-to-date measurements, which are transmitted using technologies such as 5G (5th Generation) mobile networks, LPWAN (Low-Power Wide-Area Network) and satellite links (Zanella et al., Reference Zanella, Zubelzu and Bennis2023; Kaynak et al., Reference Kaynak, Mermer, Sermet and Demir2025). Cloud-based data pipelines ensure temporal consistency and spatial alignment across distributed monitoring networks, whereas space–time synchronization algorithms maintain real-time coherence between the physical and digital domains (Placidi et al., Reference Placidi, Gasperini, Grassi, Cecconi and Scorzoni2020). Robust integration tools manage raw input cleaning, standardization, fusion across heterogeneous data types and real-time anomaly detection to ensure data reliability and model validity (Demir et al., Reference Demir, Conover, Krajewski, Seo, Goska, He, McEniry, Graves and Petersen2015; dos et al., Reference dos Reis, Weerts, Ramos, Wetterhall and dos Santos Fernandes2022; Ramirez et al., Reference Ramirez, Sermet, Shahid and Demir2024).
Visualization and communication: Effective visualization bridges technical analysis with decision-making and public understanding. Hydrological DTs employ interactive dashboards, GIS-based mapping tools, and 3D/4D visual analytics environments that allow users to explore scenarios, monitor performance and simulate management strategies (Sermet and Demir, Reference Sermet and Demir2022; Sajja et al., Reference Sajja, Mermer, Sermet and Demir2025b). Immersive visualization technologies such as AR, VR, and XR environments enable intuitive comprehension of flood propagation, aquifer recharge and watershed dynamics (Mudiyanselage et al., Reference Mudiyanselage, Gonzalez, Sermet and Demir2025). These visualization layers also underpin stakeholder engagement, community outreach and disaster preparedness, translating scientific complexity into accessible spatial narratives (Oyshi et al., Reference Oyshi, Maleska, Schanze, Bormann, Dachselt and Gumhold2022).
Modeling and analysis: The analytical core of DTs lies in their ability to simulate, predict and optimize hydrological processes. Hydrological DTs incorporate both physics-based and AI-driven models for process simulation, pattern detection, forecasting and uncertainty quantification. ML, DL and hybrid modeling frameworks simulate streamflow, flooding, water quality dynamics and groundwater fluctuations (Bayar et al., Reference Bayar, Demir and Engin2009; Kreuzer et al., Reference Kreuzer, Papapetrou and Zdravkovic2024) while supporting scenario-based optimization and early warning systems (Haq et al., Reference Haq, Lee, Rizzo and Perdrial2024). Multiobjective optimization, sensitivity analysis and “what-if” simulations strengthen the DT’s decision-support capacity. Advanced DTs employ generative and reinforcement learning models to enhance predictive precision, particularly under data scarcity or nonstationary conditions (Menzel et al., Reference Menzel, Thieken, Schwandt and Bürger2006; Keyes et al., Reference Keyes, McInnes, Woodward, Gropp, Myra, Pernice, Bell, Brown, Clo, Connors, Constantinescu, Estep, Evans, Farhat, Hakim, Hammond, Hansen, Hill, Isaac, Jiao, Jordan, Kaushik, Kaxiras, Koniges, Lee, Lott, Lu, Magerlein, Maxwell, McCourt, Mehl, Pawlowski, Randles, Reynolds, Rivière, Rüde, Scheibe, Shadid, Sheehan, Shephard, Siegel, Smith, Tang, Wilson and Wohlmuth2013; Ebrahimi et al., Reference Ebrahimi, Azizi, Khorram, Kashani and Ali2024).
Governance and operation: As DTs transition from localized projects to national or transboundary infrastructures, establishing robust governance, ethics and interoperability becomes fundamental to system success (Addor et al., Reference Addor, Do, Alvarez-Garreton, Coxon, Fowler and Mendoza2020). This pillar integrates data security and privacy protocols, utilizing cybersecurity measures to protect sensitive hydrological data and sensitive infrastructure information. To facilitate cross-sector coordination, the framework emphasizes standards and interoperability by adopting standardized data formats (e.g., Open Geospatial Consortium (OGC) WaterML), ensuring the DT can communicate with broader smart city networks (Song et al., Reference Song, Burns, Pandey and Roth2019; Jeong et al., Reference Jeong, Baek, Lim, Kim, Kim, Lee, Jung and Lee2022; Zhang, Reference Zhang2024). Operational integrity is maintained through regulatory and policy compliance, ensuring that all data handling adheres to regional water rights, quality assurance and control processes, environmental protection laws and government safety regulations, whereas stakeholder engagement and ethics frameworks manage transparency (Langenheim et al., Reference Langenheim, Sabri, Chen, Kesmanis, Felson, Mueller, Rajabifard and Zhang2022; Dihan et al., Reference MdS, Akash, Tasneem, SK, MdR, MdM, Badal, MdF, MdH, Abhi, Sarker and MdM2024; Ouedraogo et al., Reference Ouedraogo, Hawbani, Wang, Liu, Zhao, Al-qaness and Alsamhi2025).
Applications for DTs in hydrology
The core components defined earlier enable DTs to function as integrated platforms, successfully bridging the gap between theoretical frameworks and real-world execution. DTs in hydrology serve both operational (e.g., monitoring and control) and strategic (e.g., planning and resilience analysis) functions. For example, flood risk assessment can be conducted using dynamic DTs that incorporate river discharge data and rainfall forecasts to simulate inundation in real time (Seo et al., Reference Seo, Keem, Hammond, Demir and Krajewski2019; Manocha et al., Reference Manocha, Sood and Bhatia2024). Similarly, long-term water quality management can be enhanced through DTs that integrate data from drones, satellites and in situ sensors (Chen et al., Reference Chen, Wang, Wu, Wang, Huang and Wang2023; Hamzah et al., Reference Hamzah, Aqlan and Baidya2024). Hydrological DTs are now being used to simulate surface water flow through a river basin (Yang et al., Reference Yang, Xie, Fan, Xu, Melville, Liu and Hong2024), wastewater treatment (Wang et al., Reference Wang, Li, He, Tao, Wang, Yang, Savic, Daigger and Ren2024), drinking water networks (Conejos Fuertes et al., Reference Conejos Fuertes, Martínez Alzamora, Hervás Carot and Alonso Campos2020), urban drainage (Bartos and Kerkez, Reference Bartos and Kerkez2021) or stormwater infrastructure (Sharifi et al., Reference Sharifi, Beris, Javidi, Nouri, Lonbar and Ahmadi2024).
One of the most widespread applications of DTs is in flood prediction and emergency response. Open-source frameworks integrating hydrologic–hydrodynamic models are increasingly being developed to improve early warning capabilities and risk assessment in data-scarce regions (Rápalo et al., Reference Rápalo, Gomes and Mendiondo2024). Web-based DT systems using height-above-drainage metrics enhance flood inundation predictions (Li et al., Reference Li, Duque, Grout, Bates and Demir2023a). Furthermore, recent AI-enhanced DTs improve both the lead time and spatial resolution of disaster response systems by leveraging graph neural networks and DL-based satellite analysis for flood extent extraction within cloud computing environments (Roudbari et al., Reference Roudbari, Punekar, Patterson, Eicker and Poullis2024). Studies unraveling the power of neural networks for flood prediction across complex hydrological systems have further demonstrated their superior capability to map nonlinear rainfall–runoff relationships (Saberian et al., Reference Saberian, Zafarmomen, Panthi and Samadi2027). In urban infrastructure contexts, spatiotemporal Long Short-Term Memory (LSTM)-based surrogate of the U.S. Environmental Protection Agency (EPA)’s Storm Water Management Model (SWMM) allows for highly efficient node-level water depth and inflow prediction within complex urban drainage networks, replacing computationally heavy physical solvers with rapid AI approximations to support immediate decision-making (Zafarmomen et al., Reference Zafarmomen, Samadi and Borgomeo2026).
Beyond immediate disaster response, DTs play a pivotal role in watershed-scale modeling and planning. Recent DT-based watershed information modeling frameworks emphasize real-time data visualization, scenario analysis and multistakeholder collaboration for water allocation, hydrological cycle simulation and sustainable planning (Qiu et al., Reference Qiu, Duan, Xie, Ding and Jiao2022; Liu et al., Reference Liu, Guan, Tian, Cao, Chen and Shi2024). Large-scale cyberinfrastructure systems demonstrate how DTs can enable multiagency data sharing and joint decision-making across river basins such as the Upper Mississippi River (Mount et al., Reference Mount, Sermet, Jones, Schilling, Gassman, Weber, Krajewski and Demir2024).
In addition, DTs are revolutionizing aquatic ecosystem monitoring and water quality management. Drone-based DT systems now deliver high-resolution spatial and temporal data for nutrient tracking and pollution mapping, proving especially valuable in inaccessible regions (Hamzah et al., Reference Hamzah, Aqlan and Baidya2024). Urban DTs are being developed for hydrogeological risk assessment by integrating environmental monitoring with smart urban infrastructure such as adaptive lighting or traffic systems (Kušić et al., Reference Kušić, Schumann and Ivanjko2023; Barrile et al., Reference Barrile, Genovese, Maesano, Calluso and Manti2025).
DTs also contribute to infrastructure risk modeling and logistics during floods, integrating hydrological modeling with transportation analytics to optimize road accessibility, emergency facility placement and evacuation routes under flood conditions (Alabbad et al., Reference Alabbad, Mount, Campbell and Demir2024; Mitropoulos et al., Reference Mitropoulos, Karolemeas and Tsigdinos2025). These systems enhance real-time response strategies and infrastructure design for climate-resilient urban planning and management (Riaz et al., Reference Riaz, McAfee and Gharbia2023).
Another rapidly growing frontier is public engagement and education through DTs. Immersive simulations and serious games simplify complex hydrological phenomena for nonexpert audiences, facilitating better understanding and trust in water management decisions (Sermet et al., Reference Sermet, Demir and Muste2020b; Yin et al., Reference Yin, Hu, Liu, Liu, He, Zhu and Kornejady2024; Demiray et al., Reference Demiray, Sermet, Yildirim and Demir2025). Virtual labs and interactive 3D environments allow users to visualize flood wave dynamics, explore infrastructure behavior and test mitigation strategies (Dunlop et al., Reference Dunlop, Chandramouli, Moreland and Zhou2024).
The integration of conversational AI into DT platforms represents the next stage of operational hydrology. Natural language assistants linked with multimodal DT data sources enhance user accessibility, allowing managers and policymakers to query systems intuitively for training and operational purposes (Pursnani et al., Reference Pursnani, Sermet and Demir2025). Immersive and AI-enhanced DT frameworks are now supporting sustainable water reserve management and participatory platforms that address nutrient pollution and water quality (Shrestha et al., Reference Shrestha, Mount, Vald, Sermet, Samuel, Bryant, Peralta, Beck, Meyers, Muller-Karger, Cwiertny and Demir2025; Zhao et al., Reference Zhao, Song, Yu, Xing, Xu, Li and Wang2025).
At the cutting edge, federated DT systems link multiple environmental and infrastructural domains into unified decision-support ecosystems. Large-scale prototypes, such as NASA’s earth system digital twin (ESDT), demonstrate how AI, immersive visualization and multisource data integration can synchronize across global river basins, supporting comprehensive water governance and climate adaptation strategies (Le Moigne, Reference Le Moigne2025).
In summary, hydrological DTs have evolved from basic digital monitoring tools into powerful decision-support and participatory platforms. By integrating AI, immersive visualization and federated computing, DTs enable real-time insight, predictive foresight and collaborative governance of complex hydrological systems.
Technological evolution of DT
The general concept of a DT is well defined in many studies (Sjarov et al., Reference Sjarov, Lechler, Fuchs, Brossog, Selmaier, Faltus, Donhauser and Franke2020; Botín-Sanabria et al., Reference Botin-Sanabria, Mihaita, Peimbert-Garcia, Ramirez-Moreno, Ramirez-Mendoza and JdJ2022). However, aspects related to the specific technology and implementation of the DT systems are often unclear (Schroeder et al., Reference Schroeder, Steinmetz, Rodrigues, Henriques, Rettberg and Pereira2021). More recently, various DT projects related to different aspects of hydrology are modeled on city scale or larger areas (Ghaith et al., Reference Ghaith, Yosri, El-Dakhakhni, Walbridge, Nik-Bakht, Tsun Wai Ng, Shome, Shahria Alam, Damatty and Lovegrove2021; Henriksen et al., Reference Henriksen, Schneider, Koch, Ondracek, Troldborg, Seidenfaden, Kragh, Bøgh and Stisen2023; Iliuţă Reference Iliuţă, Moisescu, Pop, Ionita, Caramihai and Mitulescuet al., 2024; Lehner et al., Reference Lehner, Kordasch, Glatz and Agugiaro2023). Autonomous, federated and intelligent technologies for many types of DTs are required to optimize the real world in a large-scale DT (city or larger area scale). However, most of the existing studies focus on abstract concepts about these technologies among many DTs. It is necessary to describe the essential technologies clearly rather than abstract descriptions.
As challenges in the hydrological domain grow, a conventional DT model cannot offer a fully comprehensive solution. In this article, a conventional DT refers to an early-stage implementation that passively mirrors real-time sensor data on GIS dashboards through a unidirectional data flow, without AI-enabled predictive analytics or autonomous decision-making capabilities. Therefore, DTs in hydrology are not static products but evolving systems shaped by advancements in a range of digital technologies. To provide a structured roadmap for development and assessment, we propose a five-level technological evolution model for hydrological DTs, as illustrated in Figure 2. The functional definition of each stage is as follows:
-
Level 1: Physical DTs: representing hydrological events, phenomena or entity with physical prototypes.
-
Level 2: Web and GIS-powered DTs: modeling hydrological concepts or entities with 2D web-based platforms.
-
Level 3: Immersive and 3D DTs: representing hydrological concepts or entities with 3D interactive environments.
-
Level 4: Intelligent and AI-assisted DTs: enhancing hydrological concepts or models through advanced AI-driven predictive analytics.
-
Level 5: Autonomous and interconnected DTs: autonomously recognizing and solving complex problems in interconnected DTs and optimizing hydrological concepts or models according to the federated DT solution.
Figure 2.A two-axis technological evolution model for DTs in hydrology. This graph serves as a conceptual representation that maps the field’s progression across two simultaneous trajectories: an increase in intelligence and autonomy (vertical axis, evolving from static/passive models to fully autonomous agentic AI) and an expansion in system integration and federation (horizontal axis, evolving from isolated/single-domain applications to cross-domain global ecosystems).
To fully understand this evolutionary framework, it is essential to ground these levels in the foundational taxonomy proposed by Kritzinger et al. (Reference Kritzinger, Karner, Traar, Henjes and Sihn2018), which categorizes digital replicas based on the directionality of data flow. According to this taxonomy, a digital model features manual data flow in both directions; a digital shadow features automated one-way data flow from the physical object to the digital object; and a true DT requires fully automated, bidirectional data flow between the physical and digital domains. Our proposed five-level framework builds upon this taxonomy by mapping how this data-flow connectivity has been historically achieved and expanded in hydrology.
It is important to note that this five-level framework does not represent a single, mutually exclusive ordinal scale of autonomy. Instead, it maps the chronological integration of breakthrough technologies over the past three decades (1995–2025). As these technologies mature, they build upon one another. For instance, a modern operational system can simultaneously feature immersive visualization (Level 3) and AI-driven predictive intelligence (Level 4). Our model traces how distinct technological milestones progressively transitioned the field from manual digital models (Level 1) and early digital shadows (Levels 2 and 3) into intelligent, fully automated DTs (Levels 4 and 5).
The late 1990s and early 2000s marked the foundational era of stand-alone physical prototypes and static computational modeling, functioning strictly as Level 1. Although the early 2000s laid the conceptual groundwork for web-based spatial data, the true transition to Level 2 fully materialized in the late 2000s and early 2010s. During this period, the widespread deployment of IoT sensor networks, wireless telemetry and cloud computing enabled systems to incorporate automated one-way data feeds, transitioning from static maps to dynamic monitoring platforms.
The mid-2010s witnessed a new leap with the advent of immersive and XR technologies, symbolized by breakthroughs such as Oculus Rift and similar VR/AR platforms. This era corresponds to Level 3, introducing interactive and immersive environments that transformed the visualization and communication of real-time spatial data without fundamentally changing the system’s level of autonomy. This integration allowed scientists, policymakers and citizens to simulate hydrological scenarios, visualize flood dynamics and engage collaboratively in virtual decision environments.
The early 2020s were defined by the AI revolution, driven by the rise of DL, generative models, and large language models such as GPT-based systems. In hydrology, these technologies catalyzed the emergence of Level 4, capable of forecasting floods, identifying anomalies, optimizing infrastructure operations and supporting complex decision-making. Here, systems achieved true DT status by leveraging bidirectional data flows, ML and generative models for predictive analytics, anomaly detection and continuous adaptive learning.
Most recently, the mid-2020s have initiated the conceptual shift toward Level 5, characterized by agentic AI, edge computing and federated learning that operate as self-organizing ecosystems capable of horizontal cross-domain coordination without central human supervision. Such systems integrate hydrological, meteorological and ecological domains; self-update based on continuous feedback; and negotiate trade-offs autonomously in real time. Emerging paradigms in agentic AI, where AI systems act as collaborative agents capable of reasoning, planning and communication, signal the next stage in hydrological DT development: a shift toward self-governing, ethically guided and interoperable environmental intelligence networks.
Together, these technological milestones, spanning from the web revolution to immersive computing, AI and agentic autonomy, form the chronological backbone of our five-level DT evolution framework. Each level represents not merely an incremental improvement in modeling sophistication but also a fundamental paradigm shift in how hydrological systems are conceptualized, simulated and managed. In the following sections, we detail each level’s defining characteristics, technological enablers and implementation challenges, outlining a roadmap for the continued evolution of hydrological DTs.
Level 1: Physical DTs
At the initial level of this evolution, representing the foundational period of the late 1990s and early 2000s, hydrological systems were primarily represented through physical prototypes or static digital models. These early representations provided the fundamental understanding of system dynamics but lacked real-time connectivity and adaptive synchronization that define modern DTs. During this era, physical scale models of river basins, flumes and floodplains were widely used to study water movement, sediment transport and morphological changes (Peakall et al., Reference Peakall, Ashworth and Best1996; Daniel et al., Reference Daniel, Camp, LeBoeuf, Penrod, Dobbins and Abkowitz2011; Green, Reference Green, Cook, Clarke and Nield2014).
Simultaneously, this period marked the consolidation of static computational modeling where hydrologists relied on physically based distributed codes, such as MIKE Système Hydrologique Européen (MIKE SHE) (Refsgaard and Storm, Reference Refsgaard and Storm1995), and conceptual rainfall–runoff framework (Todini, Reference Todini1996) to simulate catchment behavior offline. These tools were instrumental for analyzing hydrological parameters, including watershed boundaries and potential flood risks under fixed conditions (Li et al., Reference Li, Unger, Sudicky, Kassenaar, Wexler and Shikaze2008; Papanicolaou et al., Reference Papanicolaou, Elhakeem, Krallis, Prakash and Edinger2008).
The integration of early GIS facilitated the spatial representation of these models, allowing for the mapping of environmental variables and land-use data (Tim, Reference Tim1995; Maidment, Reference Maidment2002). Although these systems enabled substantial advances in catchment analysis and flood forecasting (De Roo et al., Reference De Roo, Wesseling and Van Deursen2000), they remained fundamentally static. NASA’s early work with virtual models in aerospace hinted at the future potential of DTs (Glaessgen and Stargel, Reference Glaessgen and Stargel2012), but in hydrology, models were disconnected from real-time systems. They relied on manual data inputs, lacked interactivity and could not dynamically reflect environmental changes, limiting their ability to support adaptive water management or immediate disaster response.
Despite these operational limitations, 3D physical prototypes have proven highly effective in STEM (Science, Technology, Engineering, and Mathematics) education and public outreach. Physical models, such as tabletop floodplain simulation systems, are visually engaging and ideal for demonstrating hydrological processes in a hands-on manner. For instance, models like WARD’s Stormwater Floodplain Simulation System have been used in classrooms and science fairs to highlight floodplain dynamics, land-use impacts and environmental stewardship. Their accessibility makes them valuable for outreach programs, where direct observation offers meaningful learning experiences.
Although valuable for foundational studies and education, these static models were constrained by their disconnect from the living environment. This critical gap necessitated a shift toward the next stage of evolution: web-based digital models capable of integrating live data streams and providing dynamic visualization through remote platforms.
Level 2: Web and GIS-powered DTs
In the second level, with the advent of web-based technologies, remote sensing and cloud computing, hydrological models transitioned from static physical prototypes to digital platforms. Hydrological data became more accessible and dynamically visualized through 2D web platforms (Demir and Krajewski, Reference Demir and Krajewski2013). GIS played a crucial role in this transformation by enabling large-scale mapping and data integration and facilitates spatial representation of hydrological elements, such as watersheds, river networks and flood zones. Choi et al. (Reference Choi, Engel and Farnsworth2005) presented a web-based framework integrating GIS technology and hydrologic modeling to support real-time watershed management. Valjarević (Reference Valjarević2024) used GIS technique to track 60 years of climate-driven changes in Serbia’s river networks and basins. Cikmaz et al. (Reference Cikmaz, Yildirim and Demir2023) developed GIS-based flood susceptibility maps for identifying high-risk flood zones and supporting mitigation planning.
The availability of high-resolution satellite imagery and remote sensing technologies enhanced data accuracy by providing high-resolution elevation models and land cover classifications and allowed for more precise and dynamic water management solutions. Li et al. (Reference Li, Xiang, Demiray, Sit and Demir2023b) proposed MA-SARNet (Multi-scale Attention Synthetic-Aperture Radar Net), a one-shot DL framework that uses physical drivers and historical SAR imagery to predict high-resolution SAR (Synthetic-Aperture Radar) backscatter for flood nowcasting and mapping. Rogan and Chen (Reference Rogan and Chen2004) highlighted remote sensing advancements and their integration with GIS for monitoring land-cover and land-use changes to support planning and management. Bühler et al. (Reference Bühler, Marty and Ginzler2012) evaluated LiDAR (Light Detection and Ranging) and photogrammetry for high-alpine DEMs (Digital Elevation Model), finding optical methods accurate for moderate slopes and LiDAR better for steep terrain.
The rise of the Internet and cloud computing further enhanced DT capabilities by facilitating remote access to hydrological data. This level basically consists of three intermediate steps such as mirroring, monitoring and simulation. In mirroring (duplication of physical object into DT), the physical object simply replicated as digital object on the GIS-based mapping or web-based platform, and this step does not have additional processes such as simulation, testing and optimization. In monitoring step, physical entities can be monitored and controlled based on analysis of the DT, which is started to incorporate real-time data streams from remote sensors, IoT devices and high-resolution satellite imagery.
This step ensures that the virtual representation of hydrological systems remains accurate and reflective of real-time conditions. Real-time synchronizations are usually considered for DTs because DTs need to manage and synchronize various sensors to ensure consistency. Therefore, in this stage, the synchronization engine can manage many sensors and maintain consistency between the replicated twin world and the real world. IoT-enabled sensor technology plays a pivotal role in capturing real-time hydrological parameters such as river flow, rainfall, soil moisture and groundwater levels (Abdulbaqi and Hashim, Reference Abdulbaqi and Hashim2024; Kaynak et al., Reference Kaynak, Mermer, Sermet and Demir2025).
These sensors transmit data using high-speed data transmission technologies, including 5G networks, LPWAN and satellite communication, ensuring seamless and reliable data flow even from remote monitoring stations. Zanella et al. (Reference Zanella, Zubelzu and Bennis2023) explored how sensor networks can address challenges in hydrological monitoring and modeling, emphasizing the need for adaptive sensing technologies. HydroLang, a modular, web-based framework, enables real-time flood risk and mitigation assessment through interactive visualizations, geospatial analysis and customizable modeling tools, aiming to support decision-making, public engagement and community resilience across diverse flooding scenarios (Ramirez et al., Reference Ramirez, Sermet, Molkenthin and Demir2022; Sufi et al., Reference Sufi, Ramirez and Demir2025). In the final step, the physical entities can be optimized by executing various simulations of the DT.
Multiphysics simulations integrate multiple physical processes such as atmospheric dynamics, water prediction and hydraulic flow into a unified modeling framework (Keyes et al., Reference Keyes, McInnes, Woodward, Gropp, Myra, Pernice, Bell, Brown, Clo, Connors, Constantinescu, Estep, Evans, Farhat, Hakim, Hammond, Hansen, Hill, Isaac, Jiao, Jordan, Kaushik, Kaxiras, Koniges, Lee, Lott, Lu, Magerlein, Maxwell, McCourt, Mehl, Pawlowski, Randles, Reynolds, Rivière, Rüde, Scheibe, Shadid, Sheehan, Shephard, Siegel, Smith, Tang, Wilson and Wohlmuth2013). These simulations leverage high-performance computing (HPC) or distributed volunteer computing infrastructures to process large-scale, high-resolution hydrological models efficiently (Agliamzanov et al., Reference Agliamzanov, Sit and Demir2020). Scenario-based simulations also play a crucial role in evaluating various hydrological systems by running different “what-if” scenarios (Ebrahimi et al., Reference Ebrahimi, Azizi, Khorram, Kashani and Ali2024; Lobanova et al., Reference Lobanova, Liersch, Tàbara, Koch, Hattermann and Krysanova2017; Menzel et al., Reference Menzel, Thieken, Schwandt and Bürger2006). These simulations help policymakers and engineers test the effectiveness of flood control infrastructure, water quality and emergency response strategies under varying hydrological conditions and climate change.
This advancement enabled more accurate monitoring of hydrological variables such as precipitation, river flow and groundwater levels (Xu et al., Reference Xu, Muste and Demir2019a). Web-based models allowed several real-word applications such as flood monitoring systems, cloud-based watershed management platforms and decision-support frameworks, providing stakeholders with up-to-date information on river levels, precipitation patterns and watershed conditions (Qiu et al., Reference Qiu, Duan, Xie, Ding and Jiao2022).
However, despite the ability to visualize real-time data, these web-based systems remained largely passive, offering limited interaction between the digital and physical counterparts. This limitation paved the way for the next advancement, where immersive technologies such as VR and AR would transform these platforms into interactive, three-dimensional environments, enabling a more intuitive understanding of complex hydrological behaviors.
Level 3: Immersive and 3D DTs
With improvements in computational capabilities and new immersive technologies such as VR, AR or mixed reality (XR), DTs evolved into interactive systems capable of real-time data processing, simulation and visualization. In hydrology, this level introduced 3D simulations with AR/VR environments, which allow for high-fidelity modeling of river flows, groundwater movements and precipitation–runoff interactions (Sermet and Demir, Reference Sermet and Demir2022). These tools enabled the creation of highly interactive, three-dimensional models that allowed for a more intuitive understanding of hydrological behaviors under different scenarios (Chandramouli et al., Reference Chandramouli, Zhou, Moreland, Teegavarapu and Chandramouli2022). Some studies are related to decision support, disaster awareness and emergency response training (Sermet and Demir, Reference Sermet and Demir2022). These immersive platforms allow users to simulate and visualize extreme weather events and flooding in dynamic environments, thereby enhancing both technical preparedness and public communication. For example, researchers have used these tools to simulate flood projections, run flood models and demonstrate hydrometeorological extremes in virtual settings (Puertas et al., Reference Puertas, Hernández-Ibáñez, Cea, Regueiro-Picallo, Barneche-Naya and Varela-García2020; Rink et al., Reference Rink, Sen, Hannemann, Ködel, Nixdorf, Weber, Werban, Schrön, Kalbacher, Kolditz, Dutta, Feige, Rink and Zeckzer2021; Oyshi et al., Reference Oyshi, Maleska, Schanze, Bormann, Dachselt and Gumhold2022).
A recent and notable extension of this immersive paradigm includes the use of gaming engines and metaverse platforms to host DT environments (Xu et al., Reference Xu, Demir, Koylu and Muste2019b; Uchimiya, Reference Uchimiya2024). Game development technologies such as Unity and Unreal Engine have been leveraged to create engaging, physics-based virtual hydrological worlds, enabling both scientific exploration and stakeholder participation. In parallel, the rise of the metaverse has opened the door for persistent, multiuser DT spaces, where planners, engineers and citizens can collaboratively explore flood scenarios, codesign infrastructure and test policies in shared digital ecosystems (Sermet et al., Reference Sermet, Villanueva, Sit and Demir2020a,b; Xu et al., Reference Xu, Berres, Shao, Wang, New, Omitaomu and Raval2023). These developments have shown significant potential for participatory planning and education, particularly in urban water management contexts.
These interactive DTs allow hydrologists and policymakers to visualize flood scenarios, test mitigation strategies and optimize water resource management in a virtual and interactive environment before implementing changes in the real world. These advancements have proven particularly valuable in urban water management, where policymakers can visualize the impacts of various infrastructure projects before implementation. Despite these advancements, interoperability and scalability remained key challenges, as integrating different hydrological models and data sources seamlessly remains complex. In addition, immersive DT applications often require high computational power and specialized equipment such as headsets and motion controller, which may limit broader dissemination and accessibility, particularly in resource-constrained regions.
Although these immersive systems offered unparalleled user interaction and intuitive visualization for scenario testing, they still relied on predefined models and lacked inherent intelligence. Their inability to learn from data or provide predictive analytics on their own marked a clear boundary, setting the stage for the next major leap: the integration of AI to create intelligent DTs capable of forecasting, anomaly detection and continuous learning.
Level 4: Intelligent and AI-assisted DTs
In the fourth level, the intelligent DT represents a major advancement in hydrological modeling by integrating AI and advanced ML algorithms. These DTs enable predictive analytics, allowing for more accurate forecasting of hydrological variables such as precipitation, river flow, groundwater levels and flood risk assessment (Kreuzer et al., Reference Kreuzer, Papapetrou and Zdravkovic2024; Miau and Hung, Reference Miau and Hung2020; Sajja et al., Reference Sajja, Xiong, Mermer, Sermet and Demir2025a). At the core of this layer is AI-driven analytics, which employs sophisticated DL and ML algorithms to detect, learn and predict complex hydrological patterns (Lv and Xie, Reference Lv and Xie2024; Abdulameer et al., Reference Abdulameer, Al-Khafaji, Al-Awadi, Al Maimuri, Al-Shammari and Al-Dujaili2025).
These predictive tools support a range of operational goals, including streamflow prediction, disaster preparedness, water quality forecasting and water allocation optimization (Kumar et al., Reference Kumar, Agarwal, Deshpande, Cooper, Khosravi, Rathnayake, Hoshino, Kantamaneni and Rathnayake2024; Sanikhani et al., Reference Sanikhani, Nikpour and Jamshidi2025). In parallel, anomaly detection plays a crucial role in identifying irregularities in hydrological data, such as unexpected water level fluctuations, sensor malfunctions or water quality dynamics (Leigh et al., Reference Leigh, Alsibai, Hyndman, Kandanaarachchi, King, McGree, Neelamraju, Strauss, Talagala, Turner, Mengersen and Peterson2019; Haq et al., Reference Haq, Lee, Rizzo and Perdrial2024).
A key development within this level is the rise of AI agents, capable of simulating unseen hydrological conditions, filling gaps in sparse datasets and generating synthetic training data to improve model robustness (Foroumandi et al., Reference Foroumandi, Moradkhani, Sanchez-Vila, Singha, Castelletti and Destouni2023; Kadiyala et al., Reference Kadiyala, Mermer, Samuel, Sermet and Demir2024). These agents contribute to enhanced situational awareness and support more informed and timely interventions. They also reduce reliance on expert-driven calibration, allowing predictive systems to adapt more readily to emerging data (Kizilkaya et al., Reference Kizilkaya, Sajja, Sermet and Demir2025). The integration of AI agents further enhances situational awareness, reduces human intervention and enables more efficient resource allocation.
In recent years, foundation models, large-scale pretrained AI models such as transformers, have begun to influence the development of intelligent DTs. These models, trained on broad multimodal datasets, can be fine-tuned for hydrological applications, including rainfall estimation, soil moisture prediction or even semantic interpretation of environmental reports (Syed et al., Reference Syed, Khan, Jan, Albouq, Alqahtany and Naqash2024). As these interactive systems evolve, researchers are actively evaluating whether large language models can effectively reason for adverse weather conditions, ensuring that these agents can accurately process meteorological warnings, interpret complex risk factors and provide reliable, context-aware logic during extreme hydrometeorological events (Zafarmomen and Samadi, Reference Zafarmomen and Samadi2025). Their adaptability makes them well-suited for tasks that require transfer learning across basins, regions or variable data types.
Moreover, generative AI techniques are increasingly used to simulate hydrological scenarios in data-scarce environments. Generative adversarial networks, variational autoencoders and diffusion models have been applied to generate synthetic hydrographs, climate inputs and land-use scenarios, thus addressing data sparsity, uncertainty and extreme-event modeling challenges (Ji et al., Reference Ji, Mirzaei, Lai, Dehghani and Dehghani2024; Sun et al., Reference Sun, Xia, She, Ding, Jiang, Liu and Zhao2024). These capabilities enhance the resilience and adaptability of hydrological DTs, especially under nonstationary conditions driven by climate change and urbanization.
Although these intelligent DTs represented a major advancement in predictive analytics, they still largely functioned as sophisticated decision-support tools, requiring human oversight for high-stakes decisions and operating as siloed systems. This reliance on human intervention and the lack of cross-domain integration defined their primary limitation, paving the way for the ultimate level of maturity: autonomous and federated DTs capable of independent decision-making and seamless collaboration within a larger digital ecosystem.
Level 5: Autonomous and interconnected DTs
At the highest level of technological advancement, DTs operate autonomously and are interconnected across domains. Although Levels 1–4 primarily represent a vertical trajectory toward greater system intelligence and predictive capability, Level 5 represents the realization of the horizontal trajectory: seamless integration and federation across diverse environmental and infrastructural boundaries. These systems are capable of making independent decisions and integrating multiple DTs across hydrological, meteorological, ecological and infrastructural networks. Such advanced DTs go beyond reactive monitoring; they function as intelligent agents that manage, predict and optimize system behavior in real time using integrated data streams and machine intelligence.
In experimental models or limited, bounded infrastructure settings, these autonomous systems leverage real-time sensor data, DL algorithms and adaptive control mechanisms to dynamically manage hydrological processes. Although fully autonomous operation in natural environments remains conceptual, AI-powered self-learning capabilities are being developed to eventually allow DTs to autonomously adjust flood mitigation strategies, simulate rainfall scenarios and optimize irrigation schedules with minimal human intervention (Redel-Macías et al., Reference Redel-Macías, Marín-Moreno, Bajo, Nichols, Pérez, Bellido, Taguas and Daniela2021; Yoon and Ahn, Reference Yoon and Ahn2024). In many implementations, intelligent DTs actively synthesize simulations, real-time data and policy constraints to generate actionable recommendations for water management, flood response, infrastructure planning and investment decisions (Nie and Liu, Reference Nie and Liu2025).
However, it is important to note that fully autonomous, operational Level 5 systems within complex, natural hydrological environments remain largely conceptual. Currently, the most advanced practical implementations of these interconnected systems are found in urban smart-city frameworks and artificial infrastructure networks, where bounded parameters make autonomous control more feasible (Dembski et al., Reference Dembski, Wössner, Letzgus, Ruddat and Yamu2020; Ivanov et al., Reference Ivanov, Nikolskaya, Radchenko, Sokolinsky and Zymbler2020; Lehtola et al., Villani et al., Reference Lehtola, Koeva, Oude Elberink, Raposo, Virtanen, Vahdatikhaki and Borsci2022; Reference Villani, Gugliermetti, Barucco and Cinquepalmi2025). Translating these capabilities to negotiate trade-offs autonomously in large-scale, unbounded natural river basins remains a critical frontier for future research.
As the number and scale of DTs grow, the concept of Interconnected (Federated) DTs emerges as a critical architectural innovation driving this horizontal trajectory. These systems enable cooperation between multiple domain-specific DTs, for example, river basin models, urban flood control systems and climate forecast engines, within an integrated digital ecosystem. Federated DTs allow for multiagency collaboration, connecting stakeholders such as government bodies, utilities, researchers and non-governmental organizations (NGOs) into a shared information and decision-making space. Large-scale prototypes are pioneering this federated approach. NASA’s ESDT integrates AI, immersive visualization and multisource data to enable synchronized monitoring of global river basins (Reference Huang, David, Oadia, Roberts, Kumar, Stackhouse, Borges, Baillarin, Blanchet and KettigHuang et al., 2022; Le Moigne, Reference Le Moigne2025). Alongside NASA’s system, the European Union’s Destination Earth initiative stands as one of the most comprehensive operational efforts to date, demonstrating how federated multisource data integration can synchronize across global domains to support comprehensive climate adaptation and policymaking (EC, 2025; Doblas-Reyes et al., Reference Doblas-Reyes, Kontkanen, Sandu, Acosta, Al Turjmam, Alsina-Ferrer, Andres-Martinez, Anerdi, Arriola, Axness, Batlle Martin, Bauer, Becker, Beltan, Beyer, Bockelmann, Bretonniere, Cabaniols, Caprioli, Castrillo, Chandrasekar, Cheedela, Correal, Danovaro, Davini, Enkovaara, Frauen, Fruh, Gaya Avila, Ghinassi, Ghosh, Ghosh, Gonzalez, Grayson, Griffith, Hadade, Haine, Hartick, Haus, Hearne, Jarvinen, Jimenez, John, Juchem, Jung, Kegel, Kelbling, Keller, Kinoshita, Kiszler, Klocke, Kluft, Koldunov, Kolling, Kolstela, Kornblueh, Kosukhin, Lacima-Nadolnik, Leal Rojas, Lehtiranta, Lunttila, Luoma, Manninen, Medvedev, Milinski, Mohammed, Muller, Naryanappa, Nazarova, Niemela, Niraula, Nortamo, Nummelin, Nurisso, Ortega, Paronuzzi, Pedruzo-Bagazgoitia, Pelletier, Pena, Polade, Pradhan, Quintanilla, Quintino, Rackow, Raisanen, Rajput, Redler, Reuter, Rocha, Roura-Adserias, Ruppert, Sayed, Schnur, Sharma, Sidorenko, Sievi-Korte, Soret, Steger, Stevens, Streffing, Sunny, Tenorio, Thober, Tigerstedt, Tinto, Tonttila, Tuomenvirta, Tuppi, Van Thielen, Vitali, von Hardenberg, Wagner, Wedi, Wehner, Willner, Yepes-Arbos, Ziemen and Zimmermann2026).
The implementation of autonomous and federated DTs is further enhanced by edge intelligence, which brings computational power closer to data sources such as field sensors, unmanned aerial vehicles and smart infrastructure. This reduces latency and enables localized decision-making for time-critical applications like flood control operations or water resource management (Chen et al., Reference Chen, Jiang, Zhou, Lv, Liang and Wan2022). Moreover, federated learning plays a vital role in enabling privacy-preserving AI model training across distributed DTs (Akbulut et al., Reference Akbulut, Cifci and Aslan2023). This approach addresses concerns around data ownership, confidentiality and regulatory compliance while enabling cross-regional model generalization, allowing a regional flood model to benefit from patterns learned in adjacent basins without direct data transfer (Farooq et al., Reference Farooq, Tehseen, Qureshi, Omer, Yaqoob, Tanweer and Atal2023).
Despite their transformative potential, autonomous and federated DTs introduce significant challenges. Key concerns include data governance, model transparency, interoperability and computational scalability. Implementing such systems requires robust legal frameworks, interoperable data standards, strong cybersecurity measures and ethical oversight mechanisms to ensure their safe, equitable and accountable use in environmental decision-making.
Although Level 5 represents the current conceptual horizon, DT technologies will likely continue evolving. Future hydrological DTs may incorporate neuromorphic AI, self-evolving model architecture and quantum simulation, as demonstrated by recent advancements such as HydroQuantum, a quantum-driven Python package that offers new computational paradigms for complex hydrological simulations (Saberian et al., Reference Saberian, Zafarmomen, Neupane, Panthi and Samadi2025). The trajectory will not be linear but iterative, guided by regulatory shifts, ecological priorities and the emergence of global water governance challenges.
Key considerations and recommendations
The evolution of DT technologies in hydrology and environmental science presents a complex interplay of opportunities and constraints. Effective implementation requires balancing several interdependent factors, data availability, computational infrastructure, technical expertise and sustainable maintenance. Each of these dimensions influences the scalability, reliability and longevity of DT systems. The following considerations outline these key aspects and provide targeted recommendations for advancing hydrological DT design and deployment.
Data availability and interoperability: Reliable, high-quality and continuously updated data form the foundation of every hydrological DT. DTs depend on diverse sources, including in situ sensors, satellites, IoT networks and historical databases. However, data heterogeneity, inconsistent standards and limited coverage in many regions undermine system reliability and scalability. Datasets are often fragmented across institutions, subject to proprietary restrictions and poorly documented, leading to integration challenges and reduced confidence in model outputs. Addressing these issues requires coordinated efforts to improve data infrastructure and governance.
The adoption of open data standards that follow findable, accessible, interoperable and reusable principles is essential for ensuring compatibility across platforms. Establishing federated data-sharing frameworks can facilitate collaboration among institutions while maintaining control over sensitive datasets. Expanding sensor networks, especially in data-scarce regions through IoT or crowdsourced citizen sensors, will enhance spatial and temporal resolution, whereas automated data validation and anomaly detection algorithms can safeguard data quality. Policy mechanisms should support transparent data sharing while balancing security and privacy, thereby building trust among agencies and stakeholders. A well-structured, interoperable data ecosystem is critical to the reliability and adaptability of hydrological DTs.
Computational infrastructure and scalability: The computational demands of DTs increase rapidly as systems evolve from static visualization platforms to intelligent, federated networks. High-resolution hydrological simulations, real-time analytics and AI-driven forecasting require significant processing power, data throughput and storage capacity. Reliance on centralized computing environments may lead to bottlenecks, latency issues and high energy consumption. Scalable architectures that distribute computational workloads are essential for maintaining real-time responsiveness. A hybrid approach that combines cloud computing for large-scale simulations with edge intelligence for local processing near data sources can significantly enhance efficiency. HPC should be leveraged for multiphysics modeling and data assimilation, whereas modular, open-source software frameworks can reduce dependence on proprietary systems. Investments in energy-efficient infrastructure and sustainable computing practices will also minimize the environmental footprint of large-scale DT deployments. Shared regional or national DT infrastructure hubs can provide equitable access to computational resources, ensuring that smaller institutions and developing regions can participate in and benefit from DT technology.
Technical expertise and capacity building: The successful implementation of hydrological DTs depends not only on technology but also on human expertise. Developing and maintaining DTs require interdisciplinary collaboration among hydrologists, data scientists, engineers, software developers and policy experts. Many organizations, however, lack the integrated skill sets needed to design, operate and interpret DT systems effectively. Bridging this gap requires sustained investment in training and capacity development. Universities and research institutions should establish cross-disciplinary curricula that combine environmental modeling, data analytics and AI methodologies. Open-access training materials, workshops and online platforms can expand participation and lower entry barriers. International collaborations and joint test beds can accelerate learning and promote knowledge transfer between institutions. Capacity-building initiatives should also prioritize inclusiveness by supporting participation from regions with limited technical infrastructure. Long-term sustainability will depend on cultivating communities of practice that continuously exchange expertise and best practices across hydrological DT networks.
Cost, maintenance and sustainability: DT projects often face challenges in sustaining operations beyond their initial development phase. High upfront costs for sensors, computing infrastructure and software deployment are compounded by recurring expenses for data acquisition, model updating and maintenance. Without clear financial planning, many DT systems risk becoming short-lived pilot projects. Achieving financial and operational sustainability requires both modular design and collaborative funding strategies. Modular architecture allows incremental scaling and targeted upgrades rather than large, one-time investments. Cost-sharing mechanisms among governmental agencies, research organizations and private stakeholders can distribute maintenance responsibilities more equitably. The use of open-source software and shared digital infrastructure can reduce licensing costs and promote transparency. Long-term funding programs should be established to support data infrastructure upkeep, hardware replacement and system modernization. Sustainability assessments must also evaluate the environmental and economic impacts of DT operation, promoting efficient use of computational and material resources. Building DT as living, adaptable system rather than static project ensures that they remain valuable assets for long-term water governance and climate resilience.
Integrated perspective and strategic planning: The interplay among data quality, computational power, expertise and financial resources defines the feasibility and maturity of hydrological DTs. As systems progress from simple static representations to intelligent, federated and autonomous ecosystems, their benefits grow, but so do the associated costs and technical demands. Table 1 summarizes this continuum, highlighting how data, computation, expertise and cost evolve across the five technological evolution levels. The resource intensity is highly variable and depends heavily on the specific scale and scope of the DT’s development and deployment rather than just its technological level. For sustainable implementation, institutions must align their DT ambitions with available infrastructure, workforce and funding capacity. Strategic planning that integrates these considerations from the outset will ensure that DT development remains both achievable and enduring.
A comprehensive comparison of various factors affecting the success of digital twins across all five technological evolution levels
Table 1. Long description
Table summarizes this continuum, highlighting how data, computation, visualization, expertise and cost evolve across the five technological evolution levels (Physical DT, Web and GIS DT, Immersive DT, Intelligent DT and Autonomous DT).
The long-term success of hydrological DTs depends on viewing them as evolving sociotechnical ecosystems rather than stand-alone software projects. Open standards, modular architectures and sustained collaboration across institutions will be vital for creating robust, adaptive and trustworthy DT frameworks. By integrating these considerations such as data, computation, expertise and sustainability, hydrological DTs can mature into resilient platforms that enable climate-informed water management, disaster mitigation and inclusive environmental governance.
Conclusion and future directions
DT technologies are poised to revolutionize the field of hydrology by enabling real-time, data-driven and collaborative management of water systems. By synthesizing technological milestones into a structured five-level model, this article provides both a roadmap and a research agenda for hydrological DT evolution, from basic data visualization and simulation to fully autonomous system coordination. The integration of web systems, GIS, IoT sensors, cloud computing, AI and immersive technologies have transformed DTs into sophisticated, interactive and predictive platforms. These platforms are now central to diverse hydrological applications, including flood risk assessment, watershed monitoring, water quality forecasting and infrastructure resilience planning.
The development of DTs in hydrology follows a clear and accelerating evolutionary trajectory. Beginning with static physical and conceptual models, DTs have progressively incorporated real-time sensing, immersive environments and AI-driven intelligence, culminating in the vision of federated and autonomous systems. This progression reflects key technological milestones, such as the rise of web-based systems, the adoption of AR/VR tools for visualization and training and the growing reliability and availability of ML for predictive modeling. By organizing these developments within a structured implementation and maturity framework, this manuscript offers a flexible conceptual guide for navigating the current state and envisioning the future potential of hydrological DTs.
Looking ahead, several transformative directions will define the next generation of DTs in hydrology. The integration of conversational AI and natural language interfaces is expected to make DT systems significantly more accessible to nonexpert users, enabling broader participation in environmental decision-making. AI agents powered by large-scale foundation models and generative techniques will enhance forecasting accuracy, generate synthetic data for data-scarce regions and simulate unobserved hydrological scenarios, greatly expanding the predictive capacity and robustness of DTs. Quantum computing, while still emerging, promises to revolutionize computational hydrology by enabling real-time, high-resolution and multiphysics simulations that are beyond the reach of current technologies.
At a systems level, federated DT networks, such as NASA’s ESDT, will play a crucial role in enabling cross-domain coordination and global-scale water governance. These platforms will combine Earth observation data, cloud infrastructures, immersive visualization and AI analytics to monitor and manage water systems across political and ecological boundaries. As user interfaces evolve with AR/XR technologies, citizen engagement will become more immersive and dynamic, fostering participatory science, real-time education and hyperlocal policy feedback loops. Emerging technologies such as edge intelligence and 6G (6th Generation) mobile networks will further reduce latency and enhance system responsiveness especially critical for applications like disaster risk reduction and emergency response.
Beyond technical advancements, the convergence of DTs with global digital public goods, such as open-source platforms, shared ontologies and open-access hydrological datasets, offers a pathway toward democratizing access to environmental intelligence. International collaboration will be essential for addressing shared water challenges, including transboundary river basin governance, global flood monitoring and long-term climate adaptation planning. By embracing principles of open science, ethical and transparent AI and inclusive design, the hydrological community can ensure that DT technologies evolve not only as powerful technical solutions but also as enablers of sustainability, equity and resilience in a rapidly changing world.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/wat.2026.10035.
Data availability statement
The original contributions presented in the study are included in the article/supplementary material, and further inquiries can be directed to the corresponding author.
Acknowledgements
The authors thank the Hydroinformatics Lab members for their contributions to the article. The authors used Gemini to assist with improving the readability, clarity and grammatical accuracy of the manuscript. All conceptual content and interpretations were developed entirely by the authors. The AI tool was not used to generate novel scientific ideas, results or conclusions.
Author contributions
Conceptualization: O.M. and I.D.; methodology: O.M.; data curation: O.M.; visualization: O.M.; writing – original draft: O.M.; writing – review and editing: O.M. and I.D.; project administration, supervision: I.D.; all authors approved the final submitted draft.
Funding statement
This research did not receive any specific grants from public, commercial or nonprofit funding agencies.
Competing interests
The authors declare no conflicts of interest.
Open access
Comments
April 21, 2026
Dear Editor-in-Chief,
I am pleased to submit our manuscript, titled “The Evolution of Digital Twins in Hydrology and Environmental Science: From Physical Models to AI-Assisted Autonomous Systems,” for consideration as a review (position) article in Cambridge Prisms: Water.
This manuscript directly aligns with the journal’s mission to translate hydrological research into effective, implementation-ready solutions. We propose a novel five-level technological evolution framework for hydrological Digital Twins (DTs) that maps the field’s progression from static physical prototypes to fully autonomous and interconnected AI-driven systems. Given the increasing urgency of climate-driven flooding and water management challenges, our work provides a structured roadmap for developing the next generation of use-inspired hydroinformatics.
Crucially, our study moves beyond abstract conceptualization by introducing a layered implementation framework and a cross-level capability matrix. This matrix analyzes DT systems in terms of data requirements, computational demand, and cost, providing a practical decision-support tool for engineers and policymakers seeking to implement scalable disaster resilience strategies. We believe this work is particularly timely for your readership, following the recent publication of related high-impact work in your journal by my co-author, Dr. Ibrahim Demir, regarding flood risk communication.
As requested by the journal’s guidelines, we have included an Impact Statement that details the wider beneficial use of this research for stakeholders across the water sector.
This manuscript is an original work that has not been published elsewhere and is not currently under consideration by any other journal. All authors have approved the manuscript and agree with its submission to Cambridge Prisms: Water.
Thank you for your time and for the opportunity to share our work with the Cambridge Prisms community.
Sincerely,
Omer Mermer
On behalf of all co-authors:
Ibrahim Demir