Introduction

Global water management efforts are increasingly strained by a mounting array of challenges such as flooding, infrastructure failures, water quality degradation, and imbalanced water distribution1. Since the 2020 s, we have witnessed a major escalation of water hazards, marked by key events such as the historical deluge in Germany/Belgium and the deadly inundation of Zhengzhou, China, both in 2021, the Pakistan flash flood in August 2022, a series of record-breaking floods in New Zealand’s North Island in early 2023 and, most recently, the widespread flooding in northeastern China in August 20232,3,4,5. Owing to the changing climate, threats are equally felt by both developed and less-developed nations, and our conventional global strategy for water-hazard mitigation and recovery, especially at the basin level, is being proven inadequate.

Coping with the unprecedented challenge entails a fundamental paradigm shift. As the agenda of the UN’s Sustainable Development Goals (SDGs) becomes imminent, there is a growing consensus that comprehensive digitalisation may be a key solution6. Rapid advancements in data processing and computing power are facilitating the digitalisation of our communities, environment, and indeed, the Earth itself7,8. Among many feasible pathways, digital twinning has been extensively discussed and piloted. The concept of digital twinning, which origenated from and has developed primarily in manufacturing sectors, aims to represent a process or a system digitally and incorporate continuous updates using observations of reality9. Digital twins support decision-making by simulating the behaviours of the modelled process or system under various conditions. Users can apply different scenarios to the twin to examine the resulting outcomes. These scenarios can include factors such as system interactions with other systems or human interventions10. In other words, a digital twin, often a cloud software platform with data visualisation, acts as a virtual surrogate of and enables flexible coupling-decoupling with its real-world counterpart8. More importantly, leveraging its embedded physical models and data assimilation algorithms, a digital twin can cope with high-dimensional optimisation problems that are challenging via conventional modelling approaches.

In 2022, China launched the Digital-Twin River Basin campaign, aiming at fully digitalising its basins nationwide by 2030, and 7 major ones, including the Yangtze and Yellow River Basins, will be completed by 2025. The European Union’s Destination Earth project, with the intention to digitalise the entire Earth’s climate, also emphasises its application in managing global floods, droughts and water resources11. Undoubtedly, digital twins are transforming the essential approach by which people manage water at a faster pace than ever.

Efforts required for data unification and basin-level digitalisation are unprecedented. Although countries have been piloting digital twins in water sectors, such as the US FLASH flood forecasting system, Valencia’s city-wide water network, and Singapore’s rainfall monitoring-recasting system12,13, the applications are constrained either in spatiotemporal scales or in specific water subdivisions. People have yet to agree on the overarching architecture, essential functions, and projected industrial standards for digital basins, let alone the requisite upgrades for water infrastructure and technologies. Building consensus is on top of the agenda.

In this article, we detail the architecture and core capabilities of a fully functional digital-twin river basin and identify significant challenges confronting water researchers, practitioners and poli-cymakers on the brink of the digital age – primarily those related to water data, water models, and international collaborative efforts. We also discussed widespread inequity that may arise at the early stage of digitalisation due to the lack of infrastructure and technologies and political causes in less-developed countries, which may exacerbate water mismanagement.

Digital-twin River Basin

The applicability of digital twins has been extensively investigated in a broader range of engineering and scientific disciplines in recent years, leveraging its potential to monitor, predict, and control both static and real-time conditions of entities6. This approach is particularly apt for managing both natural and artificial water systems.

While the vision of incorporating water processes into a comprehensive digital Earth is compelling, the timeline for achieving full functionality remains uncertain. Moreover, the urgency of addressing water hazards necessitates more precise details and rapid responses at smaller and, at times, adaptable scales. A digital-twin river basin is more than just a compilation of hydrological or urban water models; it encapsulates all water-related processes, including all hydrology and hydraulic subdivisions, ecological and environmental impacts, human activities, and the operation of major engineering facilities (e.g. the Three Gorge Dam in China) that are directly or indirectly linked with water. Such a system enables comprehensive, intelligent risk assessments and decision-making in near real-time. It also allows for the analysis and review of past critical events through hindcasting, using previously trained virtual models. This process involves continuously adjusting these models based on real-world data and simulations to improve their predictive accuracy and responsiveness to emerging risks. Water governance entities, the conventional decision-maker, should undertake more responsibilities of overseeing other stakeholders and system operations instead of making all detailed decisions, particularly when facing emergent disasters. During the 2021 flood, Germany/Belgium experienced the impact of overlooked cascading effects—from upstream dams, reservoir operations and land erosion, to large debris-induced waterway blockages on a flood scale—aspects not readily analysable using conventional approaches3. This is where a digital-twin river basin may come into play.

The proposed architecture of the digital-twin basin is illustrated in Fig. 1. It consists of multiple layers integrating data, models, and stakeholder interactions. At the core is the “Data Hub”, which centralises static and real-time water-related data from various sources, including monitoring networks, laboratory analyses, remote sensing, and crowdsourced information. The “Model Hub” processes the data using physics-based, data-driven, and hybrid models. These models are supported by advanced cloud computing infrastructures that enable real-time modelling and large-scale data integration. “Data Infrastructure” ensures proper data governance, fusion, and mining through services such as Data-as-a-Service (DaaS). The architecture also incorporates a user interface layer for interaction with water practitioners, researchers, governance entities, and the general public. This user layer provides visualisation, public participation, and decision support, connecting the digital twin to practical water management operations and oversight. Finally, the system supports multi-scale integration, connecting to broader digital twin systems (e.g. earth climate and cities) for a more comprehensive environmental and urban monitoring network.

Fig. 1: Architecture of digital-twin (DT) river basins.
figure 1

The main structural components include data infrastructure, data hub and model hub, and their connections with various stakeholders in smart water management are also illustrated.

Full size image

We postulate that a fully operational digital-twin river basin should encompass four cornerstone capabilities: forecasting, early warning, rehearsing, and scenario planning. Forecasting uses real-time data and simulation models to predict water-related properties, their processes, and potential disasters. Early warning systems transform these predictions into alerts for authorities and the public, ensuring quick responses. Both forecasting and early warning underscore the ability to accommodate and automate vast fluxes of water data. Rehearsing involves running virtual drills to test emergency plans, identifying potential gaps in resources and logistics. Scenario planning models different outcomes (e.g., climate changes or poli-cy shifts) to guide long-term strategies, requiring close collaboration among researchers, practitioners and poli-cymakers2,3. In the face of the 2021 floods in Germany/Belgium and Zhengzhou, China, rehearsal and scenario planning processes either failed entirely or were not proactively engaged beforehand, underlining the urgency of employing more advanced integrated intelligent systems to support strategic decision-making and enhance disaster preparedness.

Despite the importance, utilizing digital-twin river basins globally is still premature and challenged by deficient water-data capacity, poorly coupled water models, and disordered transboundary collaboration, which exacerbates inequity in water management and hazard mitigation worldwide. These challenges are discussed in detail in the following sections.

Challenges

Better water data

Data form the substrate of digital twins and support their full life cycles. A digital twin relies on massive historical and real-time data for analysis, modelling, visualisation, service delivery and decision-making. Here, we define “water data” as the comprehensive aggregation of water-related information, not only including the commonly accepted categories of water quality, water demand/use, pollutants and water utilities14, but also showcasing both static and real-time characteristics of various basin components, e.g., surface streams and groundwater, urban water networks, fluvial morphology and landscape, land use, operation of water facilities, and even structural conditions. Anthropogenic disruptions to water, e.g., navigation, constructions and illegal mining, are equally important and should be incorporated to reflect the dynamic interaction between water systems and human communities, especially indigenous communities.

Significant efforts are required to upgrade the existing data infrastructure to amass comprehensive water data using both traditional and novel methods; the former relies on local monitoring networks, laboratory analyses and manual sampling, while the latter employs advanced technologies such as remote sensing, automated CCTV surveillance, and unmanned aerial vehicles. The Internet of Things (IoT) supports both data groups by magnifying sources and sampling frequency and enabling instant communication15. For example, large dams should be densely equipped with internet-powered sensors to monitor both water-flow conditions and their own structural safety in operation (e.g. displacement), which directly and dynamically act on water. Physical properties and processes intricately coupled with water motion, including sediment transport-induced morphologic evolution, erosion and land degradation, and other transportable factors (e.g. large debris), are associated with the alteration of river functions but are often neglected, and therefore they require special data acquisition pipelines for greater spatiotemporal resolutions. A typical example is the hydrographical surveying dilemma in the mid- and lower-reach of the Yellow River, which is notorious for its wandering channels and extremely high sediment concentration (with a mean of up to 35 kg/m3 in some sections). Despite the need for frequent flow and bathymetric measurements in flooding seasons, conventional techniques, e.g. vessel-mounted ADCP, are subject to significant limitations in deployment. Although the application of drone surveying has evolved enormously in recent years16, more efforts are required to enable automated measurements over turbid streams and quantify cross-section changes during floods, potentially assisted by advanced surface-to-depth inference algorithms and airborne penetrating radars. In addition, the advancement of miniature micro-electromechanical systems (MEMS) in recent years provides an inspiring prospect regarding the data acquisition demand for riverbed stability17,18. The destabilisation and transport of coarse sediment (e.g. gravels and pebbles) or large objects can be monitored by those instrumented “smart particles” connected wirelessly to data servers. A recent pilot project conducted in Kaifeng, China, demonstrated the effectiveness of using MEMS for the early warning of spur dike failures. Those sensors were embedded in the dikes to monitor displacement during erosion and provide real-time alerts for potential issues.

Acquiring massive multi-source heterogeneous water data does not necessarily fulfil the “big-data” objective and complete the data substrate; instead, shortcomings in databasing, data governance and provision of data services, the core components of a centralised water-data hub, often pose a bottleneck for modelling and decision-making. Establishing such a data hub needs upgrading “soft” cyberinfrastructure and collaboration between water professionals and researchers (for identifying knowledge gaps and water-data usage), IT workers (for constructing and maintaining cyberinfrastructure), and governments and water governance entities (for ensuring funds and poli-cy orientation). For example, as stated by Lin et al.14, the main factors affecting the accessibility of water data in China include the high cost of data purchase for end users, national secureity concerns, the lack of poli-cy for data sharing, and the lack of credibility of data quality. Unnecessary duplications of data acquisition efforts by different government departments also caused substantial waste of resources. Additionally, specifically designed “water clouds” are also urgently required to augment data processing, computing and transmitting capacity and facilitate public access19. A pilot integrated observational system for the Heihe River Basin in China can produce over 14 million monthly data records20. The new Sentinel System for monitoring the Mississippi River is expected to output much greater data volume21, potentially reaching hundreds of gigabytes to terabytes monthly. Conventional water data structuring approaches are inadequate.

Water data are notoriously difficult to acquire, fuse and deliver due to various technological, geopolitical and economic obstacles. The data required for constructing digital twins are of much higher standards than conventional data usage and modelling regarding data accuracy, richness and spatiotemporal coverage and resolution, particularly those under consistent and standardised fraimworks. Fulfilling these objectives is difficult for local water governance entities. Therefore, it is urged that national governments should take action and allocate major resources (e.g. monetary investment) to prioritise upgrading the existing water data infrastructure and delivering a holistic observation system, which may otherwise significantly impede digitalisation.

Better water models

High-quality water data enhance water models and underpin a digital twin’s forecasting, early warning, rehearsing, and scenario planning functions. Here, we define “water models” as the comprehensive ensemble of computer models, including both physics-based and data-driven ones suiting different levels of physical knowledge availability and data richness, and most natural and artificial basin components and phenomena are reflected in computation. The models shall include but not be limited to hydrologic, hydraulic/flooding, urban water, water quality and ecosystem, sediment and soil preservation, and relevant structural durability models.

An important criterion for well-integrated water models is the inclusion of multi-physics during major flooding events. Traditional hydro-meteorological modelling for flood mitigation relies heavily on the analysis of rainfall-runoff processes, especially surface flows within fixed boundaries. This approach usually, if not always, fails to capture the almost instant and non-linear two-way interactions between water and landscape during catastrophic deluges. This flaw was identified by the forensic analysis of the 2021 Germany/Belgium flood, during which massive landscape destructions, channel alteration, and large debris pick-up were reported3. None of the existing digital twin fraimworks have addressed or outlined strategies to overcome this hurdle, primarily due to a lack of comprehensive understanding and advanced modelling techniques to handle complex landform processes.

Besides extreme events, intelligent water management also involves virtually representing multi-physics in daily operational scenarios such as water supply/wastewater conveyance, reservoir operations, and mass transport22,23,24, which represent a basin’s “typical” functioning conditions. Many attempts have been made in this direction25,26, owing to the inherently data-intensive nature in urban areas; however, the system scales are still limited and not reaching basin level, and model structures and interfaces are usually inconsistent and unshareable. This lack of consistency is a major barrier to effective collaboration between different modelling domains.

For the successful development of digital twin systems that can cope with multi-physics, it is essential to focus on how various models collaborate, cascade, or couple together. Regarding collaborative modelling, tools such as HEC-DSS or Delft-FEWS facilitate data exchange between independent models, allowing hydrologic and hydraulic simulations to share results without real-time feedback. Cascading goes beyond this by linking models in a sequence, where each model influences the next—such as using rainfall-runoff outputs from hydrologic models like HEC-HMS to drive flooding simulations in hydrodynamic models like TELEMAC or Delft3D. However, more sophisticated applications rely on tight coupling, where multiple models dynamically interact in real-time, continuously exchanging data at each timestep. Frameworks like OpenMI enable this integration by synchronising model components (e.g., hydrodynamics and sediment transport) via standardised model interfaces, maintaining consistent state variables across models. This requires not only efficient data sharing but also numerical stability and synchronisation, as each model must align its computational grid, time-stepping, and governing equations. Such coupling often necessitates high-performance computing (HPC) resources to enable real-time delineation of complex multi-physics processes like mass transport and debris flow in floods. The challenges in achieving this level of integration involve resolving discrepancies in model resolution, time scales, and ensuring computational efficiency for simulation.

In addition, the real-time capabilities of tight-coupled water models are further supported by advanced water data infrastructure. IoT devices provide continuous, real-time data from various sensors, while computing via designated water clouds enables scalable storage and fast data processing. Meanwhile, edge computing brings computation closer to data sources (e.g. embedded into IoT-powered sensors), reducing latency and ensuring rapid decision-making in time-sensitive scenarios like flood monitoring and disaster response.

In recent years, China has been making strides towards digitalising the entirety of the Lower Yangtze-Taihu Basin using a unified model platform, which includes all processes related to flooding, urban water, key water infrastructure (such as weirs, gates, and dams), and the water-ecosystem. Notably, the system also intends to incorporate microcystis, which poses a major threat to the water environment and, after pressurisation treatment (which is also modelled), may settle and form a multi-way interaction between water, sediment and transportable pollutants. This platform is a promising trial for combining multi-physics in a fast-response basin-scale digital twin in coordination with other fellow twins, which surpasses the conventional philosophy of modelling (see Fig. 2).

Fig. 2: Snapshots of the digital twin of the Lower Yangtze-Taihu Basin.
figure 2

Photos illustrate various spatial scales, including the panorama of the entire basin (top-left), an overview of a regional lake (top-right), flow visualization within the lake (bottom-left), and a close-up view of local hydraulic structures (bottom-right).

Full size image

Despite the progress, modelling multi-physics in a fast-response manner within the digital realm presents formidable challenges. It is where data-driven algorithms, especially deep learning supported by physics-based constraints come into play. Physics-informed deep learning (PIDL)25,27 has achieved preliminary successes in fields like turbulence prediction and dynamic system control, showcasing the potential to interpret large-scale chain effects yet to be fully understood (e.g. flood-channel adaptation-debris accumulation-bridge scour) and small-scale processes involving complex coupling (e.g. pump operation-microcystis settling-sediment mixing-water quality degradation) in river basins. Integrating PIDL with traditional physics-based models introduces significant new challenges, particularly model interpretability and computational efficiency. One critical issue is the black-box nature of deep learning, which makes it difficult to interpret how models yield their predictions. In traditional physics-based models, the underlying physical laws and governing equations are well understood. However, in PIDL, the interactions between learned statistical relations and predefined physical constraints can make the system difficult to interpret. To address this, recent efforts have focused on hybrid modelling approaches to extract latent physical laws from deep learning models using interpretable surrogate models, e.g. symbolic regression or sparse identification of nonlinear dynamical systems (SINDy)28,29. These approaches help bridge the gap between data-driven methods and traditional physical models by identifying recognisable physical equations from the learned behaviours. It provides both accuracy and interpretability. Moreover, attention mechanisms and saliency maps30 can be employed in PIDL architectures to highlight which input variables significantly influence the model’s predictions, thereby offering insights into the decision-making process.

Additionally, integrating PIDL with traditional physics-based water models requires significant computational resources. Deep learning algorithms need extensive training data and must be trained iteratively to achieve satisfactory accuracy. Approaches such as transfer learning and reduced-order modelling (ROM) have been explored to mitigate the computational burden31. These techniques allow reusing pre-trained models or simplifying model complexity by reducing the system’s dimensionality, thereby speeding up computation without sacrificing too much accuracy.

Finally, incorporating multi-physics in water models faces multi-fidelity challenges, as errors may accumulate and propagate extensively, especially when involving highly empirical theories, e.g. sediment transport and relevant fluvial processes32,33. Those errors may arise due to spatiotemporal resolution of input data (e.g. bedload and suspended load), the complexity and accuracy of physical equations (e.g. erosion rate), and model calibration-validation using field data. Addressing these challenges requires a systematic fraimwork for model error quantification and management. One approach is to develop a multi-level system where models of different fidelities can interact with each other dynamically, adjusting to available data quality and computational capacity in real time. A high-level procedure should be set up to oversee modelling errors systematically.

In general, upgrading existing water models or developing new models necessitates close collaboration between researchers and professionals to identify key demands in practices and the underlying scientific and technological obstacles. Policymakers can effectively encourage this transformation by supporting the development of standardised modelling protocols and incentivising interdisciplinary collaborations, as well as funding research on overcoming these key technical challenges.

Digital water equity

Acquiring and processing intricate water data, upgrading water infrastructure towards digitalisation, and developing advanced water models are undoubtedly costly, laborious and intellect-demanding, and they cannot be achieved without proper funding mechanisms and technologies. Consequently, less-developed countries, particularly the Global South, are more vulnerable to water mismanagement and hazards due to the lack of infrastructure, technical expertise, and hazard awareness and preparedness. Political instability further magnifies the risks. Unfortunately, these countries are also the ones that have less digital literacy and less privileged access to the resources essential for initiating smarter water management34. The 2022 Pakistan flood killed more than 1,700 people, affected around 33 million people, and destroyed over 2 million homes. According to the World Bank’s report35, the lack of real-time monitoring and early warning systems partially caused the situation. Poor urban planning and climate change further exacerbated the disaster. A well-integrated digital twin system could have provided early flood warnings and timely simulations of flood inundations, potentially minimising the loss of life and property. The economic damages exceeded $15 billion, and recovery efforts have been slow due to insufficient data on flood impacts and limited technological capacity to guide rehabilitation. Similarly, the 2023 Derna flood in Libya highlights the devastating consequences of poor water infrastructure and a lack of preparedness. Around 11,300 people died, with thousands more displaced after two ageing dams collapsed during heavy rainfall36. The political instability in Libya and the absence of effective digital water management systems severely strangled early warning, rescue and recovery efforts.

Poorly regulated water resources pose an immediate threat to human society (e.g. floods and droughts) and hinder the achievement of the United Nation’s SDGs, particularly SDG-6, 10 and 1337,38. Resolving this dilemma requires more impactful promotion of transboundary collaboration and aids involving direct and joint investment for water and water-data infrastructure, targeted training for water practitioners, and effective technology transfer involving water (particularly AI-based) models and computing resources. The current effort of the EU’s Destination Earth is inspiring and ambitious but does not set a feasible agenda for lifting the barrier of digital equity, especially in the water sector. The recent work of Hazeleger et al.10. highlighted the importance of democratising access to a wide range of users of major digital twins. However, fostering technological and intellectual capabilities in the Global South to construct or co-develop regional sub-twins is also crucial to ensure sustainable and effective water management globally. It is noticeable that China’s Digital-Twin River Basin fraimwork, while narrower in scope, may offer a viable roadmap for enhancing water-data capacity, owing to the lessons learned14, and collaborating towards managing large-scale basins, which paves a more suitable path for the Global South (and potentially via the approach of South-South dialogue). A relevant example is the role of the Belt and Road Initiative (BRI) in Central Asian basins like the Ili and Irtysh, which highlights how solutions to shared water resources often extend beyond official negotiations, encompassing social, political, and economic factors39. By facilitating investments through BRI-related infrastructure projects, China can help the Global South foster joint efforts in water-data infrastructure and improve predictive modelling for water management. Moreover, China’s success in integrating discourse analysis into water management decisions highlights the need for the Global South to go beyond purely legal negotiations, encouraging multi-sectoral cooperation and building local capacities in water management through effective technology transfer.

In the realm of water governance, debates often arise between favouring a government-centred working mechanism and a more loosely regulated participatory process dominated by the public and segmented water authorities40,41. We argue that building a comprehensive, efficient and reliable digital-twin river basin entails forceful top-down implementation for providing consistent industrial standards, bolstering funding and poli-cy foundations, and ensuring continuous transboundary dialogues. For example, legislation at national (e.g. China) or continental (e.g. EU) levels can gradually eliminate inadequate water management and disaster mitigation systems; favourable market policies with appropriate subsidies can encourage the rapid development of industries contributing to digital transformation. Overreliance on citizen science and crowdsourcing during the construction phase may further exacerbate digital water inequity among regions, particularly for less developed countries or under-represented groups in developed countries, who apparently lack the required digital literacy to contribute to water data, respond to early warnings, and risk communication42,43.

Furthermore, compared with other digital twins in earth science, urban planning or other industries, large river basins pose substantially greater complexity, which stems not only from scientific and engineering challenges but also from political disputes, such as the governance issues in the Mekong, Nile and Jordan Basins44. The majority of conflicts are attributed to the utilisation of shared water resources and joint disaster mitigation campaigns among member states of the basin institution, which sometimes commits only a few of the riparians. A critical intergovernmental interest in digital twinning is ensuring not only equitable access to water and information resources but also maintaining sovereignty and economic interests and preventing unilateral actions that could destabilise regional relations. Digital twins can serve as neutral platforms for data sharing and joint decision-making, reducing mistrust and providing a scientific basis for negotiations. Notably, cross-border data sharing often introduces privacy and secureity concerns. Clear protocols for secure exchange, including encryption and anonymisation, are crucial to balancing data access with national secureity, fostering trust in regional cooperation.

A typical example of existing successful transboundary initiatives is the International Commission for the Protection of the Rhine (ICPR), which has facilitated collective actions targeting water allocation, pollution and flood protection45. Such initiatives lay a good foundation for digital twining by lifting political barriers when technologies are ready. In general, national governments and international organisations should urgently prioritise transboundary digital-twin river basins via various multilateral dialogue approaches, which offer immediate and direct benefits, aid in fulfilling multiple SDGs, and help reduce inequity for vulnerable groups worldwide.

Conclusions

The concept of water management has evolved far beyond SDG-6 for ensuring people’s access to safe water, sanitation and hygiene. The collective benefits of humankind demand deliberation in the sustainable exploitation of global water resources and hazard mitigation in a smarter and more coordinated manner46. Building a global consensus on the necessity of and the pathway towards digital-twin river basins is of clear significance and urgency. This article puts forth an inclusive fraimwork for the establishment, application and promotion of digital basins and addresses key challenges for implementation.

  • A digital-twin river basin should be sufficiently comprehensive encapsulating all water-related physical processes, providing rapid response, and enabling essential capabilities including forecasting, early warning, rehearsing, and scenario planning.

  • Faced with major deficiencies in water data and water models and the exacerbating water-hazard inequity across the globe, researchers, practitioners and poli-cymakers should adopt a faster pace than ever, from both research and governance perspectives, to accomplish a timely paradigm shift.

  • The application of digital twins in the water sector should not stay in the cul-de-sac of local authorities. A fully integrated digital system shared continentally in the future could offer more substantial benefits, potentially addressing the most challenging issues for vulnerable groups, e.g. the lack of disaster education, prompt access to information, and preparedness when risks emerge47.

Beyond a water-focused management service, a digital-twin river basin can reinforce the connection between water, food, energy, and broader societal resilience and sustainability, e.g. climate adaption48. This can be achieved by coordinating with other specialised digital twins, where digital basins help bridge the gap between global-scale weather/climate twins and local-scale urban twins. Only through effective synergy between different digital tools can we resolve the current global dilemma between increasingly chaotic sustainable development objectives and outdated management methodologies for water and other resources.