2025 年 13 巻 4 号 p. 61-74
Urban flood early warning systems (EWS) are increasingly recognized as vital components of climate resilience strategies in rapidly urbanizing areas. With intensifying hydrometeorological risks and infrastructural vulnerabilities, timely and accurate flood alerts are essential to minimizing disaster impacts. This systematic review synthesizes evidence from over 40 peer-reviewed studies published between 2015 and 2025, spanning technological innovations, modeling techniques, governance frameworks, and community engagement approaches in urban EWS.
Guided by PRISMA methodology, the review identifies key advances in real-time monitoring (IoT, remote sensing), data-driven flood forecasting (AI/ML), and integrative decision-support tools. However, persistent barriers, including fragmented institutional coordination, limited model generalizability, and inequitable risk communication, continue to hinder EWS effectiveness, especially in low-resource settings.
The paper proposes a multi-dimensional framework for future research and implementation, emphasizing hybrid modeling, participatory design, and inclusive governance. Findings call for integrated, context-sensitive EWS architectures that balance technical sophistication with social trust and equity.
Urban flooding is an increasingly critical threat to cities worldwide, driven by the combined pressures of climate change, rapid urbanization, and insufficient drainage infrastructure [1, 2, 3, 4]. As the frequency and intensity of extreme weather events escalate, urban areas, especially those with aging or poorly planned infrastructure, face mounting risks of catastrophic inundation and socioeconomic disruption.
Early warning systems (EWS) have emerged as a key non-structural strategy to mitigate such risks. These systems provide timely alerts to enable preparedness actions, reduce loss of life, and minimize property damage [5, 6, 7, 8, 9]. Technological advancements have significantly influenced the evolution of EWS. Innovations in the Internet of Things (IoT), artificial intelligence (AI), machine learning (ML), and remote sensing have substantially improved real-time flood monitoring, forecasting accuracy, and the efficiency of communication protocols [10, 11, 12, 13, 14, 15, 16, 17].
However, despite these advances, the effectiveness of EWS remains uneven. Significant challenges persist regarding system interoperability, data integration, institutional governance, and community engagement, particularly in low-resource settings [18, 19, 20]. As noted by Perera et al. [7] and Samansiri et al. [18], socio-political barriers such as low public trust, fragmented institutional mandates, and lack of clear standard operating procedures often undermine the translation of warnings into timely, life-saving actions.
This review aims to synthesize cutting-edge research on urban flood early warning systems by exploring the interplay between technological innovation, modeling approaches, societal dynamics, and implementation barriers. By integrating insights from over 40 peer-reviewed studies, including systematic reviews, case studies, and critical frameworks, this paper seeks to highlight current best practices, identify persistent research gaps, and propose future directions for developing more resilient, inclusive, and adaptive EWS frameworks.
This review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to ensure methodological rigor and transparency.
2.1 Search strategyThe literature search was conducted across three major databases: Scopus, Semantic Scholar, and PubMed to maximize coverage of interdisciplinary studies on urban flood EWS. The Boolean string applied was: (“urban flooding” OR “urban flood risk”) AND (“early warning system” OR “EWS”) AND (“IoT” OR “AI” OR “remote sensing” OR “machine learning”).
To improve coverage, we considered synonymous keywords and wildcards (e.g., flood*, system*, remote sens*) and recognized that omission of non-English works introduces a potential language bias. The decision to exclude conference proceedings and grey literature was deliberate, to ensure methodological transparency, though it may have excluded recent AI/IoT applications often published first in conferences.
2.2 Inclusion and exclusion criteriaEligible studies explicitly addressed urban flood EWS through empirical investigations, systematic or structured reviews, or conceptual modeling frameworks. Exclusion criteria were: (i) studies focusing solely on rural or river-basin contexts without urban relevance; (ii) editorials, commentaries, or non-peer-reviewed reports; (iii) duplicates or incomplete records.
2.3 Screening and selection processThe initial search yielded 1,050 records. After duplicate removal (n = 724; 68.9%), 326 articles remained for title and abstract screening. Following a standardized eligibility assessment, 40 articles were retained for final inclusion (Fig. 1).
2.4 Data extraction and synthesisA structured extraction matrix was employed to capture information on research objectives, study design, geographic scope, data sources, technologies used (IoT, AI/ML, remote sensing), and governance/communication aspects. The synthesis was thematically organized into four domains: (i) technological innovations, (ii) modeling and forecasting, (iii) governance and risk communication, and (iv) community engagement.
2.5 LimitationsThe search was limited to English-language, peer-reviewed journal articles, which may have introduced language and publication bias. Exclusion of conference papers likely omitted specific cutting-edge AI/IoT applications. Moreover, while duplicates were systematically removed, potential overlap in database indexing cannot be completely ruled out. These limitations notwithstanding, the use of PRISMA methodology and transparent reporting of search parameters enhances reproducibility and minimizes subjective bias.
Over the past decade, urban flood EWS have been transformed by IoT, AI/ML, and remote sensing advances. These technologies now enable real-time monitoring, higher-resolution forecasts, and faster communication of alerts [10, 11, 12, 13, 14, 15, 16, 17, 22, 23, 24, 25, 26].
IoT-based systems are increasingly used to expand and improve flood sensing in urban areas. These platforms integrate diverse sensor arrays such as ultrasonic level detectors, rain gauges, flow meters, and barometric pressure sensors with low-power communication networks, including Long Range Wide Area Network (LoRaWAN), Zigbee, and Global System for Mobile Communications (GSM). This architecture enables low-latency data transmission to centralized analytic hubs [10, 23, 24, 25, 26]. For instance, Farabi and Sintawati (2024) developed an IoT-enabled EWS at the Jakarta Dam employing a real-time fishbone analysis method, demonstrating enhanced responsiveness to flash flood signals [23].
In parallel, AI and ML algorithms have significantly improved flood prediction by enabling data-driven, adaptive learning models. Models such as Random Forest (RF), Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and hybrid ensemble methods are increasingly used to extract nonlinear patterns from hydrological time series and spatial datasets [10, 11, 12, 14, 16, 27, 28, 29]. As reported by Bentivoglio et al. (2021), deep learning methods consistently outperform traditional hydrological models in terms of accuracy and scalability, particularly under data-rich conditions [14].
Remote sensing technologies such as Synthetic Aperture Radar (SAR), Light Detection and Ranging (LiDAR), and multispectral imagery now play a central role in mapping the extent of urban floods. These technologies enable rapid, wide-area inundation assessments, even during adverse weather conditions [30, 31, 32, 33]. Furthermore, digital visualization platforms, such as digital twins and augmented reality interfaces, are increasingly adopted to enhance operational awareness and support emergency coordination [34, 35].
The movement toward sensor fusion, integrating diverse data inputs like satellite data, field-based sensors, weather models, and even social media signals, is facilitating more robust decision support systems (DSS) for real-time flood risk management [10, 17, 26]. For example, Tao et al. highlight novel sensor architectures combining mobile sensing, IoT infrastructures, and in-situ analytics for dynamic flood impact assessment [26].
However, widespread adoption remains limited by challenges such as inconsistent data standards, calibration issues, and weak network resilience problems that are especially acute in low-resource urban settings [2, 15].
3.2 Modeling approaches and forecastingAccurate forecasting of urban flood events depends on the integration of hydrological, hydraulic, and data-driven modeling frameworks, each offering complementary strengths within EWS [11, 12, 6, 32, 34, 38].
Physically based models, including hydrological and hydraulic simulations, remain foundational to urban flood forecasting. Models such as Hydrologic Engineering Center-River Analysis System (HEC-RAS), MIKE Flood, and Storm Water Management Model (SWMM) are widely employed to simulate 1D–2D flow dynamics in complex urban drainage systems [6, 11, 12, 32, 34]. These models require granular input data on terrain elevation, land use, rainfall patterns, and drainage infrastructure, and are valued for their interpretability and regulatory acceptance.
Yet, their high computational demands make real-time use difficult, particularly in data-scarce regions [2, 36, 38]. Moreover, their calibration and validation processes are time-consuming and may exhibit limited generalizability across cities with differing hydrological conditions or infrastructure typologies.
In response, researchers have increasingly adopted data-driven and ML approaches, which enable rapid, adaptive forecasting with reduced data dependencies [11, 12, 14, 16, 27, 28, 29]. Algorithms such as Long Short-Term Memory (LSTM) networks, Support Vector Regression (SVR), and hybrid architectures combining physics-informed and statistical layers have demonstrated promising results in predicting short-term flood risks.
For instance, Anafi et al. documented the widespread application of LSTM and Artificial Neural Network (ANN) models in Southeast Asia for real-time flood forecasting, particularly in data-constrained settings [27]. Similarly, Piadeh et al. emphasized the potential of integrating real-time sensor data with AI models to enhance forecasting responsiveness and model adaptability [11].
Nevertheless, ML-based models often lack interpretability, and their accuracy tends to decline when applied beyond the conditions of their training data [14, 16, 38].
To overcome these limitations, emerging research promotes hybrid modeling paradigms, which blend deterministic simulations with data-driven components for error correction, adaptive calibration, or probabilistic risk estimation [14, 16]. For example, a hybrid model might use SWMM for hydraulic routing while applying LSTM layers to adjust forecasts based on rainfall anomalies or near-real-time inflow data. These AI-augmented hybrid models balance accuracy, computational efficiency, and real-world applicability [32, 38].
Uncertainty quantification techniques, including ensemble forecasting and Monte Carlo simulations, are increasingly integrated into urban flood models, offering probabilistic forecasts that support risk-informed, anticipatory decision-making under uncertainty [11, 14].
While physically based models such as SWMM and HEC-RAS are valued for their interpretability and regulatory acceptance, they are often computationally intensive and require extensive calibration data. Data-driven models, particularly those using LSTM or RF, offer rapid forecasting and adaptability in data-scarce settings but usually lack physical transparency and generalizability. Hybrid models attempt to bridge this gap by combining physical interpretability with data-driven adaptability, though they remain relatively underdeveloped regarding standardized frameworks and operational integration.
A detailed comparison of these approaches is presented in Table 1, highlighting their respective strengths, limitations, and typical use-case contexts.
| Model Type | Strengths | Limitations | Suitable contexts |
|---|---|---|---|
| Physically based |
- High interpretability; - Accepted by regulatory bodies; - Captures physical processes. |
- Requires high-quality input data - Computationally expensive - Difficult to calibrate in real time |
Data-rich urban areas with detailed topographic/hydraulic data. |
| Data-driven (ML/AI) |
- Fast processing; - Adaptable to real-time data; - Suitable for sparse data environments. |
- Lacks physical interpretability; - Poor generalizability; - Needs large training datasets. |
Cities with limited sensor infrastructure, tropical floods. |
| Hybrid |
- Combines the strengths of both models; - Potential for real-time correction; - Improved accuracy. |
- Still emerging conceptually; - Lack of standardized frameworks; - Integration challenges. |
Complex urban systems require both accuracy and speed. |
While technological advancements have significantly enhanced the capabilities of urban flood EWS, their practical effectiveness is frequently undermined by societal, institutional, and communication-related challenges [5], [7, 9, 18, 19, 20, 39].
A significant structural barrier lies in the governance frameworks underpinning EWS. Numerous studies have emphasized that the lack of clearly defined institutional roles, mandates, and standard operating procedures (SOPs) across agencies often leads to poor coordination and delayed response [7, 18, 19, 20]. For instance, Samansiri et al. employed Interpretive Structural Modeling (ISM) to identify institutional ambiguity and inter-agency misalignment as key failure points in flood early warning and response systems (FEWRS) [18].
In many developing countries, the absence of cohesive disaster risk governance and limited regulatory oversight has resulted in underfunded, poorly maintained EWS infrastructure, especially in informal settlements and peri-urban areas [2, 39].
Moreover, EWS can only be effective if recipients trust, understand, and act on warnings. However, low public confidence, insufficient risk awareness, and minimal community participation frequently undermine warning effectiveness [7, 8, 19, 20]. As noted by Perera et al., skepticism often arises from false alarms, ambiguous messaging, or irrelevant warning content, thereby reducing the likelihood of timely action [7].
Civil society organizations (CSOs) play a critical mediating role, facilitating dialogue between authorities and vulnerable populations through awareness campaigns, capacity-building workshops, and co-designed communication strategies [19, 20]. Their engagement is particularly crucial in addressing equity issues and ensuring that marginalized groups such as the elderly, people with disabilities, and linguistically isolated populations are not excluded from the EWS process.
Risk communication remains one of the most underestimated yet pivotal components of EWS. Kuller et al. show that warning messages’ structure, timing, delivery channel, and clarity significantly affect public responsiveness [8]. It is not sufficient to merely transmit technical data; messages must be actionable, context-sensitive, and tailored to the information-processing capacity of diverse communities.
Best practices include multilingual alerts, visual and audio cues, and mobile-based dissemination platforms [8, 20]. Increasingly, researchers advocate for two-way communication channels, including feedback loops, participatory simulations, and community-based drills, to build trust and reinforce behavioral response [9, 20].
In fragile or resource-constrained contexts, these challenges are further amplified. For example, Morshed et al. documented that despite external donors’ technological support, EWS in Afghanistan remained ineffective due to political instability, institutional fragility, and lack of technical capacity [39].
These findings highlight the need for contextualized EWS governance models. Such models account for local political economies, institutional maturity, and levels of social capital if early warning systems are to become truly inclusive, trusted, and actionable.
3.4 Integration, limitations, and future directionsThe development of urban flood EWS has been characterized by rapid technological advancement, yet their full potential remains constrained by fragmented integration, institutional inertia, and underlying structural inequalities [6, 13, 34, 35, 38]. Unlocking the transformative power of EWS requires more than technological innovation; it demands a holistic, systems-oriented approach that bridges technical, institutional, and social dimensions.
Effective EWS must function as integrated socio-technical systems, combining real-time sensing and predictive analytics with responsive governance frameworks and participatory engagement mechanisms [6, 13, 18, 20]. This entails coordinated alignment among:
• Sensing technologies and forecasting models,
• Communication platforms and institutional protocols, and
• Community-based planning and risk governance processes.
Tools such as digital twins, decision support dashboards, and GIS-integrated visualization platforms exemplify integrative technologies that facilitate real-time situational awareness and policy-oriented scenario simulation [34, 35].
Nonetheless, such integration is often hindered by institutional silos, interoperability constraints, and infrastructural fragmentation, particularly in multi-agency urban environments [7, 18, 20].
Several cross-cutting limitations continue to impede widespread EWS deployment and effectiveness:
• High implementation and maintenance costs, especially in low-income or informal urban settlements [2, 3, 15, 38];
• Data scarcity and inconsistency, particularly in areas lacking dense sensor networks or historical hydrological records [2, 36, 38];
• Lack of methodological standardization in model calibration, validation, and uncertainty quantification, affecting generalizability across urban contexts [14, 16, 38];
• Equity deficits, including unequal access to alerts, preparedness tools, and institutional support for vulnerable populations [3, 4, 9].
These challenges reflect deeper asymmetries in the governance, financing, and technological prioritization of urban climate adaptation strategies [1, 2, 40].
Moving forward, both research and policy agendas should emphasize:
1. Hybrid and probabilistic modeling approaches that combine AI with physics-based logic, while incorporating uncertainty bounds for informed decision-making [14, 16, 38];
2. Context-specific architectures that reflect localized infrastructural realities, political economies, and planning cultures [3, 7, 20];
3. Participatory EWS design frameworks, ensuring meaningful community involvement in system planning, communication strategy, and feedback integration [7, 19, 20];
4. Low-cost, modular, and open-source EWS prototypes adapted to the constraints of low-resource settings, including decentralized sensor networks and community-managed platforms [15, 23, 26];
5. Multi-scalar policy harmonization, fostering vertical coordination across municipal, regional, and national levels for interoperable and sustainable EWS ecosystems [1, 4, 40].
Addressing these intertwined challenges is essential to evolving EWS into equitable, scalable, and resilient systems that can effectively respond to the growing risks of urban flooding in a changing climate.
| Paper | Methodology | Focus | Geographic scope | Key results |
|---|---|---|---|---|
| Performance of early warning systems in mitigating flood effects. A review | Literature review | Traditional and modern EWS | Africa | EWS reduces flood impacts; community and institutional capacity are critical |
| IoT-Enabled Flood Monitoring and Early Warning Systems: A Systematic Review | Systematic review (PRISMA) | IoT, ML, sensors | Global | IoT and ML enhance real-time monitoring and prediction accuracy |
| A Critical Review of Real-Time Modelling of Flood Forecasting in Urban Drainage Systems | Critical review | Real-time modeling, AI | Global | AI and real-time data integration improve urban flood forecasting |
| Real-Time Urban Flood Forecasting Systems for Southeast Asia - A Review of Present Modelling and Its Future Prospects | Review | Real-time forecasting, AI | Southeast Asia | Data-driven AI models are most applied; computational limits are noted |
| Critical Failure Factors of Flood Early Warning and Response Systems (FEWRS): A Structured Literature Review and Interpretive Structural Modelling (ISM) Analysis | Structured review and ISM | Governance, failure factors | Global | Governance, SOPs, and engagement are key to EWS effectiveness |
The contemporary research landscape on urban flood EWS reflects a convergence of technological innovation and real-world demand. The integration of IoT, AI/ML, and remote sensing has substantially enhanced the capacity for real-time environmental monitoring, predictive flood modeling, and automated dissemination of alerts [10, 11, 12, 13, 14, 15, 16, 17, 26].
A growing diversity of modeling frameworks reinforces these technologies, ranging from physically based hydrodynamic models to deep learning architectures that have demonstrated improved forecast precision, response time, and system scalability [11, 12, 14, 16, 32, 38]. Empirical evidence consistently highlights the role of EWS in reducing disaster impacts, facilitating earlier evacuations, enhancing preparedness, and minimizing infrastructure disruption [5, 6, 9].
However, this technological progress has not consistently translated into operational effectiveness. Numerous studies emphasize that EWS performance is frequently constrained by systemic governance limitations, fragmented institutional coordination, and insufficient public engagement, particularly in resource-constrained urban settings [7, 18, 20, 39]. This implementation gap underscores a core insight: technological sophistication alone does not ensure effective risk mitigation.
Instead, the success of EWS hinges on their ability to operate as integrated socio-technical systems embedding trust, contextual sensitivity, and interoperability across technical infrastructure, institutional actors, and local communities [18, 20, 40]. Bridging this divide requires reconfiguring EWS as a technical platform and an adaptive governance instrument.
An additional set of considerations emerged from this review that remain underexplored in current scholarship but are critical for the practical deployment of EWS. First, the growing use of IoT sensors, AI-driven models, and cloud-based communication platforms raises concerns over data privacy and cybersecurity. Unauthorized data access or misuse undermines trust and may discourage community participation in system reporting and feedback mechanisms. Addressing these issues requires stronger regulatory safeguards and privacy-by-design approaches.
Second, inter-agency coordination continues to be a structural weakness of many EWS deployments. While technological integration has advanced rapidly, institutional integration often lags. Early warnings were delayed or disregarded in several case studies due to unclear mandates or fragmented responsibilities among municipal, regional, and national agencies. Therefore, multi-scalar governance frameworks and standardized protocols are essential to ensure timely, coordinated responses.
Third, system operability under extreme hydrometeorological events remains a practical bottleneck. Sensors may fail under prolonged submergence, communication networks can be disrupted during storms, and power outages can render automated systems ineffective. Building redundancy, modularity, and backup channels into EWS design is thus crucial for operational resilience.
Fourth, the uncertainty inherent in flood forecasting models has direct behavioral implications. Communities are less likely to act decisively when forecasts are perceived as unreliable, especially when false alarms have occurred. Transparent communication of uncertainty through probabilistic warnings or ensemble forecasts can help build trust and improve response compliance.
Finally, the long-term sustainability of EWS demands attention to institutional capacity, financing, and maintenance. Donor-driven pilot projects often fail once external support ends, leaving systems underfunded or non-functional. Embedding EWS within local planning budgets, training programs, and community institutions is necessary to ensure continuity beyond the project lifecycle.
Together, these dimensions highlight that technological sophistication alone is insufficient. For EWS to be reliable and equitable, future research and practice must prioritize privacy, coordination, resilience under stress, uncertainty communication, and institutional sustainability as integral components of system design and governance.
The strength of empirical evidence is especially notable in areas such as hazard detection, forecast accuracy, and decision support efficiency [10, 11, 12, 14, 16]. However, several critical challenges remain:
• Inadequate validation of vulnerability and exposure indices;
• Limited generalizability of AI models across diverse geographies and flood regimes;
• Persistent inequities in warning dissemination and access to preparedness resources [2, 3, 7, 9, 14].
Overcoming these challenges demands transdisciplinary collaboration uniting expertise from hydrology, computer science, urban planning, social psychology, and community-based organizations. Such integration is essential to ensure that EWS are technically effective, socially embedded, and contextually relevant. Table 3 synthesis of key claims identified in the reviewed literature, with assessment of evidence strength and supporting rationale for each claim.
While numerous review papers on flood EWS exist, this study provides several distinct contributions that extend beyond prior works. First, it offers a systematic synthesis that spans four interconnected dimensions: technological innovations, modeling approaches, governance challenges, and community engagement, whereas most reviews tend to focus on a single thematic area. This cross-dimensional integration allows for identifying often overlooked interdependencies, such as how governance structures influence the deployment of IoT-based systems, or how trust and equity considerations affect the adoption of advanced modeling tools.
Second, by collating insights from both highly cited reviews and diverse case studies, this paper prioritizes three critical gaps that cut across disciplinary boundaries: (i) limited generalizability and interpretability of AI-driven flood models; (ii) insufficient vulnerability mapping and validation in low-resource settings; and (iii) persistent inequities in risk communication and warning dissemination. While earlier reviews have noted aspects of these issues in isolation, this study consolidates them into a coherent research agenda.
Finally, this review contributes to methodological transparency by employing PRISMA-based selection procedures and providing a detailed account of search strategies, screening criteria, and bias minimization methods (see Section 2). This approach enhances reproducibility and ensures that the synthesis is both rigorous and transparent. Taken together, these contributions highlight the value of this study not merely as a repetition of existing reviews but as an integrative and forward-looking framework for advancing urban flood EWS research.
Compared with existing reviews, which emphasize technical or governance perspectives in isolation, this study provides an integrative framework that explicitly demonstrates how these dimensions intersect in practice. By systematically consolidating findings across technological, institutional, and social domains, the paper highlights three underexplored but critical challenges: (i) the limited transferability of AI-driven forecasting models across diverse urban contexts, (ii) the lack of validated vulnerability mapping approaches in low-resource settings, and (iii) persistent inequities in access to and trust in early warning messages. These insights are not merely repetitions of earlier reviews but represent a cross-cutting synthesis that foregrounds interdependencies and outlines future research priorities. As such, the study offers novel contributions to the literature by shifting the focus from siloed technical discussions to a more holistic, multi-dimensional agenda for advancing resilient and inclusive urban flood EWS.
| Claim | Evidence Strength | Reasoning | Papers |
|---|---|---|---|
| EWS reduces loss of life and property in urban floods | Strong | Multiple reviews and case studies show that EWS enables timely evacuation and preparedness, reducing impacts. | [5, 6, 7, 8, 9, 10, 11, 12] |
| IoT and AI/ML technologies improve real-time flood monitoring and prediction. | Strong | Systematic reviews and comparative studies demonstrate higher accuracy and responsiveness. | [10, 11, 12, 13, 14, 15, 16, 17, 26] |
| Governance, SOPs, and community engagement are critical for EWS effectiveness. | Strong | Structured reviews and ISM analyses identify these as key failure/success factors. | [7, 8, 18, 19, 20, 39] |
| Integration of advanced modeling (AI, deep learning) faces generalizability and uncertainty quantification challenges. | Moderate | Reviews highlight model limitations, data gaps, and the need for probabilistic approaches. | [2, 11, 12, 14, 16, 38] |
| High implementation costs and maintenance limit EWS adoption in low-resource settings | Moderate | Several papers note cost, infrastructure, and data challenges, especially in developing countries | [2, 3, 15, 38, 40] |
| Lack of participatory approaches and tailored communication reduces warning effectiveness. | Moderate | Studies show community engagement and risk communication gaps, especially in non-industrialized regions. | [7, 8, 9, 19, 39] |
Urban flood EWS have become an essential component of urban climate adaptation strategies, offering the potential to reduce disaster risk through timely alerts, predictive analytics, and multi-scalar coordination. This review highlights how recent advances in IoT, AI/ML, remote sensing, and hybrid modeling approaches have significantly strengthened the technical foundations of EWS, enabling more accurate forecasting, real-time monitoring, and automated dissemination.
Nevertheless, technological innovation alone does not guarantee operational success. Persistent challenges, including governance fragmentation, community disengagement, data gaps, and limited institutional capacity, continue to impede the equitable and effective implementation of EWS, especially in low-resource and politically fragile contexts.
Realizing the full potential of EWS requires a paradigm shift: from siloed technical systems to integrated, inclusive, and adaptive risk governance platforms. This involves advancing predictive capability, building trust, enhancing community co-ownership, and ensuring institutional alignment across governance levels. In doing so, EWS can be transformed into resilient, equitable, and scalable solutions capable of addressing the growing challenges of urban flooding in a changing climate.
5.1 Research gapsDespite technological progress, this review highlights three high-priority research gaps:
(i) the lack of generalizable and transferable AI-based forecasting models;
(ii) inadequate methods for vulnerability assessment and exposure mapping in data-scarce urban contexts; and
(iii) persistent inequities in early warning access and responsiveness among vulnerable populations.
Beyond urban centers, peri-urban zones and agricultural systems present unique challenges for early warning. These transitional landscapes often combine rural and urban hydrology features but are frequently excluded from current EWS design and governance frameworks. Future research should focus on (i) developing low-cost, decentralized warning systems tailored to peri-urban expansion areas; (ii) integrating agricultural water management with urban flood risk governance, as irrigation and land-use changes can exacerbate flood exposure; and (iii) cross-scalar institutional coordination to ensure warnings in peri-urban regions are consistent with urban core systems.
Figure 2 illustrates the distribution of the 40 reviewed studies across thematic topics and attributes. The darkest clusters are observed in “Forecasting and Mapping” with AI/ML modeling (15 papers) and “Real-time Monitoring” with IoT/Sensor technologies (14 papers), reflecting the dominant focus on technological advancements. By contrast, governance, policy, and community engagement intersections show considerably fewer studies (typically 2–3 papers per cell). A moderate concentration appears in “Communication and Dissemination” linked to “Community Engagement” (8 papers), suggesting growing but still limited attention to participatory approaches. The heatmap strongly emphasizes technology-driven research and comparatively limited work on governance, vulnerability assessment, and low-resource implementation.
Future research should address integrating advanced technologies with participatory governance, improving model generalizability, and ensuring equitable access to EWS. Table 3 outlines emerging research priorities from the literature review, offering a forward-looking perspective on developing more resilient, inclusive, and scalable early warning systems.
| Question | Why |
|---|---|
| How can AI and deep learning models be more generalizable and robust for flood forecasting across diverse urban environments? | Generalizability is crucial for reliable EWS deployment in varied urban contexts, mainly where data availability and urban characteristics differ. |
| What participatory approaches can enhance community engagement and trust in urban flood early warning systems? | Community engagement ensures warnings are understood and acted upon, increasing the effectiveness and equity of EWS. |
| How can low-cost, scalable EWS be implemented in low-resource urban settings to ensure equitable disaster risk reduction? | Addressing cost and infrastructure barriers is essential for protecting vulnerable populations in developing cities. |
| How can EWS be adapted to address peri-urban flood risk and agricultural-urban linkages? | Peri-urban areas and agricultural interfaces are increasingly vulnerable due to land-use change, insufficient drainage, and governance gaps, yet remain understudied in current EWS research. |
