An LCZ-Based Machine Learning Reveals Differences in Coastal High-Density Urban Flood Risk: Enhancing Interpretability and Simplifying Morphology

An LCZ-Based Machine Learning Reveals Differences in Coastal High-Density Urban Flood Risk: Enhancing Interpretability and Simplifying Morphology | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article An LCZ-Based Machine Learning Reveals Differences in Coastal High-Density Urban Flood Risk: Enhancing Interpretability and Simplifying Morphology Yongheng Wang, Qingtao Zhang, Kairong Lin, zhuochao Zhang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9078373/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Urban pluvial flooding is intensifying under rapid urbanization and climate change, yet most data-driven assessments underrepresent the role of urban morphology. We tentatively treat the Local Climate Zone (LCZ) scheme as a standardized morphological proxy rather than a purely hydrological variable. This study introduces an innovative analytical framework that integrates LCZ-based morphological indicators into a LightGBM machine learning model, enhanced by Shapley Additive Explanations (SHAP) for improved interpretability. Using data derived from Guangzhou and Shenzhen, we constructed two model scenarios: a baseline model employing traditional socio-environmental variables and an enhanced model incorporating LCZ typologies. The enhanced model demonstrated a substantial improvement in predictive accuracy, particularly in Guangzhou, where LCZ-related factors contributed over 30% to the model's importance, with a higher relative contribution rate than standalone 3D building metrics. Compared with conventional land-use classification, LCZ produced a markedly finer-grained urban form. Besides, SHAP analyses further revealed distinct threshold effects associated with specific land coverage levels. By coupling standardized morphology with interpretable machine learning, this framework is scalable across cities and provides actionable guidance for adaptive planning It prioritizes infrastructure improvements—such as street renovations and permeable upgrades—in areas exhibiting the highest morphological sensitivity. Urban flooding Local Climate Zones Machine Learning SHAP Exposure risk Adaptive flood management Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Urban flooding represents one of the most urgent and complex challenges in contemporary urban governance, intensified by the confluence of accelerating climate change and rapid urbanization. According to the IPCC’s Sixth Assessment Report, urban flood risks are rising with high confidence under climate change scenarios, driven by the increased frequency and intensity of extreme precipitation events (IPCC 2023 , Jun-Neng et al. 2022 ). In China, urban floods have overtaken other natural hazards in terms of frequency and severity, affecting over 40 million people annually and causing cumulative economic losses exceeding USD 10 billion between 2001 and 2018 (Sun et al. 2020 ). This growing risk arises from the interaction between natural factors — such as storm intensity and terrain configuration — and human factors including impervious surface expansion, land use transformation, and the inadequacy of drainage infrastructure (Luo and Zhang 2022 ). Mitigating the frequency of urban flooding is essential for advancing sustainable landscape management and climate-resilient urban development (Soh et al. 2025 , Wu et al. 2025c ). To address this challenge, policymakers have introduced integrated strategies such as Low Impact Development and the Sponge City program, aiming to enhance urban hydrological resilience (Behboudian et al. 2023 , Chan et al. 2018 , Qi et al. 2025 ). However, there are still significant differences in their implementation effectiveness. In fact, the effectiveness of current mitigation strategies depends critically on a comprehensive understanding of the drivers behind urban flooding, as well as accurate risk assessments. Previous studies have underscored the dual role of natural and human-induced factors in urban flooding, including precipitation extremes, terrain variations, land use patterns, and the functionality of drainage systems. For example, land use changes and precipitation extremes play an important role in the occurrence of surface runoff (Gong et al. 2025 , Luo and Zhang 2022 ). In recent years, studies have attempted to characterize urban morphology using two-dimensional metrics (e.g., impervious surface ratio, land use diversity) and three-dimensional metrics (e.g., building density, height, and volume). While these indicators are valuable for capturing specific aspects of urban form, they differ in definition across cities and often neglect interactions that are nonlinear. Additionally, they frequently function independently and fail to represent the integrated nature of urban surfaces (Chen et al. 2025 , Lin et al. 2024 ). Consequently, current models often underrepresent the integrated, scale-coupled role of morphology in shaping exposure and susceptibility, limiting cross-city comparability and interpretability (Wang et al. 2025 , Yuan et al. 2024 ). Fortunately, the Local Climate Zone (LCZ) framework offers a standardized, semantically meaningful representation of urban morphology that addresses these limitations (Stewart and Oke 2012 ). LCZs classify the urban landscape into standardized, semantically meaningful types based on surface cover, structure, and human activity, offering a robust bridge between morphological representation and geospatial modeling. By identifying spatial pattern approaches in this way, traditional land use classification methods often fail to detect these effects. Their adoption in urban climate research has enabled significant progress in understanding urban heat islands, thermal comfort, and air quality across diverse cities (Huang et al. 2023 , Rahmani and Sharifi 2025 , Wu et al. 2025b ). The World Urban Database and Access Portal Tools (WUDAPT) initiative has further advanced LCZ mapping and standardization efforts, promoting their global comparability and operational feasibility in geospatial modeling (Ching et al. 2018 ). However, a coherent theoretical and empirical framework linking LCZ typologies with flood susceptibility is still lacking, with few studies that systematically quantify the relationships between LCZ typologies and flood dynamics (Zhang et al. 2025 ). In response, this study proposes a novel, integrative framework to assess urban flood events through the lense of LCZ typologies. By LightGBM and SHAP, we investigate whether certain LCZ categories are disproportionately associated with flood occurrence, exposure and drivers, and whether specific LCZ combinations co-contribute to higher flood risk. The study addresses three critical research questions: (1) Do urban pluvial floods cluster disproportionately within specific LCZ morphology types? (2) Which LCZ categories are associated with elevated flood risk and exposure levels? (3) Which combinations of LCZ morphology types most significantly co-contribute to urban flood vulnerability? This study aims to extend the theoretical underpinnings of LCZ morphology research and improve its practical application in urban hydrology. Our findings are expected to provide new insights into flood-prone urban morphology, thereby facilitating more spatially nuanced and morphology-informed flood mitigation strategies in an era of climate uncertainty. 2. Materials and methods 2.1. Study area and data This study is conducted in two rapidly urbanizing megacities located in the core of the Pearl River Delta (PRD) region of southern China—Guangzhou and Shenzhen. Guangzhou, the provincial capital of Guangdong, spans an area of approximately 7,434 km² and lies between 22°26′–23°56′ N and 112°57′–114°03′ E. Shenzhen, situated directly to the southeast, covers a smaller area of 1,997 km² (22°27′–22°52′ N, 113°46′–114°37′ E) but has experienced an exceptionally rapid urban transformation since the 1980s, evolving from a constellation of fishing villages into a global innovation hub. As of 2024, the two cities are home to 18.98 million (Guangzhou) and 17.99 million (Shenzhen) residents, with corresponding GDPs of $ 4,321 billion and $ 5,124 billion, respectively, ranking them among China’s most economically dynamic and globally integrated urban regions. Both cities are influenced by subtropical marine monsoon systems, with average annual precipitation of about 1720 mm in Guangzhou and 1935 mm in Shenzhen. More than 80% of this precipitation occurs between April and September, leading to a significant seasonal concentration of rainfall, which greatly increases the risk of flooding in these areas. Both cities suffer from varying degrees of flooding due to their coastal location and topography and rapid urbanization factors (Zhang et al. 2023 ). Despite their geographic proximity and shared climatic conditions, Guangzhou and Shenzhen display pronounced differences in urban morphology, demographic density, and land-use patterns. Recent studies highlight the multidimensional spatial heterogeneity of flood risk in the PRD region, driven by the interactions among urban form, socio-economic exposure, and climate anomalies. This heterogeneity underscores the necessity of moving beyond single-city analyses to conduct comparative, cross-city investigations that can reveal how varying urbanization trajectories modulate flood vulnerability under similar climatic regimes. This study integrates a diverse range of multi-source geospatial and socio-environmental datasets to support urban flood risk modeling. The datasets include flood point records, digital elevation models (DEM), land use classifications, annual rainfall, storm events, as well as socioeconomic indicators such as population, GDP, and building footprints. Urban flooding point data is provided by the Baidu Flooding Map platform ( https://oil.baidu.com/static/rainmap/index.html#/ ), which compiles crowdsourced and official reports of urban waterlogging events. We obtained 357 waterlogged points supplied by Guangzhou and Shenzhen as of December 2024 using a Python crawler technique, cleaned out points with the same location and deleted duplicate items, totaling 331 flooding points. DEM data was downloaded from the Geospatial Data Cloud ( http://www.gscloud.cn ). Land use data were sourced from an open-access dataset available on Zenodo ( https://zenodo.org/records/4417810 ) (Yang and Huang 2021 ). Rainfall data were obtained from the National Tibetan Plateau Data Center (Peng et al. 2019 ). Supplementary stormwater-related data were downloaded from the China Scientific Data ( https://www.sciengine.com/ ) (Bai et al. 2022 ). Population data were retrieved from a curated dataset hosted on Figshare ( https://figshare.com/s/d9dd5f9bb1a7f4fd3734?file=43847643 ) (Chen et al. 2024 ), while GDP data were obtained from the Resource and Environmental Science Data Center ( https://www.resdc.cn/ ). Finally, building footprint data were sourced from the Global Urban Building Dataset ( https://www.landbigdata.info/cscproject/GlobalBuildings.html ), providing detailed representations of built-up areas (Li et al. 2022b ). 2.2. Local climate zones classification The Local Climate Zone (LCZ) data used in this study are sourced from Zendo ( https://zenodo.org/records/6364594 ) (Demuzere et al. 2022 ). This dataset provides a globally consistent LCZ map with a spatial resolution of 100 meters, constructed by integrating Earth observation imagery, auxiliary geospatial features, and expert-labeled LCZ classes through supervised classification using random forest algorithms. The LCZ map achieves an overall classification accuracy of 74.5%, which increases to over 90% in urbanized areas when validated against the World Settlement Footprint data (Javadpoor et al. 2024 ). The LCZ classification framework offers a semantically rich and physically interpretable representation of urban morphology by categorizing urban spaces into 10 built types (LCZ 1–10) and 7 natural types (LCZ A–G). Each LCZ class is defined by a distinct combination of surface cover, building height, thermal inertia, and anthropogenic activity, making it especially suitable for cross-context comparative urban studies. In this study, we focus on LCZ 1–10, the built-up categories, to systematically represent the spatial and vertical structures of urban form relevant to urban flooding. By incorporating LCZs into flood risk modeling, we aim to bridge the current gap between morphological classification and hydrological assessment, and to investigate how specific configurations—such as compact high-rise (LCZ 1), open mid-rise (LCZ 5), or scattered low-rise (LCZ 9)—influence flooding susceptibility in dense urban systems (Huang et al. 2025 ). Figure 2 illustrates the spatial distribution of built LCZ classes in Guangzhou (left) and Shenzhen (right). Despite their geographic proximity, the two cities reveal markedly divergent morphological compositions. Guangzhou is dominated by LCZ 8 (32.4%) and 6 (31.6%), suggesting a loosely organized, low- to mid-rise structure across the metropolitan area. High-density development zones such as LCZs 1–3 are confined to the historical urban core and collectively account for less than 10% of the urban area. The absence of LCZ 7 indicates a limited presence of informal or temporary structures within the formal urban fabric. In contrast, Shenzhen exhibits a markedly denser morphological profile, with LCZ 8 accounting for 34.2% of the built-up area, followed by LCZ 6 (22.5%) and LCZ 9 (8.7%). Importantly, LCZs 1–4 together constitute nearly 30% of the built environment, underscoring the city’s polycentric concentration of compact high-rise and mid-rise development, particularly in the southwestern and central districts. Like Guangzhou, LCZ 7 is entirely absent, reflecting a formalized building regime with negligible informal settlement clusters. 2.3. Potential flood exposure assessment Understanding the spatial distribution of flood-exposed elements is essential for evaluating urban flood risk in a comprehensive and policy-relevant manner. In this study, potential flood exposure is defined as the aggregated concentration of vulnerable urban components—namely, population, built infrastructure, and economic assets—that are likely to be affected by a given flood hazard intensity. The aim is to quantify the magnitude and density of exposure surrounding localized flood-prone areas, and to analyze how these vary across different LCZ types. We established 1-kilometer-radius circular buffers around each recorded urban flooding point, treating each buffer as an independent exposure unit. Within each unit, we extracted three core indicators: For each flooding location 𝑖, the following exposure metrics were defined: (1) Potential Building Exposure (PBE): $$PBE=\sum_{i=1}^{n}{Bf}_{i}$$ 1 (2) Potential Population Exposure (PPE): $$PPE=\sum_{i=1}^{n}{Pop}_{i}$$ 2 (3) Potential Economic Exposure (PEE): $$PEE=\sum_{i=1}^{n}{Eco}_{i}$$ 3 where \({Bf}_{i}\) denotes the total building footprint area (in m²) within analysis unit i , \({Pop}_{i}\) denotes the population count, and \({Eco}_{i}\) indicates the economic value (CNY) within the unit. To enable a holistic and standardized comparison of flood exposure levels across space and LCZ types, we constructed a Composite exposure index (CEI) by integrating the three dimensions of exposure (Li et al. 2022a , Wang et al. 2024 ). Each of three indicators—PBE, PPE, and PEE—was first normalized to a range of [0, 1] using min-max scaling to remove unit dependency and ensure comparability. Then the CEI for each unit i is computed as a weighted average: $$CEI={w}_{1}\bullet PBE{\prime}+{w}_{2}\bullet PPE{\prime}+{w}_{3}\bullet PEE{\prime}$$ 4 where \({w}_{1}\) = \({w}_{2}={w}_{3}=\frac{1}{3}\) , assigning equal importance to all three exposure dimensions (Li et al. 2022a ). The normalized indicators were denoted as \(PBE{\prime}\) , \(PP{E}^{{\prime}}\) , and \(PEE{\prime}\) . 2.4. Modeling urban flood risk and interpretability We employed a combination of gradient boosting machine learning and post hoc interpretability methods. The dependent variable in this study is flood point density, a spatially aggregated metric that reflects the frequency of reported urban flooding events per unit area. The set of explanatory variables includes a wide range of potential flood-driving factors encompassing topography, land use, hydrology, meteorology, infrastructure, socioeconomic attributes, and LCZ-based morphological indicators. A full list of predictor variables and their data sources is provided in Table S1 . To disentangle the complex interactions between urban morphology, especially local climate zones, and flood occurrence, we employ correlation analysis, gradient boosting machine learning, and post hoc interpretability methods. To capture nonlinear dependencies among predictors, we utilized the Light Gradient Boosting Machine (LightGBM), a state-of-the-art ensemble learning algorithm based on leaf-wise decision tree boosting (Ke et al. 2017 ). LightGBM constructs additive predictive models by minimizing a loss function via gradient descent optimization. Its capacity to handle high-dimensional data and model feature interactions makes it particularly suitable for flood risk assessment in heterogeneous urban environments (Wang et al. 2025 , Zhou et al. 2024 ). Model training incorporated stratified k-fold cross-validation to reduce overfitting risk and ensure generalizability (Zhang et al. 2026 ). Feature importance was first evaluated using gain-based metrics, which reflect each variable’s contribution to loss reduction across decision splits. To enhance the interpretability of LightGBM outputs, we adopted the SHapley Additive exPlanations (SHAP) framework—an interpretable machine learning approach grounded in cooperative game theory (Lundberg et al. 2018 , Lundberg and Lee 2017 ). SHAP assigns a unique value to each feature based on its marginal contribution to the model’s prediction for each observation, thus enabling global interpretability, local interpretability and interaction analysis. Based on the analytical framework proposed by (Lin et al. 2023 ), we designed two sets of comparative modeling experiments. (i) a hydrology-only baseline with conventional predictors (topography, land use, hydrology, meteorology, socioeconomic factors) and without LCZ variables; and (ii) an enhanced model (hydrology + LCZ morphology) that incorporates all variables, including LCZ-based indicators. By comparing the predictive performance (e.g., R², RMSE) and SHAP-derived explanatory models between the two models (Wang et al. 2025 ), we can assess whether the urban morphological structure represented by the LCZ has explanatory power beyond conventional predictors to support more spatially targeted and mechanistic flood mitigation strategies. 3. Result 3.1. Spatial patterns of urban flooding across LCZ Figure 3 presents the spatial distribution of urban flooding kernel density at the community scale, revealing the spatial clustering patterns of flood-prone area across different LCZ types. This spatial analysis uncovers why flood events are not randomly distributed, but instead exhibits strong spatial dependencies influenced by the urban morphological context. In Guangzhou, flood hotspots are primarily concentrated in the central and southwestern urban districts, notably Tianhe, Yuexiu, and Haizhu. Among these, LCZ4 (31 points) and LCZ8 (30 points) exhibit the highest number of flood points, followed by LCZ6 (29 points). These areas are characterized by a high intensity of built-up land, limited green cover, and historically dense development patterns. In Shenzhen, flooding hotspots are predominantly located in the core built-up zones of Futian, Nanshan, and Luohu, which form the city’s economic and administrative centers. Among these, LCZ8 has the highest number of flood points at 57, followed by LCZ4 with 52. 3.2. Potential flood exposure across LCZ As illustrated in Fig. 4 , all three exposure indicators—population,, population, economic, and built up area—exhibited marked differences across LCZ types and between the two cities. Overall, Shenzhen displayed consistently higher levels of potential exposure than Guangzhou. The total building exposure in Shenzhen reached approximately 300 km², representing a 50% increase over Guangzhou’s approximately 200 km². Similarly, population exposure in Shenzhen was estimated at 6.5 million, compared to 4.2 million in Guangzhou—an increase of approximately 55%. In terms of economic exposure, Shenzhen recorded an estimated 90 million CNY, surpassing Guangzhou’s 55 million CNY by about 63%. These findings reflect Shenzhen’s more intensive urban development and demographic concentration, which contribute to its heightened vulnerability to urban flooding. Across both cities, LCZ4 and LCZ8 emerged as dominant contributors to potential exposure. In Shenzhen, LCZ4 accounting for approximately 23% of total PBE (70 km²), 34% of PPE (2.2 million people), and 33% of PEE (30 million CNY). LCZ8 also contributed substantially, with around 20% of PBE (60 km²), 28% of PPE (1.8 million CNY), and 28% of PEE (25 million CNY). Collectively, these two LCZ types represented over 40% of Shenzhen’s total exposure, underscoring their strategic relevance for flood risk management. In contrast, Guangzhou exhibited a more spatially diverse exposure profile, with significant contributions from LCZ2 and LCZ1 particularly within the densely populated central urban districts. The CEI further delineates the risk composition and magnitude across LCZ spatial patterns. A pie chart summarizes the relative contributions of population, building, and economic exposure for each pattern, whereas a box plot reports the central tendency and dispersion of CEI (Fig. 5 ). Flood risk differs markedly by LCZ. LCZ1 shows the highest risk (CEI = 1.06 in Shenzhen; CEI = 0.90 in Guangzhou). Regarding composition, in Shenzhen, LCZ1 has the largest population and economic exposure, whereas LCZ5 records the largest building exposure. In Guangzhou, LCZ10 has the largest population exposure, LCZ4 the largest economic exposure, and LCZ5 the largest building exposure. In Shenzhen, LCZ4 and LCZ8 have elevated median CEI values (≈ 0.25–0.40) with maxima > 0.6, indicating morphology-driven, widespread exposure in dense urban cores. In Guangzhou, although overall CEI values are lower, LCZ1 and LCZ2 exhibit larger interquartile ranges and high-risk outliers, suggesting localized hotspots plausibly linked to aging infrastructure or rapid redevelopment. 3.3. Driving factors of urban flood risk By comparing a baseline model with an enhanced model, this study highlights the critical importance of LCZ morphological characteristics in predicting urban flood (Supplementary Table S2 ). In Guangzhou, the baseline model achieved a strong fit with an R² of 0.95, a MAE of 0.23, and a MSE of 0.15. After incorporating LCZ proportions into the feature set, the enhanced model improved the R² to 0.97 and reduced the MAE by more than 50% to 0.10, and decreased the MSE to 0.13. In Shenzhen, the baseline model yielded an R² of 0.944, with both MAE and MSE at 0.16. The enhanced model improved the R² to 0.957 and slightly reduced the MAE and MSE to 0.15. These results confirm that LCZ-based morphological descriptors significantly enhance model accuracy, particularly in cities characterized by complex spatial heterogeneity. We further analyzed the importance ranking of each predictive variable under the baseline scenario. As shown in Fig. 6 , POP consistently emerged as the most influential factor in both cities, emphasizing the central role of human exposure in shaping flood vulnerability patterns. In Guangzhou, the next most important variables were PD and TRI, pointing to the combined influence of urban imperviousness and topographic variation on flood hotspot. Conversely, in Shenzhen, DR was the second-ranked factor, followed by PD, suggesting that hydrological infrastructure plays a more decisive role in flood modulation. A cumulative contribution analysis revealed that flood risk prediction in both cities is dominated by a small subset of variables. In Guangzhou, the top three variables—POP, PD, and TRI—together accounted for approximately 58% of the model’s total explanatory power, while the top five contributed over 74%. In Shenzhen, the top three drivers—POP, DR, and PD—explained over 60%, with the top five exceeding 80% of total importance. These cumulative patterns suggest that targeted interventions focusing on the highest-contributing factors may yield substantial improvements in flood risk mitigation strategies. In the enhanced model, LCZ indicators were introduced to quantify the role of urban morphology alongside conventional socio-environmental drivers in shaping urban flood risk. The results reveal a clear spatial heterogeneity between Guangzhou and Shenzhen in terms of the dominant flood-driving mechanisms. In Guangzhou, morphological factors exhibited strong predictive power. Notably, LCZ4 emerged as a key driver, with a mean |SHAP value| of approximately 0.6, surpassing traditional predictors such as POP, PD, and TRI. LCZ-related variables collectively accounted for 30.8% of the model’s explanatory importance, with open-form zones contributing 21.1%, mixed forms 5.1%, and compact types only 1.3%. SHAP dependency plots further highlighted nonlinear threshold effects: LCZ4 triggered a sharp increase in flood risk when its coverage exceeded 20% within the buffer zone (Fig. 7 c), while LCZ8 also exhibited elevated flood contributions beyond a 10% threshold (Fig. 6 g). Meanwhile, POP displayed a near-linear positive relationship with flood density, underscoring exposure-driven risk accumulation (Fig. 6 d). In Shenzhen, by contrast, the model indicated that socioeconomic and infrastructural variables played a more dominant role. POP alone yielded a mean |SHAP value| close to 0.9, far exceeding any single morphological predictor. Drainage-related variables, particularly DR, exhibited a strong negative correlation with flood risk, reinforcing the mitigating role of drainage infrastructure in high-density urban contexts. While LCZ predictors still contributed 21.1% to the overall model, their influence was relatively muted: open-form types explained 11.6%, mixed forms 5.7%, and compact types 4.9% (Fig. 7 b). SHAP dependency analysis in Shenzhen also revealed nonlinear and threshold-based dynamics, albeit with city-specific nuances. Similar to Guangzhou, LCZ4 retained a positive association with flood risk, while POP remained the most consistent positive predictor. Additionally, TWI showed a non-monotonic pattern: SHAP values increased rapidly for TWI < 5, peaked around TWI ≈ 10, and declined thereafter (Fig. 7 n), suggesting that moderately wet zones are more flood-prone than both arid and excessively saturated areas. 4. Discussion 4.1. Advantages of LCZ over traditional land use and 3D morphology Compared to traditional LUCC classification, LCZ morphology can be used to develop more accurate urban landscape classification maps. The results indicate that LCZ exhibits a greater number of blocks than land use classification, likely because the LCZ method simultaneously considers both land cover characteristics and geometric features (Jiang et al. 2023 ). LCZ exhibits a higher edge density (ED) than LUCC classification while showing a lower largest patch index (LPI), suggesting that LCZ provides more detailed boundary information and captures more complex morphological details. This study compares the results of (Wang et al. 2022 ) in characterizing urban three-dimensional morphology, selecting four three-dimensional building indicators for comparison: Building Surface Coverage (BSC), Mean Building Height (MBH), Mean Building Volume (MBV), and Building Compactness (BCD). In both cities, LCZ demonstrated the highest relative contribution rates (30.8% and 21.7%), surpassing the combined significance of the four independent three-dimensional building metrics. This indicates that the LCZ-based morphological classification integrates building height, density, and surface characteristics—providing stronger explanatory power for urban storm flooding than any single three-dimensional variable. Thus, in flood modeling, LCZ functions more as a “composite urban form–surface response unit,” serving as a simplified 2D and 3D metric approach (Zhou et al. 2025 ). It is important to note that 3D metrics still capture detailed variations within LCZ types. At finer scales or in specific scenarios (such as different height combinations within the same LCZ), three-dimensional morphological information remains complementary (Qin et al. 2025 , Zhou et al. 2023 ). 4.2. LCZ and flood risk differentiation This study demonstrates that the integration of LCZ classifications into flood risk models substantially enhances both predictive accuracy and mechanistic interpretability. Embedding LCZ metrics within a SHAP-interpretable LightGBM framework revealed that urban morphology—in conjunction with traditional socio-environmental drivers—plays a critical role in shaping the spatial distribution of flood exposure and risk. Spatial analysis showed that flood hotspots are consistently concentrated within the LCZ4 and LCZ8 zones, which not only registered the highest number of flood incidents but also exhibited elevated KDE. Across both cities, LCZ2, LCZ4, and LCZ8 emerged as the most flood-exposed typologies across three dimensions—building footprint, population, and economic, which is consistent with Zou's findings (Zou et al. 2024 ). For example, in Shenzhen, LCZ4 alone accounted for nearly one-third of total exposure, while LCZ8 contributed an additional 20–28%, indicating that morphological configurations act as spatial amplifiers of urban flooding risk, co-locating both hazards and vulnerable assets. These exposures were further corroborated by the CEI, where LCZ4 and LCZ8 consistently showed higher median and upper quantile values—some exceeding 0.6, indicating not just frequent flooding, but high potential for damage when flooding occurs. In these areas, urban form, population concentration, and infrastructure constraints interact to exacerbate the impact of flooding (Zhou et al. 2025 ). A comparative analysis between Guangzhou and Shenzhen reveals different urbanization tracks shape different urban flood risk mechanisms. In Guangzhou, LCZ-related variables explained over 30%, highlighting the structural importance of urban morphology in shaping hydrological behavior. In contrast, Shenzhen's flood risk was driven primarily by POP and DR, with LCZ variables contributing only 21% to the model—suggesting that in newer, planned urban environments, risk profiles may be dominated by infrastructure. This is because Shenzhen, as a typical newly planned city, has a highly planned and standardized urban spatial layout, with clearly delineated functional zones and relatively uniform architectural forms, and the flood risk is more dependent on whether the infrastructure configuration can match the population growth and development intensity brought about by the rapid expansion of the city. Notably, Shenzhen’s drainage system—including its widespread adoption of separate stormwater and sewage systems—is among the most advanced in China. Its early implementation of comprehensive urban drainage master plans and high-standard engineering design has substantially enhanced its resilience to pluvial flooding. The results of the SHAP analysis reveal that LCZ4 (open high rise) and LCZ8 (large scale low rise) are not only the main bearers of the high-risk areas, but also significant morphological factors driving urban floods. These zones are typically characterized by high imperviousness, low or absent vegetation cover, and are often covered by extensive areas of hard paving, buildings, and roads. During heavy rainfall, these surfaces have minimal ability to retain or infiltrate stormwater, leading to the rapid accumulation of surface runoff. When the urban drainage system is limited in capacity or poorly maintained, localized runoff aggregation can easily result in urban flooding with high short-term intensity. This finding aligns with recent studies (Huang et al. 2024 , Wang et al. 2025 ) that demonstrate a high proportion of impervious surfaces can significantly increase the peak runoff rate and likelihood of inundation once rainfall intensity surpasses a certain threshold. In contrast, compact high-rise and mid-rise zones (LCZ1, LCZ2) ranked lower despite their higher building density and population concentration. This is likely due to their superior drainage infrastructure and vertical runoff management capabilities. These areas are typically the administrative, commercial, and economic centers of cities, so their drainage systems are designed to higher standards, updated more frequently, maintained more adequately, and have greater capacity for storage and rapid discharge (Wu et al. 2025a ). These findings underscore the importance of context-sensitive flood risk assessment, particularly in rapidly urbanizing regions. The explanatory power of LCZ variables is not uniform across cities, reflecting differences in infrastructure development, urban design, and land governance. While LCZ typologies offer strong predictive utility in morphologically heterogeneous cities like Guangzhou, their contribution is more muted in newer cities like Shenzhen, where infrastructure homogeneity dampens morphological contrasts. Nevertheless, even in Shenzhen, LCZ-based modeling captures structural vulnerabilities in industrial (LCZ8) and open mid-rise (LCZ4) zones—areas with large impervious surfaces and high-value assets. Recognizing these zones as morphology-infrastructure risk interfaces can inform spatially differentiated adaptation strategies (Zhang et al. 2023 ). 4.2. Integrating LCZ typologies into flood responsive urban planning Incorporating LCZs into flood risk governance provides a framework for understanding urban hydrometeorological vulnerability beyond land use-centered assessments of urban form (Aslam et al. 2021 , Fan et al. 2023 ). Traditionally considered as a tool for analyzing thermal heterogeneity, LCZs encompass critical urban morphological features that control flood behavior, such as impermeability, built form, surface openness, and spatial continuity. This study demonstrates that LCZs function not only as descriptive classifications, but also as operational proxies for identifying structural vulnerabilities and exposure concentrations (Dong et al. 2025 , Yang et al. 2025 ). The predictive enhancement achieved by integrating LCZ data into machine learning models substantiates their analytical value, bridging the gap between urban form and hydrological function. Zones such as LCZ4 and LCZ8 consistently emerge as compound risk hotspots, characterized by the convergence of flood susceptibility and concentrated assets or population—a spatial phenomenon that traditional land-use categories fail to capture with comparable specificity. From a planning and governance perspective, LCZ-based assessment facilitates a shift from reactive to anticipatory urban flood management. By identifying high-risk morphological zones—particularly LCZ4 and LCZ8—prior to flood events, cities can strategically prioritize investments in drainage infrastructure and green adaptation measures, integrate LCZ mapping into zoning regulations, redevelopment guidelines, and early warning systems, and establish morphology-specific design standards for critical infrastructure (Grobicki et al. 2015 , Parizi et al. 2024 ). Such spatially targeted, form-sensitive interventions enable more cost-effective and socially equitable adaptation, especially in resource-limited urban contexts. Unlike conventional flood zoning, LCZ-informed strategies are rooted in the structural logic of urban form, accounting for the co-evolution of the built environment, infrastructure systems, and exposure patterns. In summary, the integration of LCZs into flood risk governance transcends their descriptive origin. It represents a framework that is scalable, transferable, and interpretable across various cities and regions. Future research should further explore their integration with dynamic exposure data and urban climate models to develop truly morphology-sensitive resilience strategies. 4.3. Limitations Despite the potential of this framework to be applied to urban stormwater management, the study has some limitations. First, the inundation point dataset used in this study lacks critical attributes such as inundation depth and spatial extent, which somewhat limits the ability to assess the severity and impact of individual flooding events. Second, the current analysis is based on static representations of urban morphology and flood events, without incorporating the temporal evolution of LCZ types or flood exposure. Incorporating longitudinal LCZ mapping and time-series flood exposure data would provide a more nuanced understanding of how urban morphological trajectories influence risk accumulation and persistence. Third, LCZs were not integrated into physically based hydrodynamic models (e.g., SWMM or MIKE URBAN), which limits the mechanistic understanding of how morphological configurations translate into real-world flood dynamics (Mondal et al. 2025 , Xu et al. 2024 ). Future research should aim to couple LCZ-informed urban morphology with physical process models to validate and enrich the causal interpretations of flood risk mechanisms. 5. Conclusion This study demonstrates the critical value of integrating LCZ classifications into urban flood risk modeling, revealing how urban morphological structures fundamentally shape flood occurrence, exposure, and driving mechanisms across megacities. Key findings indicate that flood events are not spatially random but are strongly patterned by urban form, with open-form LCZs exhibiting elevated event density and potential exposure in both Guangzhou and Shenzhen. The enhanced predictive power of LCZ features—accounting for over 30% of the explanatory strength in Guangzhou—underscores their role as structural amplifiers of flood risk, especially where impervious surfaces, infrastructural deficits, and concentrated human activity coalesce. In addition, a comparative analysis of Guangzhou and Shenzhen shows that different urbanization LCZ patterns influence different urban flooding mechanisms. Guangzhou’s flood dynamics are more sensitive to urban form, while Shenzhen's risk is more influenced by human exposure and drainage infrastructure, reflecting divergent urbanization pathways and planning baselines. SHAP analysis further revealed nonlinear and threshold effects for key LCZs and environmental drivers, offering interpretable insights for targeted mitigation. Ultimately, our findings support a scalable, interpretable, spatially accurate and morphology-based flood management framework for managing urban hydrometeorological risks in the face of continued urban sprawl and climate intensification. Declarations Author information : Authors and Affiliations School of Civil Engineering, Sun Yat-sen University, Tangjiawan, 519082, Zhuhai, Guangdong, China Yongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang Guangdong Provincial Key Laboratory for Marine Civil Engineering, Sun Yat-sen University (Zhuhai Campus), Tangjiawan, Zhuhai 519082, China Yongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang Guangdong Engineering Technology Research Center of Water Security Regulation and Control for Southern China, Sun Yat-sen University, Guangzhou 510275, China Yongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang Conflicts of Interest: The authors declare no conflict of interest. Funding: This work was financially supported by the National Natural Science Foundation of China (31270748 and 31470707), Project funded by the Hydrological Bureau of Guangdong Province (440001-2023-10716), and the Guangzhou Bureau of Hydrology project “Research on the mechanism of hydro-ecological dynamics in a typical river network area” (SWYS2023F050). References Aslam A, Rana IA, Bhatti SS (2021) The spatiotemporal dynamics of urbanisation and local climate: A case study of Islamabad, Pakistan. Environ Impact Assess Rev 91:106666 Bai H, Ming Y, Liu Q, Huang C (2022) A dataset of rainstorm in China based on GPM precipitation product during 2001–2019. China Scientific Data 7. Behboudian M, Anamaghi S, Mahjouri N, Kerachian R (2023) Enhancing the resilience of ecosystem services under extreme events in socio-hydrological systems: A spatio-temporal analysis. J Clean Prod 397:136437 Chan FKS, Griffiths JA, Higgitt D, Xu S, Zhu F, Tang Y-T, Xu Y, Thorne CR (2018) Sponge City in China—A breakthrough of planning and flood risk management in the urban context. Land Use Policy 76:772–778 Chen Y, Ma W, Shao Y, Wang N, Yu Z, Li H, Hu Q (2025) The impacts and thresholds detection of 2D/3D urban morphology on the heat island effects at the functional zone in megacity during heatwave event. Sustainable Cities Soc 118:106002 Chen Y, Xu C, Ge Y, Zhang X, Zhou Y (2024) A 100 m gridded population dataset of China's seventh census using ensemble learning and big geospatial data. Earth Syst Sci Data 16(8):3705–3718 Ching J, Mills G, Bechtel B, See L, Feddema J, Wang X, Ren C, Brousse O, Martilli A, Neophytou M, Mouzourides P, Stewart I, Hanna A, Ng E, Foley M, Alexander P, Aliaga D, Niyogi D, Shreevastava A, Bhalachandran P, Masson V, Hidalgo J, Fung J, Andrade M, Baklanov A, Dai W, Milcinski G, Demuzere M, Brunsell N, Pesaresi M, Miao S, Mu Q, Chen F, Theeuwes N (2018) WUDAPT: An Urban Weather, Climate, and Environmental Modeling Infrastructure for the Anthropocene %J Bulletin of the American Meteorological Society. 99(9), 1907–1924 Demuzere M, Kittner J, Martilli A, Mills G, Moede C, Stewart ID, van Vliet J, Bechtel B (2022) A global map of local climate zones to support earth system modelling and urban-scale environmental science. Earth Syst Sci Data 14(8):3835–3873 Dong W, Jiang R, Dong Y, Qu A, Yuan Y (2025) Relationship between LCZ and physical activity in residential areas: A mediating role of perceptions of heat risks in climate change. Urban Clim 61:102425 Fan PY, He Q, Tao YZ (2023) Identifying research progress, focuses, and prospects of local climate zone (LCZ) using bibliometrics and critical reviews. Heliyon 9(3), e14067 Gong L, Zhang X, Guo Z, Winston R, Tao S, Smith J (2025) Urban flood resilience assessment under compounding risk: joint impacts of precipitation and river level. Sustainable Cities Soc 130:106569 Grobicki A, MacLeod F, Pischke F (2015) Integrated policies and practices for flood and drought risk management. Water Policy 17:180 Huang F, Jiang S, Zhan W, Bechtel B, Liu Z, Demuzere M, Huang Y, Xu Y, Ma L, Xia W, Quan J, Jiang L, Lai J, Wang C, Kong F, Du H, Miao S, Chen Y, Chen J (2023) Mapping local climate zones for cities: A large review. Remote Sens Environ 292:113573 Huang J, Huang Z, Liu W (2025) Combining the WRF model and LCZ scheme to assess spatiotemporal variations of thermal comfort in Shenzhen's built-up areas. Sustainable Cities Soc 122:106252 Huang S, Gan Y, Chen N, Wang C, Zhang X, Li C, Horton DE (2024) Urbanization enhances channel and surface runoff: A quantitative analysis using both physical and empirical models over the Yangtze River basin. J Hydrol 635:131194 IPCC (2023) Synthesis Report of the IPCC Sixth Assessment Report, Intergovernmental Panel on Climate Change Interlaken, Switzerland Javadpoor M, Sharifi A, Gurney KR (2024) Mapping the relationship between urban form and CO2 emissions in three US cities using the Local Climate Zones (LCZ) framework. J Environ Manage 370:122723 Jiang R, Xie C, Man Z, Afshari A, Che S (2023) LCZ method is more effective than traditional LUCC method in interpreting the relationship between urban landscape and atmospheric particles. Sci Total Environ 869:161677 Jun-Neng W, Nian-Xiu QIN, Tong J, Bu-Da SU (2022) Interpretation of IPCC AR6: impacts and adaptations of climate change on cities, settlements and key infrastructure. Adv Clim Change Res 18(4):433 Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu T-Y (2017) LightGBM: A Highly Efficient Gradient Boosting Decision Tree Li C, Liu M, Hu Y, Wang H, Zhou R, Wu W, Wang Y (2022a) Spatial distribution patterns and potential exposure risks of urban floods in Chinese megacities. J Hydrol 610:127838 Li M, Wang Y, Rosier J, Verburg P, van Vliet J (2022b) Global maps of 3D building structure for urban morphology analysis. DataverseNL Lin J, Zhang W, Wen Y, Qiu S (2023) Evaluating the association between morphological characteristics of urban land and pluvial floods using machine learning methods. Sustainable Cities Soc 99:104891 Lin Z, Xu H, Han L, Zhang H, Peng J, Yao X (2024) Day and night: Impact of 2D/3D urban features on land surface temperature and their spatiotemporal non-stationary relationships in urban building spaces. Sustainable Cities Soc 108:105507 Lundberg S, Erion G, Lee S-I (2018) Consistent Individualized Feature Attribution for Tree Ensembles Lundberg S, Lee S-I (2017) A Unified Approach to Interpreting Model Predictions Luo K, Zhang X (2022) Increasing urban flood risk in China over recent 40 years induced by LUCC. Landsc Urban Plann 219:104317 Mondal K, Ghosh M, Karmakar S (2025) Global sensitivity analysis in a complex 1D-2D coupled hydrodynamic model: Flood hazard and resilience perspectives over an urban catchment. Sustainable Cities Soc 124:106279 Parizi SM, Taleai M, Sharifi A (2024) A spatial evaluation framework of urban physical resilience considering different phases of disaster risk management. Nat Hazards 120(14):13041–13076 Peng S, Ding Y, Liu W, Li Z (2019) Earth Syst Sci Data 11(4):1931–19461 km monthly temperature and precipitation dataset for China from 1901 to 2017 Qi W, Ma C, Xu H, Xu K, Lian J (2025) Flood mitigation performance of low impact development practice in a coastal city from the perspective of catchment scale. J Hydrol 649:132466 Qin Y, Kang J, Zhou H, Xu S, Li G, Li C, Tan W (2025) Assessment of the impact of urban block morphological factors on carbon emissions introducing the different context of local climate zones. Sustainable Cities Soc 119:106073 Rahmani N, Sharifi A (2025) Urban heat dynamics in Local Climate Zones (LCZs): A systematic review. Build Environ 267:112225 Soh QY, Acha S, Shah N, O’Dwyer E (2025) Simultaneous design and control optimisation of combined rainwater harvesting and flood mitigation systems. Resources, Conservation and Recycling 222, 108459 Stewart ID, Oke TR (2012) Local Climate Zones for Urban Temperature Studies. Bull Am Meteorol Soc 93(12):1879–1900 Sun S, Zhai J, Li Y, Huang D, Wang G (2020) Urban waterlogging risk assessment in well-developed region of Eastern China. Physics and Chemistry of the Earth, Parts A/B/C 115, 102824 Wang Y, Li C, Liu M, Cui Q, Wang H, Lv J, Li B, Xiong Z, Hu Y (2022) Spatial characteristics and driving factors of urban flooding in Chinese megacities. J Hydrol 613:128464 Wang Y, Zhang Q, Lin K, Liu Z, Liang Y-s, Liu Y, Li C (2024) A novel framework for urban flood risk assessment: Multiple perspectives and causal analysis. Water Res 256:121591 Wang Y, Zhang Q, Zhang J, Lin K (2025) Impact of 2D and 3D factors on urban flooding: Spatial characteristics and interpretable analysis of drivers. Water Res 280:123537 Wu P, Wang T, Wang Z, Song C, Chen X (2025a) Impact of Drainage Network Structure on Urban Inundation Within a Coupled Hydrodynamic Model. Water 17:990 Wu R, Fang X, Liu S, Peng H, Zhao H, Zhou H, Zhao X, Yang H, Yan J, Meng Q (2025b) Carbon footprint mapping in LCZs: An integrated view of urban thermal environments through coupling spatial and human activities. Sustainable Cities Soc 128:106435 Wu W, Jamali B, Marshall L, Deletic A, Zhang K (2025c) A water sensitive urban design (WSUD) planning framework for catchment-scale urban pluvial flood mitigation targets. Water Res 285:124095 Xu Y, Lin K, Hu C, Chen X, Zhang J, Mingzhong X, Xu C-Y (2024) Uncovering the Dynamic Drivers of Floods Through Interpretable Deep Learning. Earth's Future 12 Yang J, Huang X (2021) The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019. Earth Syst Sci Data 13(8):3907–3925 Yang J, Yu W, Baklanov A, He B-J, Ge Q (2025) Mainstreaming the local climate zone framework for climate-resilient cities. Nat Commun 16 Yuan B, Zhou L, Hu F, Wei C (2024) Effects of 2D/3D urban morphology on land surface temperature: Contribution, response, and interaction. Urban Clim 53:101791 Zhang H, Gao J, Zhao J, Guo F, Bai J, Wang Z, Zhu P (2025) Applicability of local climate zones in assessing urban heat risk - a survey of coastal city. Cities 164:106068 Zhang J, Zhang Q, Wang Y, Wu X, Zhang Q (2026) Global lipid production over submarginal lands can offset anthropogenic carbon emissions. Resources, Conservation and Recycling 225, 108642 Zhang Q, Wu Z, Cao Z, Guo G, Hui Z, Li C, Tarolli P (2023) How to develop site-specific waterlogging mitigation strategies? Understanding the spatial heterogeneous driving forces of urban waterlogging. J Clean Prod 422:138595 Zhou S, Geng X, Zhao J, Hei J, Wu T, Chen Z, Wu Z (2025) An LCZ-based machine learning framework for revealing spatial heterogeneity of thermal comfort in high-density areas: Enhancing explainability and fine-grid scale resolution. Sustainable Cities Soc 133:106873 Zhou S, Wang Y, Jia W, Wang M, Wu Y, Qiao R, Wu Z (2023) Automatic responsive-generation of 3D urban morphology coupled with local climate zones using generative adversarial network. Build Environ 245:110855 Zhou S, Zhang D, Wang M, Liu Z, Gan W, Zhao Z, Xue S, Müller B, Zhou M, Ni X, Wu Z (2024) Risk-driven composition decoupling analysis for urban flooding prediction in high-density urban areas using Bayesian-Optimized LightGBM. J Clean Prod 457:142286 Zou B, Nie Y, Liu R, Wang M, Li J, Fan C, Zhou X (2024) Assessing the Impact of Urban Morphologies on Waterlogging Risk Using a Spatial Weight Naive Bayes Model and Local. Clim Zones Classif 16(17):2464 Supplementary Files supplementarymaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 08 Apr, 2026 Editor invited by journal 24 Mar, 2026 Editor assigned by journal 10 Mar, 2026 First submitted to journal 09 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9078373","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":619701303,"identity":"70f3ba31-7125-4316-a02a-ce5d82d3138c","order_by":0,"name":"Yongheng Wang","email":"","orcid":"","institution":"Sun Yat-Sen University School of Civil Engineering","correspondingAuthor":false,"prefix":"","firstName":"Yongheng","middleName":"","lastName":"Wang","suffix":""},{"id":619701304,"identity":"6b2aacab-f988-43f3-a660-8c6a7417585a","order_by":1,"name":"Qingtao Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABE0lEQVRIiWNgGAWjYDACCQglxw8kDjyAifIQ1mJgLNkA1JJAipZEgwNAiigt8rObnz38UvYnwfja4YdAW+oS589IYHzwto1B3hyHFsY5x8yNZc4Z5JndTjMAajmcuOFGArPh3DYGw50N2LUwSySYSUu2GRSb3U4AaTmQuEEigU2at40hAexULIBNIv0bSEvi5tnpH2AOY/+NTwuPRI6Z5Eeglg3SOSBbmBMbbiSwMePTIiGRUybNcM7YWOJ2TsGBBIPDxhvOPGyWnHNOwnADDi3yM9K3Sf4ok5Pjn52++cOHijrZ+e3JBz+8KbORx2ULOAh42GBMAwbHBgbGBgZ4qsABGH+wITj2eJWOglEwCkbBiAQA+1Fc89wiqwwAAAAASUVORK5CYII=","orcid":"","institution":"Sun Yat-Sen University School of Civil Engineering","correspondingAuthor":true,"prefix":"","firstName":"Qingtao","middleName":"","lastName":"Zhang","suffix":""},{"id":619701305,"identity":"a4861d8f-1944-495f-b1fe-c8df993d566d","order_by":2,"name":"Kairong Lin","email":"","orcid":"","institution":"Sun Yat-Sen University School of Civil Engineering","correspondingAuthor":false,"prefix":"","firstName":"Kairong","middleName":"","lastName":"Lin","suffix":""},{"id":619701306,"identity":"83fe590d-c815-49ef-860c-ca75d70d3a8b","order_by":3,"name":"zhuochao Zhang","email":"","orcid":"","institution":"Sun Yat-Sen University School of Civil Engineering","correspondingAuthor":false,"prefix":"","firstName":"zhuochao","middleName":"","lastName":"Zhang","suffix":""}],"badges":[],"createdAt":"2026-03-10 03:01:41","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9078373/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9078373/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107078743,"identity":"e4a028ce-8724-4c40-b359-017ca8d5e5bb","added_by":"auto","created_at":"2026-04-16 13:47:49","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1672539,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGeographic location of the study area\u003c/strong\u003e \u003cstrong\u003eand distribution of flooding points in Guangzhou (a) and Shenzhen (b).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/3d85ef8c7077dfb040a6495b.jpg"},{"id":107481404,"identity":"2228cc50-67dd-49ac-9bee-010ecf85ed15","added_by":"auto","created_at":"2026-04-22 02:17:42","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1397096,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBuilt-up LCZs in Guangzhou (a) and Shenzhen (b).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/fcba111ae4b21717954ad86c.jpg"},{"id":107078745,"identity":"6a7a645d-ad5c-4c99-a72b-f42ce65ce57f","added_by":"auto","created_at":"2026-04-16 13:47:49","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":3171425,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial distribution of kernel density of flooding points at the community scale. (a, b) Kernel density distribution of flooding points in Guangzhou and Shenzhen; (c) Number of flooding points across different LCZ types in Guangzhou and Shenzhen.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/48e4951ec378ea3331d557c1.png"},{"id":107482618,"identity":"d0525f61-4709-4076-8e47-dab45968c28f","added_by":"auto","created_at":"2026-04-22 02:24:12","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":801973,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBased on the potential spatial distribution and exposure risk components of urban flooding in LCZ. Population (PPE, 10⁴), economy (PEE, 10⁴), and building exposure (PBE, km²) in Shenzhen (left) and Guangzhou (right).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/3f85e558eade0314accc05be.jpg"},{"id":107480920,"identity":"3fff69a1-2ae1-4319-bfcc-a4d665f7017b","added_by":"auto","created_at":"2026-04-22 02:14:27","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":950882,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlood exposure risk composition and levels based on LCZ morphological types. a and b represent the CEI spatial distribution and composition shares for Shenzhen and Guangzhou, respectively; c indicates the CEI risk level.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/9216a7a64e313f25c1e38930.jpg"},{"id":107078748,"identity":"24e1103e-2738-41f8-9b35-eab45b0719e7","added_by":"auto","created_at":"2026-04-16 13:47:49","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":470589,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP-based method evaluating the importance of ranking of predictive variables for flood density estimates under the baseline scenario: (a) Guangzhou; (b) Shenzhen.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/c8bbeee780c867679c2ce1e7.jpg"},{"id":107078749,"identity":"34f48248-0037-4316-a6e5-b2981e1680c6","added_by":"auto","created_at":"2026-04-16 13:47:49","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":854264,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP-based interpretation of predictor importance and marginal effects on urban flood density. (a-b) illustrate the mean absolute SHAP values for Guangzhou and Shenzhen, respectively. The donut charts display the proportional contributions of LCZ types: compact, open, and mixed forms. (c–h)\u003c/strong\u003e \u003cstrong\u003epresent SHAP dependency plots for key predictors in Guangzhou; (i–n) present SHAP dependency plots for key predictors in Shenzhen.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/78feca07bcfb8f0b8acaa2b7.jpg"},{"id":107485395,"identity":"348e05c3-5995-4adc-b5fd-c9473ce5059b","added_by":"auto","created_at":"2026-04-22 02:34:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":9815259,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/a8d823ef-3b85-4253-bbf4-7084579ae81a.pdf"},{"id":107078747,"identity":"6ee20fd8-31ac-481b-83e9-483e20f0e2ca","added_by":"auto","created_at":"2026-04-16 13:47:49","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":22302,"visible":true,"origin":"","legend":"","description":"","filename":"supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9078373/v1/9b6bbdeb89ae3214495eef4b.docx"}],"financialInterests":"","formattedTitle":"An LCZ-Based Machine Learning Reveals Differences in Coastal High-Density Urban Flood Risk: Enhancing Interpretability and Simplifying Morphology","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eUrban flooding represents one of the most urgent and complex challenges in contemporary urban governance, intensified by the confluence of accelerating climate change and rapid urbanization. According to the IPCC\u0026rsquo;s Sixth Assessment Report, urban flood risks are rising with high confidence under climate change scenarios, driven by the increased frequency and intensity of extreme precipitation events (IPCC \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, Jun-Neng et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In China, urban floods have overtaken other natural hazards in terms of frequency and severity, affecting over 40\u0026nbsp;million people annually and causing cumulative economic losses exceeding USD 10\u0026nbsp;billion between 2001 and 2018 (Sun et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis growing risk arises from the interaction between natural factors \u0026mdash; such as storm intensity and terrain configuration \u0026mdash; and human factors including impervious surface expansion, land use transformation, and the inadequacy of drainage infrastructure (Luo and Zhang \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Mitigating the frequency of urban flooding is essential for advancing sustainable landscape management and climate-resilient urban development (Soh et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Wu et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2025c\u003c/span\u003e). To address this challenge, policymakers have introduced integrated strategies such as Low Impact Development and the Sponge City program, aiming to enhance urban hydrological resilience (Behboudian et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, Chan et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2018\u003c/span\u003e, Qi et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, there are still significant differences in their implementation effectiveness.\u003c/p\u003e \u003cp\u003eIn fact, the effectiveness of current mitigation strategies depends critically on a comprehensive understanding of the drivers behind urban flooding, as well as accurate risk assessments. Previous studies have underscored the dual role of natural and human-induced factors in urban flooding, including precipitation extremes, terrain variations, land use patterns, and the functionality of drainage systems. For example, land use changes and precipitation extremes play an important role in the occurrence of surface runoff (Gong et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Luo and Zhang \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In recent years, studies have attempted to characterize urban morphology using two-dimensional metrics (e.g., impervious surface ratio, land use diversity) and three-dimensional metrics (e.g., building density, height, and volume). While these indicators are valuable for capturing specific aspects of urban form, they differ in definition across cities and often neglect interactions that are nonlinear. Additionally, they frequently function independently and fail to represent the integrated nature of urban surfaces (Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Lin et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Consequently, current models often underrepresent the integrated, scale-coupled role of morphology in shaping exposure and susceptibility, limiting cross-city comparability and interpretability (Wang et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Yuan et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFortunately, the Local Climate Zone (LCZ) framework offers a standardized, semantically meaningful representation of urban morphology that addresses these limitations (Stewart and Oke \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). LCZs classify the urban landscape into standardized, semantically meaningful types based on surface cover, structure, and human activity, offering a robust bridge between morphological representation and geospatial modeling. By identifying spatial pattern approaches in this way, traditional land use classification methods often fail to detect these effects. Their adoption in urban climate research has enabled significant progress in understanding urban heat islands, thermal comfort, and air quality across diverse cities (Huang et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, Rahmani and Sharifi \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Wu et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e). The World Urban Database and Access Portal Tools (WUDAPT) initiative has further advanced LCZ mapping and standardization efforts, promoting their global comparability and operational feasibility in geospatial modeling (Ching et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). However, a coherent theoretical and empirical framework linking LCZ typologies with flood susceptibility is still lacking, with few studies that systematically quantify the relationships between LCZ typologies and flood dynamics (Zhang et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn response, this study proposes a novel, integrative framework to assess urban flood events through the lense of LCZ typologies. By LightGBM and SHAP, we investigate whether certain LCZ categories are disproportionately associated with flood occurrence, exposure and drivers, and whether specific LCZ combinations co-contribute to higher flood risk. The study addresses three critical research questions: (1) Do urban pluvial floods cluster disproportionately within specific LCZ morphology types? (2) Which LCZ categories are associated with elevated flood risk and exposure levels? (3) Which combinations of LCZ morphology types most significantly co-contribute to urban flood vulnerability? This study aims to extend the theoretical underpinnings of LCZ morphology research and improve its practical application in urban hydrology. Our findings are expected to provide new insights into flood-prone urban morphology, thereby facilitating more spatially nuanced and morphology-informed flood mitigation strategies in an era of climate uncertainty.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study area and data\u003c/h2\u003e \u003cp\u003eThis study is conducted in two rapidly urbanizing megacities located in the core of the Pearl River Delta (PRD) region of southern China\u0026mdash;Guangzhou and Shenzhen. Guangzhou, the provincial capital of Guangdong, spans an area of approximately 7,434 km\u0026sup2; and lies between 22\u0026deg;26\u0026prime;\u0026ndash;23\u0026deg;56\u0026prime; N and 112\u0026deg;57\u0026prime;\u0026ndash;114\u0026deg;03\u0026prime; E. Shenzhen, situated directly to the southeast, covers a smaller area of 1,997 km\u0026sup2; (22\u0026deg;27\u0026prime;\u0026ndash;22\u0026deg;52\u0026prime; N, 113\u0026deg;46\u0026prime;\u0026ndash;114\u0026deg;37\u0026prime; E) but has experienced an exceptionally rapid urban transformation since the 1980s, evolving from a constellation of fishing villages into a global innovation hub. As of 2024, the two cities are home to 18.98\u0026nbsp;million (Guangzhou) and 17.99\u0026nbsp;million (Shenzhen) residents, with corresponding GDPs of \u003cspan\u003e$\u003c/span\u003e 4,321\u0026nbsp;billion and \u003cspan\u003e$\u003c/span\u003e 5,124\u0026nbsp;billion, respectively, ranking them among China\u0026rsquo;s most economically dynamic and globally integrated urban regions. Both cities are influenced by subtropical marine monsoon systems, with average annual precipitation of about 1720 mm in Guangzhou and 1935 mm in Shenzhen. More than 80% of this precipitation occurs between April and September, leading to a significant seasonal concentration of rainfall, which greatly increases the risk of flooding in these areas. Both cities suffer from varying degrees of flooding due to their coastal location and topography and rapid urbanization factors (Zhang et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite their geographic proximity and shared climatic conditions, Guangzhou and Shenzhen display pronounced differences in urban morphology, demographic density, and land-use patterns. Recent studies highlight the multidimensional spatial heterogeneity of flood risk in the PRD region, driven by the interactions among urban form, socio-economic exposure, and climate anomalies. This heterogeneity underscores the necessity of moving beyond single-city analyses to conduct comparative, cross-city investigations that can reveal how varying urbanization trajectories modulate flood vulnerability under similar climatic regimes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study integrates a diverse range of multi-source geospatial and socio-environmental datasets to support urban flood risk modeling. The datasets include flood point records, digital elevation models (DEM), land use classifications, annual rainfall, storm events, as well as socioeconomic indicators such as population, GDP, and building footprints. Urban flooding point data is provided by the Baidu Flooding Map platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://oil.baidu.com/static/rainmap/index.html#/\u003c/span\u003e\u003cspan address=\"https://oil.baidu.com/static/rainmap/index.html#/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), which compiles crowdsourced and official reports of urban waterlogging events. We obtained 357 waterlogged points supplied by Guangzhou and Shenzhen as of December 2024 using a Python crawler technique, cleaned out points with the same location and deleted duplicate items, totaling 331 flooding points. DEM data was downloaded from the Geospatial Data Cloud (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.gscloud.cn\u003c/span\u003e\u003cspan address=\"http://www.gscloud.cn\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Land use data were sourced from an open-access dataset available on Zenodo (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://zenodo.org/records/4417810\u003c/span\u003e\u003cspan address=\"https://zenodo.org/records/4417810\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Yang and Huang \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Rainfall data were obtained from the National Tibetan Plateau Data Center (Peng et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Supplementary stormwater-related data were downloaded from the China Scientific Data (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciengine.com/\u003c/span\u003e\u003cspan address=\"https://www.sciengine.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Bai et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Population data were retrieved from a curated dataset hosted on Figshare (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://figshare.com/s/d9dd5f9bb1a7f4fd3734?file=43847643\u003c/span\u003e\u003cspan address=\"https://figshare.com/s/d9dd5f9bb1a7f4fd3734?file=43847643\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Chen et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), while GDP data were obtained from the Resource and Environmental Science Data Center (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.resdc.cn/\u003c/span\u003e\u003cspan address=\"https://www.resdc.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Finally, building footprint data were sourced from the Global Urban Building Dataset (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.landbigdata.info/cscproject/GlobalBuildings.html\u003c/span\u003e\u003cspan address=\"https://www.landbigdata.info/cscproject/GlobalBuildings.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), providing detailed representations of built-up areas (Li et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2022b\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Local climate zones classification\u003c/h2\u003e \u003cp\u003eThe Local Climate Zone (LCZ) data used in this study are sourced from Zendo (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://zenodo.org/records/6364594\u003c/span\u003e\u003cspan address=\"https://zenodo.org/records/6364594\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Demuzere et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This dataset provides a globally consistent LCZ map with a spatial resolution of 100 meters, constructed by integrating Earth observation imagery, auxiliary geospatial features, and expert-labeled LCZ classes through supervised classification using random forest algorithms. The LCZ map achieves an overall classification accuracy of 74.5%, which increases to over 90% in urbanized areas when validated against the World Settlement Footprint data (Javadpoor et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The LCZ classification framework offers a semantically rich and physically interpretable representation of urban morphology by categorizing urban spaces into 10 built types (LCZ 1\u0026ndash;10) and 7 natural types (LCZ A\u0026ndash;G). Each LCZ class is defined by a distinct combination of surface cover, building height, thermal inertia, and anthropogenic activity, making it especially suitable for cross-context comparative urban studies. In this study, we focus on LCZ 1\u0026ndash;10, the built-up categories, to systematically represent the spatial and vertical structures of urban form relevant to urban flooding. By incorporating LCZs into flood risk modeling, we aim to bridge the current gap between morphological classification and hydrological assessment, and to investigate how specific configurations\u0026mdash;such as compact high-rise (LCZ 1), open mid-rise (LCZ 5), or scattered low-rise (LCZ 9)\u0026mdash;influence flooding susceptibility in dense urban systems (Huang et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrates the spatial distribution of built LCZ classes in Guangzhou (left) and Shenzhen (right). Despite their geographic proximity, the two cities reveal markedly divergent morphological compositions. Guangzhou is dominated by LCZ 8 (32.4%) and 6 (31.6%), suggesting a loosely organized, low- to mid-rise structure across the metropolitan area. High-density development zones such as LCZs 1\u0026ndash;3 are confined to the historical urban core and collectively account for less than 10% of the urban area. The absence of LCZ 7 indicates a limited presence of informal or temporary structures within the formal urban fabric. In contrast, Shenzhen exhibits a markedly denser morphological profile, with LCZ 8 accounting for 34.2% of the built-up area, followed by LCZ 6 (22.5%) and LCZ 9 (8.7%). Importantly, LCZs 1\u0026ndash;4 together constitute nearly 30% of the built environment, underscoring the city\u0026rsquo;s polycentric concentration of compact high-rise and mid-rise development, particularly in the southwestern and central districts. Like Guangzhou, LCZ 7 is entirely absent, reflecting a formalized building regime with negligible informal settlement clusters.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Potential flood exposure assessment\u003c/h2\u003e \u003cp\u003eUnderstanding the spatial distribution of flood-exposed elements is essential for evaluating urban flood risk in a comprehensive and policy-relevant manner. In this study, potential flood exposure is defined as the aggregated concentration of vulnerable urban components\u0026mdash;namely, population, built infrastructure, and economic assets\u0026mdash;that are likely to be affected by a given flood hazard intensity. The aim is to quantify the magnitude and density of exposure surrounding localized flood-prone areas, and to analyze how these vary across different LCZ types. We established 1-kilometer-radius circular buffers around each recorded urban flooding point, treating each buffer as an independent exposure unit. Within each unit, we extracted three core indicators: For each flooding location \u0026#119894;, the following exposure metrics were defined:\u003c/p\u003e \u003cp\u003e(1) Potential Building Exposure (PBE):\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$PBE=\\sum_{i=1}^{n}{Bf}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e(2) Potential Population Exposure (PPE):\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$PPE=\\sum_{i=1}^{n}{Pop}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e(3) Potential Economic Exposure (PEE):\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$PEE=\\sum_{i=1}^{n}{Eco}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Bf}_{i}\\)\u003c/span\u003e\u003c/span\u003e denotes the total building footprint area (in m\u0026sup2;) within analysis unit \u003cem\u003ei\u003c/em\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Pop}_{i}\\)\u003c/span\u003e\u003c/span\u003e denotes the population count, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Eco}_{i}\\)\u003c/span\u003e\u003c/span\u003e indicates the economic value (CNY) within the unit.\u003c/p\u003e \u003cp\u003eTo enable a holistic and standardized comparison of flood exposure levels across space and LCZ types, we constructed a Composite exposure index (CEI) by integrating the three dimensions of exposure (Li et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022a\u003c/span\u003e, Wang et al. \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Each of three indicators\u0026mdash;PBE, PPE, and PEE\u0026mdash;was first normalized to a range of [0, 1] using min-max scaling to remove unit dependency and ensure comparability. Then the CEI for each unit \u003cem\u003ei\u003c/em\u003e is computed as a weighted average:\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$CEI={w}_{1}\\bullet PBE{\\prime}+{w}_{2}\\bullet PPE{\\prime}+{w}_{3}\\bullet PEE{\\prime}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({w}_{1}\\)\u003c/span\u003e\u003c/span\u003e = \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({w}_{2}={w}_{3}=\\frac{1}{3}\\)\u003c/span\u003e\u003c/span\u003e, assigning equal importance to all three exposure dimensions (Li et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022a\u003c/span\u003e). The normalized indicators were denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(PBE{\\prime}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(PP{E}^{{\\prime}}\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(PEE{\\prime}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Modeling urban flood risk and interpretability\u003c/h2\u003e \u003cp\u003eWe employed a combination of gradient boosting machine learning and post hoc interpretability methods. The dependent variable in this study is flood point density, a spatially aggregated metric that reflects the frequency of reported urban flooding events per unit area. The set of explanatory variables includes a wide range of potential flood-driving factors encompassing topography, land use, hydrology, meteorology, infrastructure, socioeconomic attributes, and LCZ-based morphological indicators. A full list of predictor variables and their data sources is provided in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eTo disentangle the complex interactions between urban morphology, especially local climate zones, and flood occurrence, we employ correlation analysis, gradient boosting machine learning, and post hoc interpretability methods. To capture nonlinear dependencies among predictors, we utilized the Light Gradient Boosting Machine (LightGBM), a state-of-the-art ensemble learning algorithm based on leaf-wise decision tree boosting (Ke et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). LightGBM constructs additive predictive models by minimizing a loss function via gradient descent optimization. Its capacity to handle high-dimensional data and model feature interactions makes it particularly suitable for flood risk assessment in heterogeneous urban environments (Wang et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Zhou et al. \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Model training incorporated stratified k-fold cross-validation to reduce overfitting risk and ensure generalizability (Zhang et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Feature importance was first evaluated using gain-based metrics, which reflect each variable\u0026rsquo;s contribution to loss reduction across decision splits.\u003c/p\u003e \u003cp\u003eTo enhance the interpretability of LightGBM outputs, we adopted the SHapley Additive exPlanations (SHAP) framework\u0026mdash;an interpretable machine learning approach grounded in cooperative game theory (Lundberg et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2018\u003c/span\u003e, Lundberg and Lee \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). SHAP assigns a unique value to each feature based on its marginal contribution to the model\u0026rsquo;s prediction for each observation, thus enabling global interpretability, local interpretability and interaction analysis.\u003c/p\u003e \u003cp\u003eBased on the analytical framework proposed by (Lin et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), we designed two sets of comparative modeling experiments. (i) a hydrology-only baseline with conventional predictors (topography, land use, hydrology, meteorology, socioeconomic factors) and without LCZ variables; and (ii) an enhanced model (hydrology\u0026thinsp;+\u0026thinsp;LCZ morphology) that incorporates all variables, including LCZ-based indicators. By comparing the predictive performance (e.g., R\u0026sup2;, RMSE) and SHAP-derived explanatory models between the two models (Wang et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), we can assess whether the urban morphological structure represented by the LCZ has explanatory power beyond conventional predictors to support more spatially targeted and mechanistic flood mitigation strategies.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Result","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Spatial patterns of urban flooding across LCZ\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the spatial distribution of urban flooding kernel density at the community scale, revealing the spatial clustering patterns of flood-prone area across different LCZ types. This spatial analysis uncovers why flood events are not randomly distributed, but instead exhibits strong spatial dependencies influenced by the urban morphological context. In Guangzhou, flood hotspots are primarily concentrated in the central and southwestern urban districts, notably Tianhe, Yuexiu, and Haizhu. Among these, LCZ4 (31 points) and LCZ8 (30 points) exhibit the highest number of flood points, followed by LCZ6 (29 points). These areas are characterized by a high intensity of built-up land, limited green cover, and historically dense development patterns. In Shenzhen, flooding hotspots are predominantly located in the core built-up zones of Futian, Nanshan, and Luohu, which form the city\u0026rsquo;s economic and administrative centers. Among these, LCZ8 has the highest number of flood points at 57, followed by LCZ4 with 52.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Potential flood exposure across LCZ\u003c/h2\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, all three exposure indicators\u0026mdash;population,, population, economic, and built up area\u0026mdash;exhibited marked differences across LCZ types and between the two cities. Overall, Shenzhen displayed consistently higher levels of potential exposure than Guangzhou. The total building exposure in Shenzhen reached approximately 300 km\u0026sup2;, representing a 50% increase over Guangzhou\u0026rsquo;s approximately 200 km\u0026sup2;. Similarly, population exposure in Shenzhen was estimated at 6.5\u0026nbsp;million, compared to 4.2\u0026nbsp;million in Guangzhou\u0026mdash;an increase of approximately 55%. In terms of economic exposure, Shenzhen recorded an estimated 90\u0026nbsp;million CNY, surpassing Guangzhou\u0026rsquo;s 55\u0026nbsp;million CNY by about 63%. These findings reflect Shenzhen\u0026rsquo;s more intensive urban development and demographic concentration, which contribute to its heightened vulnerability to urban flooding.\u003c/p\u003e \u003cp\u003eAcross both cities, LCZ4 and LCZ8 emerged as dominant contributors to potential exposure. In Shenzhen, LCZ4 accounting for approximately 23% of total PBE (70 km\u0026sup2;), 34% of PPE (2.2\u0026nbsp;million people), and 33% of PEE (30\u0026nbsp;million CNY). LCZ8 also contributed substantially, with around 20% of PBE (60 km\u0026sup2;), 28% of PPE (1.8\u0026nbsp;million CNY), and 28% of PEE (25\u0026nbsp;million CNY). Collectively, these two LCZ types represented over 40% of Shenzhen\u0026rsquo;s total exposure, underscoring their strategic relevance for flood risk management. In contrast, Guangzhou exhibited a more spatially diverse exposure profile, with significant contributions from LCZ2 and LCZ1 particularly within the densely populated central urban districts.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe CEI further delineates the risk composition and magnitude across LCZ spatial patterns. A pie chart summarizes the relative contributions of population, building, and economic exposure for each pattern, whereas a box plot reports the central tendency and dispersion of CEI (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Flood risk differs markedly by LCZ. LCZ1 shows the highest risk (CEI\u0026thinsp;=\u0026thinsp;1.06 in Shenzhen; CEI\u0026thinsp;=\u0026thinsp;0.90 in Guangzhou). Regarding composition, in Shenzhen, LCZ1 has the largest population and economic exposure, whereas LCZ5 records the largest building exposure. In Guangzhou, LCZ10 has the largest population exposure, LCZ4 the largest economic exposure, and LCZ5 the largest building exposure. In Shenzhen, LCZ4 and LCZ8 have elevated median CEI values (\u0026asymp;\u0026thinsp;0.25\u0026ndash;0.40) with maxima\u0026thinsp;\u0026gt;\u0026thinsp;0.6, indicating morphology-driven, widespread exposure in dense urban cores. In Guangzhou, although overall CEI values are lower, LCZ1 and LCZ2 exhibit larger interquartile ranges and high-risk outliers, suggesting localized hotspots plausibly linked to aging infrastructure or rapid redevelopment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Driving factors of urban flood risk\u003c/h2\u003e \u003cp\u003eBy comparing a baseline model with an enhanced model, this study highlights the critical importance of LCZ morphological characteristics in predicting urban flood (Supplementary Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e). In Guangzhou, the baseline model achieved a strong fit with an R\u0026sup2; of 0.95, a MAE of 0.23, and a MSE of 0.15. After incorporating LCZ proportions into the feature set, the enhanced model improved the R\u0026sup2; to 0.97 and reduced the MAE by more than 50% to 0.10, and decreased the MSE to 0.13. In Shenzhen, the baseline model yielded an R\u0026sup2; of 0.944, with both MAE and MSE at 0.16. The enhanced model improved the R\u0026sup2; to 0.957 and slightly reduced the MAE and MSE to 0.15. These results confirm that LCZ-based morphological descriptors significantly enhance model accuracy, particularly in cities characterized by complex spatial heterogeneity.\u003c/p\u003e \u003cp\u003eWe further analyzed the importance ranking of each predictive variable under the baseline scenario. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, POP consistently emerged as the most influential factor in both cities, emphasizing the central role of human exposure in shaping flood vulnerability patterns. In Guangzhou, the next most important variables were PD and TRI, pointing to the combined influence of urban imperviousness and topographic variation on flood hotspot. Conversely, in Shenzhen, DR was the second-ranked factor, followed by PD, suggesting that hydrological infrastructure plays a more decisive role in flood modulation. A cumulative contribution analysis revealed that flood risk prediction in both cities is dominated by a small subset of variables. In Guangzhou, the top three variables\u0026mdash;POP, PD, and TRI\u0026mdash;together accounted for approximately 58% of the model\u0026rsquo;s total explanatory power, while the top five contributed over 74%. In Shenzhen, the top three drivers\u0026mdash;POP, DR, and PD\u0026mdash;explained over 60%, with the top five exceeding 80% of total importance. These cumulative patterns suggest that targeted interventions focusing on the highest-contributing factors may yield substantial improvements in flood risk mitigation strategies.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the enhanced model, LCZ indicators were introduced to quantify the role of urban morphology alongside conventional socio-environmental drivers in shaping urban flood risk. The results reveal a clear spatial heterogeneity between Guangzhou and Shenzhen in terms of the dominant flood-driving mechanisms. In Guangzhou, morphological factors exhibited strong predictive power. Notably, LCZ4 emerged as a key driver, with a mean |SHAP value| of approximately 0.6, surpassing traditional predictors such as POP, PD, and TRI. LCZ-related variables collectively accounted for 30.8% of the model\u0026rsquo;s explanatory importance, with open-form zones contributing 21.1%, mixed forms 5.1%, and compact types only 1.3%. SHAP dependency plots further highlighted nonlinear threshold effects: LCZ4 triggered a sharp increase in flood risk when its coverage exceeded 20% within the buffer zone (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec), while LCZ8 also exhibited elevated flood contributions beyond a 10% threshold (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eg). Meanwhile, POP displayed a near-linear positive relationship with flood density, underscoring exposure-driven risk accumulation (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ed).\u003c/p\u003e \u003cp\u003eIn Shenzhen, by contrast, the model indicated that socioeconomic and infrastructural variables played a more dominant role. POP alone yielded a mean |SHAP value| close to 0.9, far exceeding any single morphological predictor. Drainage-related variables, particularly DR, exhibited a strong negative correlation with flood risk, reinforcing the mitigating role of drainage infrastructure in high-density urban contexts. While LCZ predictors still contributed 21.1% to the overall model, their influence was relatively muted: open-form types explained 11.6%, mixed forms 5.7%, and compact types 4.9% (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb). SHAP dependency analysis in Shenzhen also revealed nonlinear and threshold-based dynamics, albeit with city-specific nuances. Similar to Guangzhou, LCZ4 retained a positive association with flood risk, while POP remained the most consistent positive predictor. Additionally, TWI showed a non-monotonic pattern: SHAP values increased rapidly for TWI\u0026thinsp;\u0026lt;\u0026thinsp;5, peaked around TWI\u0026thinsp;\u0026asymp;\u0026thinsp;10, and declined thereafter (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003en), suggesting that moderately wet zones are more flood-prone than both arid and excessively saturated areas.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Advantages of LCZ over traditional land use and 3D morphology\u003c/h2\u003e \u003cp\u003eCompared to traditional LUCC classification, LCZ morphology can be used to develop more accurate urban landscape classification maps. The results indicate that LCZ exhibits a greater number of blocks than land use classification, likely because the LCZ method simultaneously considers both land cover characteristics and geometric features (Jiang et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). LCZ exhibits a higher edge density (ED) than LUCC classification while showing a lower largest patch index (LPI), suggesting that LCZ provides more detailed boundary information and captures more complex morphological details.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study compares the results of (Wang et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) in characterizing urban three-dimensional morphology, selecting four three-dimensional building indicators for comparison: Building Surface Coverage (BSC), Mean Building Height (MBH), Mean Building Volume (MBV), and Building Compactness (BCD). In both cities, LCZ demonstrated the highest relative contribution rates (30.8% and 21.7%), surpassing the combined significance of the four independent three-dimensional building metrics. This indicates that the LCZ-based morphological classification integrates building height, density, and surface characteristics\u0026mdash;providing stronger explanatory power for urban storm flooding than any single three-dimensional variable. Thus, in flood modeling, LCZ functions more as a \u0026ldquo;composite urban form\u0026ndash;surface response unit,\u0026rdquo; serving as a simplified 2D and 3D metric approach (Zhou et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). It is important to note that 3D metrics still capture detailed variations within LCZ types. At finer scales or in specific scenarios (such as different height combinations within the same LCZ), three-dimensional morphological information remains complementary (Qin et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Zhou et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.2. LCZ and flood risk differentiation\u003c/h2\u003e \u003cp\u003eThis study demonstrates that the integration of LCZ classifications into flood risk models substantially enhances both predictive accuracy and mechanistic interpretability. Embedding LCZ metrics within a SHAP-interpretable LightGBM framework revealed that urban morphology\u0026mdash;in conjunction with traditional socio-environmental drivers\u0026mdash;plays a critical role in shaping the spatial distribution of flood exposure and risk.\u003c/p\u003e \u003cp\u003eSpatial analysis showed that flood hotspots are consistently concentrated within the LCZ4 and LCZ8 zones, which not only registered the highest number of flood incidents but also exhibited elevated KDE. Across both cities, LCZ2, LCZ4, and LCZ8 emerged as the most flood-exposed typologies across three dimensions\u0026mdash;building footprint, population, and economic, which is consistent with Zou's findings (Zou et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For example, in Shenzhen, LCZ4 alone accounted for nearly one-third of total exposure, while LCZ8 contributed an additional 20\u0026ndash;28%, indicating that morphological configurations act as spatial amplifiers of urban flooding risk, co-locating both hazards and vulnerable assets. These exposures were further corroborated by the CEI, where LCZ4 and LCZ8 consistently showed higher median and upper quantile values\u0026mdash;some exceeding 0.6, indicating not just frequent flooding, but high potential for damage when flooding occurs. In these areas, urban form, population concentration, and infrastructure constraints interact to exacerbate the impact of flooding (Zhou et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA comparative analysis between Guangzhou and Shenzhen reveals different urbanization tracks shape different urban flood risk mechanisms. In Guangzhou, LCZ-related variables explained over 30%, highlighting the structural importance of urban morphology in shaping hydrological behavior. In contrast, Shenzhen's flood risk was driven primarily by POP and DR, with LCZ variables contributing only 21% to the model\u0026mdash;suggesting that in newer, planned urban environments, risk profiles may be dominated by infrastructure. This is because Shenzhen, as a typical newly planned city, has a highly planned and standardized urban spatial layout, with clearly delineated functional zones and relatively uniform architectural forms, and the flood risk is more dependent on whether the infrastructure configuration can match the population growth and development intensity brought about by the rapid expansion of the city. Notably, Shenzhen\u0026rsquo;s drainage system\u0026mdash;including its widespread adoption of separate stormwater and sewage systems\u0026mdash;is among the most advanced in China. Its early implementation of comprehensive urban drainage master plans and high-standard engineering design has substantially enhanced its resilience to pluvial flooding.\u003c/p\u003e \u003cp\u003eThe results of the SHAP analysis reveal that LCZ4 (open high rise) and LCZ8 (large scale low rise) are not only the main bearers of the high-risk areas, but also significant morphological factors driving urban floods. These zones are typically characterized by high imperviousness, low or absent vegetation cover, and are often covered by extensive areas of hard paving, buildings, and roads. During heavy rainfall, these surfaces have minimal ability to retain or infiltrate stormwater, leading to the rapid accumulation of surface runoff. When the urban drainage system is limited in capacity or poorly maintained, localized runoff aggregation can easily result in urban flooding with high short-term intensity. This finding aligns with recent studies (Huang et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Wang et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) that demonstrate a high proportion of impervious surfaces can significantly increase the peak runoff rate and likelihood of inundation once rainfall intensity surpasses a certain threshold. In contrast, compact high-rise and mid-rise zones (LCZ1, LCZ2) ranked lower despite their higher building density and population concentration. This is likely due to their superior drainage infrastructure and vertical runoff management capabilities. These areas are typically the administrative, commercial, and economic centers of cities, so their drainage systems are designed to higher standards, updated more frequently, maintained more adequately, and have greater capacity for storage and rapid discharge (Wu et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese findings underscore the importance of context-sensitive flood risk assessment, particularly in rapidly urbanizing regions. The explanatory power of LCZ variables is not uniform across cities, reflecting differences in infrastructure development, urban design, and land governance. While LCZ typologies offer strong predictive utility in morphologically heterogeneous cities like Guangzhou, their contribution is more muted in newer cities like Shenzhen, where infrastructure homogeneity dampens morphological contrasts. Nevertheless, even in Shenzhen, LCZ-based modeling captures structural vulnerabilities in industrial (LCZ8) and open mid-rise (LCZ4) zones\u0026mdash;areas with large impervious surfaces and high-value assets. Recognizing these zones as morphology-infrastructure risk interfaces can inform spatially differentiated adaptation strategies (Zhang et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Integrating LCZ typologies into flood responsive urban planning\u003c/h2\u003e \u003cp\u003eIncorporating LCZs into flood risk governance provides a framework for understanding urban hydrometeorological vulnerability beyond land use-centered assessments of urban form (Aslam et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e, Fan et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Traditionally considered as a tool for analyzing thermal heterogeneity, LCZs encompass critical urban morphological features that control flood behavior, such as impermeability, built form, surface openness, and spatial continuity. This study demonstrates that LCZs function not only as descriptive classifications, but also as operational proxies for identifying structural vulnerabilities and exposure concentrations (Dong et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Yang et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The predictive enhancement achieved by integrating LCZ data into machine learning models substantiates their analytical value, bridging the gap between urban form and hydrological function. Zones such as LCZ4 and LCZ8 consistently emerge as compound risk hotspots, characterized by the convergence of flood susceptibility and concentrated assets or population\u0026mdash;a spatial phenomenon that traditional land-use categories fail to capture with comparable specificity.\u003c/p\u003e \u003cp\u003eFrom a planning and governance perspective, LCZ-based assessment facilitates a shift from reactive to anticipatory urban flood management. By identifying high-risk morphological zones\u0026mdash;particularly LCZ4 and LCZ8\u0026mdash;prior to flood events, cities can strategically prioritize investments in drainage infrastructure and green adaptation measures, integrate LCZ mapping into zoning regulations, redevelopment guidelines, and early warning systems, and establish morphology-specific design standards for critical infrastructure (Grobicki et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, Parizi et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Such spatially targeted, form-sensitive interventions enable more cost-effective and socially equitable adaptation, especially in resource-limited urban contexts. Unlike conventional flood zoning, LCZ-informed strategies are rooted in the structural logic of urban form, accounting for the co-evolution of the built environment, infrastructure systems, and exposure patterns. In summary, the integration of LCZs into flood risk governance transcends their descriptive origin. It represents a framework that is scalable, transferable, and interpretable across various cities and regions. Future research should further explore their integration with dynamic exposure data and urban climate models to develop truly morphology-sensitive resilience strategies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Limitations\u003c/h2\u003e \u003cp\u003eDespite the potential of this framework to be applied to urban stormwater management, the study has some limitations. First, the inundation point dataset used in this study lacks critical attributes such as inundation depth and spatial extent, which somewhat limits the ability to assess the severity and impact of individual flooding events. Second, the current analysis is based on static representations of urban morphology and flood events, without incorporating the temporal evolution of LCZ types or flood exposure. Incorporating longitudinal LCZ mapping and time-series flood exposure data would provide a more nuanced understanding of how urban morphological trajectories influence risk accumulation and persistence. Third, LCZs were not integrated into physically based hydrodynamic models (e.g., SWMM or MIKE URBAN), which limits the mechanistic understanding of how morphological configurations translate into real-world flood dynamics (Mondal et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Xu et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Future research should aim to couple LCZ-informed urban morphology with physical process models to validate and enrich the causal interpretations of flood risk mechanisms.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study demonstrates the critical value of integrating LCZ classifications into urban flood risk modeling, revealing how urban morphological structures fundamentally shape flood occurrence, exposure, and driving mechanisms across megacities. Key findings indicate that flood events are not spatially random but are strongly patterned by urban form, with open-form LCZs exhibiting elevated event density and potential exposure in both Guangzhou and Shenzhen. The enhanced predictive power of LCZ features\u0026mdash;accounting for over 30% of the explanatory strength in Guangzhou\u0026mdash;underscores their role as structural amplifiers of flood risk, especially where impervious surfaces, infrastructural deficits, and concentrated human activity coalesce. In addition, a comparative analysis of Guangzhou and Shenzhen shows that different urbanization LCZ patterns influence different urban flooding mechanisms. Guangzhou\u0026rsquo;s flood dynamics are more sensitive to urban form, while Shenzhen's risk is more influenced by human exposure and drainage infrastructure, reflecting divergent urbanization pathways and planning baselines. SHAP analysis further revealed nonlinear and threshold effects for key LCZs and environmental drivers, offering interpretable insights for targeted mitigation. Ultimately, our findings support a scalable, interpretable, spatially accurate and morphology-based flood management framework for managing urban hydrometeorological risks in the face of continued urban sprawl and climate intensification.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e \u003cb\u003eAuthor information\u003c/b\u003e:\u003c/h2\u003e \u003cp\u003e \u003cstrong\u003eAuthors and Affiliations\u003c/strong\u003e \u003cp\u003eSchool of Civil Engineering, Sun Yat-sen University, Tangjiawan, 519082, Zhuhai, Guangdong, China\u003c/p\u003e \u003cp\u003eYongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang\u003c/p\u003e \u003cp\u003eGuangdong Provincial Key Laboratory for Marine Civil Engineering, Sun Yat-sen University (Zhuhai Campus), Tangjiawan, Zhuhai 519082, China\u003c/p\u003e \u003cp\u003eYongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang\u003c/p\u003e \u003cp\u003eGuangdong Engineering Technology Research Center of Water Security Regulation and Control for Southern China, Sun Yat-sen University, Guangzhou 510275, China\u003c/p\u003e \u003cp\u003eYongheng Wang, Qingtao Zhang, Kairong Lin, Zhuochao Zhang\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eConflicts of Interest:\u003c/h2\u003e \u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis work was financially supported by the National Natural Science Foundation of China (31270748 and 31470707), Project funded by the Hydrological Bureau of Guangdong Province (440001-2023-10716), and the Guangzhou Bureau of Hydrology project \u0026ldquo;Research on the mechanism of hydro-ecological dynamics in a typical river network area\u0026rdquo; (SWYS2023F050).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAslam A, Rana IA, Bhatti SS (2021) The spatiotemporal dynamics of urbanisation and local climate: A case study of Islamabad, Pakistan. Environ Impact Assess Rev 91:106666\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai H, Ming Y, Liu Q, Huang C (2022) A dataset of rainstorm in China based on GPM precipitation product during 2001\u0026ndash;2019. China Scientific Data 7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBehboudian M, Anamaghi S, Mahjouri N, Kerachian R (2023) Enhancing the resilience of ecosystem services under extreme events in socio-hydrological systems: A spatio-temporal analysis. J Clean Prod 397:136437\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChan FKS, Griffiths JA, Higgitt D, Xu S, Zhu F, Tang Y-T, Xu Y, Thorne CR (2018) Sponge City in China\u0026mdash;A breakthrough of planning and flood risk management in the urban context. Land Use Policy 76:772\u0026ndash;778\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Ma W, Shao Y, Wang N, Yu Z, Li H, Hu Q (2025) The impacts and thresholds detection of 2D/3D urban morphology on the heat island effects at the functional zone in megacity during heatwave event. Sustainable Cities Soc 118:106002\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Xu C, Ge Y, Zhang X, Zhou Y (2024) A 100 m gridded population dataset of China's seventh census using ensemble learning and big geospatial data. Earth Syst Sci Data 16(8):3705\u0026ndash;3718\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChing J, Mills G, Bechtel B, See L, Feddema J, Wang X, Ren C, Brousse O, Martilli A, Neophytou M, Mouzourides P, Stewart I, Hanna A, Ng E, Foley M, Alexander P, Aliaga D, Niyogi D, Shreevastava A, Bhalachandran P, Masson V, Hidalgo J, Fung J, Andrade M, Baklanov A, Dai W, Milcinski G, Demuzere M, Brunsell N, Pesaresi M, Miao S, Mu Q, Chen F, Theeuwes N (2018) WUDAPT: An Urban Weather, Climate, and Environmental Modeling Infrastructure for the Anthropocene %J Bulletin of the American Meteorological Society. 99(9), 1907\u0026ndash;1924\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDemuzere M, Kittner J, Martilli A, Mills G, Moede C, Stewart ID, van Vliet J, Bechtel B (2022) A global map of local climate zones to support earth system modelling and urban-scale environmental science. Earth Syst Sci Data 14(8):3835\u0026ndash;3873\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong W, Jiang R, Dong Y, Qu A, Yuan Y (2025) Relationship between LCZ and physical activity in residential areas: A mediating role of perceptions of heat risks in climate change. Urban Clim 61:102425\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan PY, He Q, Tao YZ (2023) Identifying research progress, focuses, and prospects of local climate zone (LCZ) using bibliometrics and critical reviews. Heliyon 9(3), e14067\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong L, Zhang X, Guo Z, Winston R, Tao S, Smith J (2025) Urban flood resilience assessment under compounding risk: joint impacts of precipitation and river level. Sustainable Cities Soc 130:106569\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrobicki A, MacLeod F, Pischke F (2015) Integrated policies and practices for flood and drought risk management. Water Policy 17:180\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang F, Jiang S, Zhan W, Bechtel B, Liu Z, Demuzere M, Huang Y, Xu Y, Ma L, Xia W, Quan J, Jiang L, Lai J, Wang C, Kong F, Du H, Miao S, Chen Y, Chen J (2023) Mapping local climate zones for cities: A large review. Remote Sens Environ 292:113573\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang J, Huang Z, Liu W (2025) Combining the WRF model and LCZ scheme to assess spatiotemporal variations of thermal comfort in Shenzhen's built-up areas. Sustainable Cities Soc 122:106252\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang S, Gan Y, Chen N, Wang C, Zhang X, Li C, Horton DE (2024) Urbanization enhances channel and surface runoff: A quantitative analysis using both physical and empirical models over the Yangtze River basin. J Hydrol 635:131194\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIPCC (2023) Synthesis Report of the IPCC Sixth Assessment Report, Intergovernmental Panel on Climate Change Interlaken, Switzerland\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJavadpoor M, Sharifi A, Gurney KR (2024) Mapping the relationship between urban form and CO2 emissions in three US cities using the Local Climate Zones (LCZ) framework. J Environ Manage 370:122723\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang R, Xie C, Man Z, Afshari A, Che S (2023) LCZ method is more effective than traditional LUCC method in interpreting the relationship between urban landscape and atmospheric particles. Sci Total Environ 869:161677\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJun-Neng W, Nian-Xiu QIN, Tong J, Bu-Da SU (2022) Interpretation of IPCC AR6: impacts and adaptations of climate change on cities, settlements and key infrastructure. Adv Clim Change Res 18(4):433\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe G, Meng Q, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu T-Y (2017) LightGBM: A Highly Efficient Gradient Boosting Decision Tree\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi C, Liu M, Hu Y, Wang H, Zhou R, Wu W, Wang Y (2022a) Spatial distribution patterns and potential exposure risks of urban floods in Chinese megacities. J Hydrol 610:127838\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi M, Wang Y, Rosier J, Verburg P, van Vliet J (2022b) Global maps of 3D building structure for urban morphology analysis. DataverseNL\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin J, Zhang W, Wen Y, Qiu S (2023) Evaluating the association between morphological characteristics of urban land and pluvial floods using machine learning methods. Sustainable Cities Soc 99:104891\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin Z, Xu H, Han L, Zhang H, Peng J, Yao X (2024) Day and night: Impact of 2D/3D urban features on land surface temperature and their spatiotemporal non-stationary relationships in urban building spaces. Sustainable Cities Soc 108:105507\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg S, Erion G, Lee S-I (2018) Consistent Individualized Feature Attribution for Tree Ensembles\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg S, Lee S-I (2017) A Unified Approach to Interpreting Model Predictions\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuo K, Zhang X (2022) Increasing urban flood risk in China over recent 40 years induced by LUCC. Landsc Urban Plann 219:104317\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMondal K, Ghosh M, Karmakar S (2025) Global sensitivity analysis in a complex 1D-2D coupled hydrodynamic model: Flood hazard and resilience perspectives over an urban catchment. Sustainable Cities Soc 124:106279\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParizi SM, Taleai M, Sharifi A (2024) A spatial evaluation framework of urban physical resilience considering different phases of disaster risk management. Nat Hazards 120(14):13041\u0026ndash;13076\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng S, Ding Y, Liu W, Li Z (2019) Earth Syst Sci Data 11(4):1931\u0026ndash;19461 km monthly temperature and precipitation dataset for China from 1901 to 2017\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQi W, Ma C, Xu H, Xu K, Lian J (2025) Flood mitigation performance of low impact development practice in a coastal city from the perspective of catchment scale. J Hydrol 649:132466\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQin Y, Kang J, Zhou H, Xu S, Li G, Li C, Tan W (2025) Assessment of the impact of urban block morphological factors on carbon emissions introducing the different context of local climate zones. Sustainable Cities Soc 119:106073\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahmani N, Sharifi A (2025) Urban heat dynamics in Local Climate Zones (LCZs): A systematic review. Build Environ 267:112225\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSoh QY, Acha S, Shah N, O\u0026rsquo;Dwyer E (2025) Simultaneous design and control optimisation of combined rainwater harvesting and flood mitigation systems. Resources, Conservation and Recycling 222, 108459\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStewart ID, Oke TR (2012) Local Climate Zones for Urban Temperature Studies. Bull Am Meteorol Soc 93(12):1879\u0026ndash;1900\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun S, Zhai J, Li Y, Huang D, Wang G (2020) Urban waterlogging risk assessment in well-developed region of Eastern China. Physics and Chemistry of the Earth, Parts A/B/C 115, 102824\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y, Li C, Liu M, Cui Q, Wang H, Lv J, Li B, Xiong Z, Hu Y (2022) Spatial characteristics and driving factors of urban flooding in Chinese megacities. J Hydrol 613:128464\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y, Zhang Q, Lin K, Liu Z, Liang Y-s, Liu Y, Li C (2024) A novel framework for urban flood risk assessment: Multiple perspectives and causal analysis. Water Res 256:121591\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y, Zhang Q, Zhang J, Lin K (2025) Impact of 2D and 3D factors on urban flooding: Spatial characteristics and interpretable analysis of drivers. Water Res 280:123537\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu P, Wang T, Wang Z, Song C, Chen X (2025a) Impact of Drainage Network Structure on Urban Inundation Within a Coupled Hydrodynamic Model. Water 17:990\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu R, Fang X, Liu S, Peng H, Zhao H, Zhou H, Zhao X, Yang H, Yan J, Meng Q (2025b) Carbon footprint mapping in LCZs: An integrated view of urban thermal environments through coupling spatial and human activities. Sustainable Cities Soc 128:106435\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu W, Jamali B, Marshall L, Deletic A, Zhang K (2025c) A water sensitive urban design (WSUD) planning framework for catchment-scale urban pluvial flood mitigation targets. Water Res 285:124095\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu Y, Lin K, Hu C, Chen X, Zhang J, Mingzhong X, Xu C-Y (2024) Uncovering the Dynamic Drivers of Floods Through Interpretable Deep Learning. Earth's Future 12\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Huang X (2021) The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019. Earth Syst Sci Data 13(8):3907\u0026ndash;3925\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Yu W, Baklanov A, He B-J, Ge Q (2025) Mainstreaming the local climate zone framework for climate-resilient cities. Nat Commun 16\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan B, Zhou L, Hu F, Wei C (2024) Effects of 2D/3D urban morphology on land surface temperature: Contribution, response, and interaction. Urban Clim 53:101791\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang H, Gao J, Zhao J, Guo F, Bai J, Wang Z, Zhu P (2025) Applicability of local climate zones in assessing urban heat risk - a survey of coastal city. Cities 164:106068\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Zhang Q, Wang Y, Wu X, Zhang Q (2026) Global lipid production over submarginal lands can offset anthropogenic carbon emissions. Resources, Conservation and Recycling 225, 108642\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Q, Wu Z, Cao Z, Guo G, Hui Z, Li C, Tarolli P (2023) How to develop site-specific waterlogging mitigation strategies? Understanding the spatial heterogeneous driving forces of urban waterlogging. J Clean Prod 422:138595\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou S, Geng X, Zhao J, Hei J, Wu T, Chen Z, Wu Z (2025) An LCZ-based machine learning framework for revealing spatial heterogeneity of thermal comfort in high-density areas: Enhancing explainability and fine-grid scale resolution. Sustainable Cities Soc 133:106873\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou S, Wang Y, Jia W, Wang M, Wu Y, Qiao R, Wu Z (2023) Automatic responsive-generation of 3D urban morphology coupled with local climate zones using generative adversarial network. Build Environ 245:110855\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou S, Zhang D, Wang M, Liu Z, Gan W, Zhao Z, Xue S, M\u0026uuml;ller B, Zhou M, Ni X, Wu Z (2024) Risk-driven composition decoupling analysis for urban flooding prediction in high-density urban areas using Bayesian-Optimized LightGBM. J Clean Prod 457:142286\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou B, Nie Y, Liu R, Wang M, Li J, Fan C, Zhou X (2024) Assessing the Impact of Urban Morphologies on Waterlogging Risk Using a Spatial Weight Naive Bayes Model and Local. Clim Zones Classif 16(17):2464\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"water-resources-management","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"warm","sideBox":"Learn more about [Water Resources Management](https://www.springer.com/journal/11269)","snPcode":"11269","submissionUrl":"https://submission.nature.com/new-submission/11269/3","title":"Water Resources Management","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Urban flooding, Local Climate Zones, Machine Learning, SHAP, Exposure risk, Adaptive flood management","lastPublishedDoi":"10.21203/rs.3.rs-9078373/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9078373/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eUrban pluvial flooding is intensifying under rapid urbanization and climate change, yet most data-driven assessments underrepresent the role of urban morphology. We tentatively treat the Local Climate Zone (LCZ) scheme as a standardized morphological proxy rather than a purely hydrological variable. This study introduces an innovative analytical framework that integrates LCZ-based morphological indicators into a LightGBM machine learning model, enhanced by Shapley Additive Explanations (SHAP) for improved interpretability. Using data derived from Guangzhou and Shenzhen, we constructed two model scenarios: a baseline model employing traditional socio-environmental variables and an enhanced model incorporating LCZ typologies. The enhanced model demonstrated a substantial improvement in predictive accuracy, particularly in Guangzhou, where LCZ-related factors contributed over 30% to the model's importance, with a higher relative contribution rate than standalone 3D building metrics. Compared with conventional land-use classification, LCZ produced a markedly finer-grained urban form. Besides, SHAP analyses further revealed distinct threshold effects associated with specific land coverage levels. By coupling standardized morphology with interpretable machine learning, this framework is scalable across cities and provides actionable guidance for adaptive planning It prioritizes infrastructure improvements\u0026mdash;such as street renovations and permeable upgrades\u0026mdash;in areas exhibiting the highest morphological sensitivity.\u003c/p\u003e","manuscriptTitle":"An LCZ-Based Machine Learning Reveals Differences in Coastal High-Density Urban Flood Risk: Enhancing Interpretability and Simplifying Morphology","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-16 13:47:44","doi":"10.21203/rs.3.rs-9078373/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-08T13:43:50+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"Water Resources Management","date":"2026-03-24T14:55:07+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-10T04:35:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Water Resources Management","date":"2026-03-09T23:01:04+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"water-resources-management","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"warm","sideBox":"Learn more about [Water Resources Management](https://www.springer.com/journal/11269)","snPcode":"11269","submissionUrl":"https://submission.nature.com/new-submission/11269/3","title":"Water Resources Management","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"505ef286-52d5-4479-aa37-82b23c402ca3","owner":[],"postedDate":"April 16th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-16T13:47:44+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-16 13:47:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9078373","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9078373","identity":"rs-9078373","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}