Study area

The study area consisted of a typical high-density urban community located in the eastern region. The community is characterized by high building density. Residential buildings, road networks, paved surfaces, auxiliary hardened spaces, green patches, and small water bodies are interspersed within the area. This configuration creates a complex surface cover structure with a high proportion of impervious surfaces. The overall terrain is relatively flat, although short slopes and micro-topographic variations occur locally. After rainfall events, surface flow paths remain short, which leads to rapid runoff formation. Localized water accumulation risk is therefore relatively prominent. Climatic conditions show clear seasonal variation, with rainfall concentrated during the wet season. Short-duration rainfall events place continuous pressure on the community’s drainage and storage capacity.

The case area has three representative characteristics. First, high-density construction coexists with a large proportion of hardened surfaces, which reflects common structural features of stormwater regulation challenges in urban communities. Second, green-space patches are relatively small and spatially dispersed, which allows identification of how landscape spatial structure influences runoff processes. Third, the community boundary is clearly defined. This condition facilitates high-resolution digital landscape identification, monitoring point deployment, and scenario-based simulation analysis. The study area therefore provides an appropriate setting for evaluating runoff regulation effects and developing landscape optimization strategies at the community scale.

Data sources

To improve the clarity and organization of the data description, the data sources were categorized into four groups: landscape identification data, meteorological data, observed runoff data, and spatial preprocessing data.

This study selected typical high-density urban communities in eastern China as the research area, and the data mainly used publicly available authoritative datasets, including the NLCD (2022 edition) published by the United States Geological Survey (USGS), long-term meteorological data from municipal meteorological stations in the research area, and runoff data observed by the research team in the field. NLCD data can be accessed through the USGS Earth Explorer platform (access link: https://earthexplorer.usgs.gov/). It has an original spatial resolution of 30 meters and containing 16 types of land cover. In this study, it was cropped and resampled to a resolution of 1 meter for cross validation with GIS interpreted landscape data. Meantime, the impermeable surface and green space data extracted by NLCD were input into the InVEST model, and compared with the runoff depth simulation results based on GIS interpreted data. In this study, the NLCD data were primarily used for cross-validation and result comparison. Digital landscape classification at the community scale was mainly based on the interpretation of high-resolution remote sensing imagery.

The evaluation of landscape classification relies on two core datasets: the 0.5m high-resolution remote sensing images obtained in 2022 (used for GIS interpretation to generate 8 types of landscape data), and the NLCD 2022 version of land cover data. The overall accuracy (OA = 92.3%) and Kappa coefficient (K = 0.89) were calculated through confusion matrix to verify the classification accuracy, ensuring that the landscape classification results meet the input requirements of the model.

Meteorological data is obtained through the official data sharing platform of the municipal meteorological department within a 20 km radius of the research area. The time resolution is monthly, the time range is from 2018 to 2023, and the time step is monthly. The data includes annual precipitation (1215 mm), monthly average precipitation (195 mm from June to September in the rainy season and 45 mm from December to February in the dry season), and annual average rainfall duration (1200 hours). They are used to calculate the rainfall erosivity factor (R) in the InVEST model and the potential runoff parameters in the runoff depth derivation formula, ensuring that the model reflects the long-term hydro meteorological conditions of the study area. The 42 sets of measured runoff data used for model calibration are rainfall event scale data collected by the research team at five monitoring points in the community from 2022 to 2023 (with a time step of the entire process of a single rainfall event), covering three scenarios: light rain, moderate rain, and heavy rain. It should be noted that the land use data in this study is static data (based on 2022 remote sensing images and NLCD 2022 version), and does not involve multi-period land use changes. The analysis of temporal changes is limited to climate factors (monthly and inter-annual fluctuations in rainfall), and does not include the temporal dynamic evolution of land use. Five monitoring points were installed along major hardened flow paths, green-space transition zones, and local low-lying units. This configuration ensured coverage of different surface types and runoff response locations. The monitoring layout captured runoff variation under the spatial heterogeneity of the community and provided a data basis for parameter calibration and model validation.

All datasets were spatially coupled through GIS, and a unified CGCS2000 coordinate system and 1m × 1m grid resolution were used to construct an integrated database of “terrain landscape soil meteorology observation.”

Digital landscape classification and land-cover definition

The term digital landscape in this study does not directly adopt the original categories from general land-cover databases. Instead, it refers to a functional spatial representation system developed for community-scale runoff regulation analysis. The system was established through the interpretation of high-resolution remote sensing imagery, terrain information, and surface-cover characteristics. Conventional land-cover classifications mainly serve regional-scale surface type identification. Their category boundaries are relatively coarse and cannot adequately capture micro-scale paved structures, green patch variation, or auxiliary hardened spaces within communities. In contrast, digital landscape classification emphasizes differences in the hydrological roles of surface units during runoff generation, storage, infiltration, and concentration. It therefore considers not only cover attributes but also spatial organization and hydrological response characteristics.

Based on the internal surface structure of the community, eight categories of digital landscape units were defined: building-covered space, roads and linear paved space, plazas and auxiliary hardened areas, tree-covered green space, shrub green space, grassland, water bodies, and bare soil. Building-covered space refers to residential and public-service buildings together with the permanent roof projection area. Roads and linear paved space include vehicular lanes, pedestrian paths, and other continuous linear hardened corridors. Plazas and auxiliary hardened areas include paved activity spaces, parking areas, and hardened surfaces surrounding buildings. Tree-covered green space, shrub green space, and grassland distinguish vegetation layers that differ in rainfall interception, root-mediated water conduction, and surface infiltration capacity. Water bodies include landscape water surfaces and stable water storage units. Bare soil represents temporarily exposed ground without pavement or stable vegetation cover.

Compared with the NLCD, this classification system differs in two main aspects. First, it operates at a finer spatial scale and can identify small hardened units and transitional green spaces within the community. Second, the classification criteria are closely aligned with runoff regulation analysis. This structure provides more suitable inputs for impervious surface connectivity analysis, green-patch aggregation measurement, and scenario-based optimization. NLCD data were used only for cross-validation and simulation result comparison and did not replace the community-scale digital landscape classification.

Green-space types were defined using a hierarchical structure. Tree-covered green space corresponds to vegetation units with clear canopy coverage and well-developed root systems. Shrub green space refers to mid- and low-layer woody vegetation units. Grassland represents areas dominated by herbaceous surface vegetation. These three categories differ significantly in rainfall interception capacity, root water conduction, surface roughness, and soil disturbance. They were therefore treated separately in model parameterization and landscape pattern analysis rather than being merged into a single green-space category.

The definitions of the eight digital landscape units are presented in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Definition of digital landscape units.

https://doi.org/10.1371/journal.pone.0352335.t001

Model framework construction and core algorithm design

This study adopted the SDR module of the InVEST model. The selection was based on the research objective. The study aimed to evaluate the relative effects of landscape structure on surface runoff regulation at the community scale rather than to identify urban flood inundation risk areas. The Urban Flood Risk Mitigation module is more suitable for analyses related to stormwater inundation risk, exposed flooded units, and disaster mitigation demand. Its outputs primarily emphasize flood-risk reduction capacity. In contrast, this study focused on the combined influence of landscape patch type, spatial configuration, slope conditions, and surface cover differences on runoff generation, storage, and transport. Therefore, a process-based module capable of integrating terrain factors, land-cover characteristics, soil attributes, and spatial structural information was required.

The SDR module was originally developed to represent surface material transport and retention processes. However, its parameter system effectively reflects the combined influence of slope gradient, slope length, surface cover conditions, soil erodibility, and surface protection factors on surface flow accumulation and transport capacity. At the community scale, these factors are closely associated with runoff generation processes. In this study, SDR outputs were integrated with infiltration correction, potential runoff derivation, and landscape pattern indices to construct an indirect quantification pathway for runoff regulation effects. This approach does not equate sediment transport with runoff processes. Instead, it uses the module’s ability to characterize spatial resistance, transport sensitivity, and slope-response dynamics to represent differences in runoff regulation under varying community landscape structures.

To reduce potential bias caused by a single proxy variable, observed runoff data were further introduced for parameter calibration. Landscape pattern indices were also analyzed together with simulation outputs. This strategy enhanced both the applicability of the model at the community scale and the interpretability of the results.

The evaluation framework and scenario simulation in this study serve different functions. The evaluation framework is used for community-scale baseline identification, parameter calibration, runoff regulation quantification, and extraction of key spatial factors, forming the foundational structure of the entire analysis. The scenario simulation is built upon this framework and is used to compare runoff response differences under different optimization strategies by adjusting green space coverage, patch aggregation, and spatial configuration. Therefore, the former is designed for baseline assessment and mechanism identification, while the latter focuses on optimization evaluation and pathway comparison.

The runoff regulation model developed in this study is based on GIS spatial analysis for data processing and integrates the ecological process simulation functionality of the InVEST-SDR module. Using 22 core formulas, the model implements a complete logical chain from landscape data interpretation to runoff regulation efficiency quantification [19]. The model input layer uses a 1 m resolution Digital Elevation Model (DEM) and 0.5 m high-resolution remote sensing imagery. GIS is employed to extract topographic parameters and classify landscape types. Slope (, in radians) is calculated based on the elevation change rate within a 3 × 3 grid window using the vector composition method:

(1)

where and represent the elevations of neighboring grid cells in the x and y directions for thse target cell (), and [20]. To meet the InVEST model requirements for topographic input, slope in radians is converted to degrees:

(2)

Slope length (L) is calculated using a modified algorithm suitable for microtopography at the community scale, incorporating flow accumulation (A, in m²) for quantification:

(3)

The formula introduces a 22.13 m² flow accumulation threshold to correct biases in short slope regions (<50 m), matching the actual terrain characteristics of the study area [21].

After landscape type classification, accuracy is validated using a confusion matrix. Overall Accuracy (OA) is calculated as:

(4)

The Kappa coefficient (K) is:

(5)

where n = 8 is the number of landscape types, is the diagonal element of the confusion matrix (correctly classified pixels), N is the total number of pixels, and and are the actual and classified total pixel counts of type i. OA and Kappa were used to assess whether the landscape classification satisfied the accuracy requirements for model input [22]. The InVEST-SDR module quantifies the ecosystem’s sediment retention capacity using the Universal Soil Loss Equation (USLE) factors, indirectly representing runoff regulation efficiency [23]. The rainfall erosivity factor (R) considers the seasonality of the study area, with 70% of precipitation occurring from June to September. Monthly R values are calculated as:

(6)

stands for the monthly total precipitation in the study area (unit: mm). Annual R values is shown in Equation (7):

(7)

where is the monthly precipitation (mm). Monthly precipitation data from 2018 to 2023 were used to calculate seasonal R values for the rainy and dry periods [24].

Soil erodibility factor K is determined from soil particle composition (sand S, silt Si, clay Cl, in %) and organic matter (OM, in %):

(8)

Clay (clay content, unit:%) is a key indicator that characterizes the composition of soil particles, and together with sand particles (S, sand) and silt particles (Si, silt), determines the soil texture characteristics. The final K value is obtained from the organic matter correction calculation:

(9)

Slope-length LS factor is calculated in segments:

For < 9%:

(10)

For ≥ 9%:

(11)

The setting criteria for the segmentation threshold ( = 9%) and coefficients (0.065, 0.045, 0.0025; 0.138, 0.153, 0.0033) are as follows:

1. (1). Segmented threshold ( = 9%): Referring to the “Soil Erosion Classification and Grading Standards” and urban micro terrain characteristics, combined with the statistical results of community slope in the study area (80% of the area slope < 9%), the slope is divided into two categories: gentle slope (< 9%) and steep slope (≥ 9%), mainly adapted to the terrain characteristics of community short slope (< 50 m);
2. (2). Coefficient values: A community scale correction scheme based on the general soil erosion equation [25], referring to the parameter calibration experience of similar high-density urban community runoff simulation studies, the coefficient was adjusted to compensate for the amplification effect of runoff acceleration on erosion in steep slope areas. Finally, 42 sets of measured runoff data were verified to ensure that the LS factor calculation was consistent with the actual erosion law in the region. The segmentation accounts for the amplification effect of steep slopes on erosion. Vegetation cover factor C is linked to NDVI and fractional vegetation cover :

(12)

where and represent the reflectance values of the near-infrared and red bands, respectively. Fractional vegetation cover () is calculated as:

(13)

where = 0.05 for bare soil, and = 0.85 for dense vegetation. The final vegetation cover factor () is computed as:

(14)

The soil conservation factor () is determined based on the efficiency of implemented measures ():

(15)

where = 0.6 for grassed swales, = 0.8 for permeable pavement, and = 0 when no measure is applied, quantifying the erosion reduction effectiveness of different engineering interventions [26].

Potential soil loss is calculated using the USLE factors:

(16)

Next, the sediment delivery ratio () for each grid cell x is computed using the core InVEST-SDR formula:

(17)

Runoff depth () is derived by coupling potential runoff with actual infiltration. The potential runoff for cell x is:

(18)

where = 1215 mm represents the annual average precipitation, and is the proportion of green space in grid cell x. The base soil infiltration rate is related to the impervious surface fraction ():

(19)

The actual infiltration rate, adjusted for vegetation root effects, is:

(20)

The annual actual infiltration volume is then:

(21)

where = 1200 h represents the annual rainfall duration. Finally, runoff depth is calculated as:

(22)

where 0.72 is the correlation coefficient between SDR and runoff, and 1.35 is a base runoff adjustment factor, both calibrated using measured data.

Model parameter calibration and statistical analysis

Model parameter calibration was based on 42 sets of measured data collected from five monitoring points within the community during 2022–2023, covering light rain (5–15 mm), moderate rain (15–30 mm), and heavy rain (>30 mm). Sensitivity analysis was conducted to identify key parameters, with the sensitivity coefficient calculated as:

(23)

where represents the relative change in simulated runoff depth, and =5% is the relative change in the parameter. Model accuracy was evaluated using the coefficient of determination () and mean relative error (MRE):

(24)

(25)

where m = 42 is the number of data sets, and are the observed and simulated runoff depths, respectively, and is the mean observed runoff depth [27]. The association between landscape pattern and runoff regulation effectiveness was quantified using Pearson correlation analysis, with the correlation coefficient calculated as:

(26)

where k = 125 represents the number of community grid cells, is the landscape pattern index, is the corresponding grid cell runoff depth (), and and are the mean values of the variables [28,29]. Statistical significance was tested using the t-statistic:

(27)

The statistical analysis comprised four parts. First, sensitivity analysis was conducted to identify key parameters, including vegetation cover factor (C), slope-length and slope-gradient factor (LS), and infiltration rate, which then guided parameter calibration. Second, model fit was evaluated using the coefficient of determination (R²) and mean relative error, based on 42 rainfall events measured at the community scale between 2022 and 2023. Third, Pearson correlation analysis was used to examine linear relationships between landscape fragmentation, patch aggregation, and runoff response indicators, with significance levels assessed via t-tests [30,31]. Fourth, the regulatory effect of landscape optimization scenarios was assessed by calculating the differences and relative changes in runoff depth before and after optimization.

Model validation and parameter calibration results

Comparison of runoff simulation results based on the NLCD and GIS-interpreted data showed an average runoff depth deviation of 2.1 mm, indicating good consistency between the two sources for community-scale landscape identification. Cross-validation results confirmed that the digital landscape classification derived from high-resolution remote sensing imagery was suitable for subsequent model inputs and scenario simulations.

Sensitivity analysis revealed that the vegetation C, LS, and infiltration rate were highly influential, with sensitivity coefficients of 0.82, 0.65, and 0.43, respectively. These factors were identified as key variables for parameter calibration. During calibration, the C factor for tree-covered green space was adjusted from 0.06 to 0.05, reducing the relative error at the corresponding monitoring points from 8.7% to 1.9%. In areas with slopes exceeding 8%, the LS factor was adjusted from 0.43 to 0.48, which more accurately reflected runoff acceleration under steep-slope conditions. After calibration, the model achieved a R² of 0.89 and an average relative error of 1.3%, demonstrating high fitting accuracy at the community scale.

Fig 1 presents the model calibration results. Simulated runoff depths under different typical scenarios closely matched observed values, with most samples ranging from 10 to 50 mm. Absolute errors were narrow, and relative errors remained generally low, indicating that the model maintains good stability and applicability across varying rainfall conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Performance evaluation of the GIS–InVEST coupled model.

https://doi.org/10.1371/journal.pone.0352335.g001

Fig 1 shows that the simulated runoff depths under six representative scenarios are in strong agreement with the observed values, with most samples concentrated in the 10–50 mm range. The absolute error remains within a narrow interval, and the relative error is generally low. These results indicate that the model errors are not randomly distributed but are primarily concentrated in spatial units with rapid slope variation and pronounced differences in vegetation coverage. The C-factor and LS-factor exhibit higher sensitivity, suggesting that runoff processes at the community scale are more strongly influenced by vegetation interception capacity and short-slope flow accumulation conditions. In contrast, although infiltration rate is also an important factor, its marginal regulatory effect is weaker under high impervious surface conditions, where land cover type and slope structure differences play a more dominant role.

Current runoff regulation results

Fig 2 presents the assessment results of runoff regulation performance under current conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Current runoff regulation effectiveness assessment.

https://doi.org/10.1371/journal.pone.0352335.g002

Fig 2 shows that under the current landscape configuration, the proportion of impervious surfaces in the community is 37.2%, with an annual mean runoff depth of 41.8 mm and a runoff coefficient of 0.52. These results indicate a relatively high level of surface runoff generation under the existing spatial configuration, suggesting that the community still has considerable potential for improving stormwater retention and regulation capacity. The current results reveal two key characteristics. First, the relatively high proportion of impervious surfaces enables rapid conversion of rainfall into surface runoff, significantly limiting opportunities for infiltration and rainwater detention. Second, the regulatory effect of green spaces is largely confined to localized areas and has not yet formed a continuous buffering structure. As a result, the overall spatial organization efficiency remains low.

Optimization scenario and landscape pattern response results

To evaluate the regulatory effects of landscape optimization, an optimized scenario was established in which green-space coverage increased by 15% and patch aggregation increased by 23%. Under this scenario, the community’s mean annual runoff depth decreased from 41.8 mm to 27.3 mm, corresponding to a 34.7% improvement in runoff regulation efficiency. Fig 3 presents the results of the optimized scenario assessment.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Optimized scenario runoff regulation assessment results.

https://doi.org/10.1371/journal.pone.0352335.g003

Fig 3 shows a significant positive correlation between landscape fragmentation and runoff volume, as well as a strong positive correlation between patch aggregation and runoff reduction rate. Comparative results of different configuration strategies indicate that a dispersed green space layout achieves 18.6% higher regulation efficiency than a concentrated layout. The observed improvements under optimized scenarios are not solely driven by an increase in total green space. The increase in green coverage expands the spatial capacity for rainfall interception and surface water storage. As patch aggregation increases, previously isolated green units gradually form more continuous ecological buffering structures, which extend surface flow paths and reduce runoff concentration speed. Meanwhile, increased connectivity of impervious surfaces strengthens rapid runoff pathways, thereby amplifying runoff generation effects. The baseline results are used to identify the current runoff regulation status of the community, while the optimized scenario results are used to compare improvement magnitudes under different landscape adjustment strategies.

High consistency in landscape classification

The small average runoff depth deviation of 2.1 mm between NLCD data and GIS-interpreted data indicates that community-scale landscape identification is largely consistent across different data sources. This consistency does not imply that the two datasets are identical, but rather that they accurately capture key regulatory units, particularly impervious surfaces, green spaces, and water bodies. The areas most sensitive to runoff regulation are those with significant differences in surface infiltration capacity. As long as the spatial delineation between hardened surfaces and vegetated areas remains stable, model outputs show minimal deviation. High-resolution remote sensing imagery further enhances the identification of micro-scale surface features, such as road edges, auxiliary paved areas, and scattered green patches, which are incorporated into the raster representation. This high-precision spatial data underpins the strong agreement observed in cross-validation and highlights the need for accurate landscape classification in community-scale runoff assessment.

Mechanisms underlying key parameter sensitivity differences

The high sensitivity of the C factor (vegetation cover), LS factor (slope length–gradient), and infiltration rate reflects that community runoff processes are primarily controlled by surface cover, micro-topography, and infiltration capacity. The C factor shows the highest sensitivity due to its direct relation to green-space coverage. Vegetation alters surface roughness, rainfall interception, root-mediated water conduction, and topsoil structure, thereby exerting strong front-end regulation on runoff formation. The LS factor ranks second in sensitivity, indicating that even within short slopes and micro-topography typical of communities, the coupled effect of slope length and gradient accelerates surface flow. When slope increases, surface water residence time decreases, local infiltration opportunities diminish, and runoff depth rises sharply. Although infiltration rate also significantly influences runoff, its sensitivity is slightly lower. In high-density communities with extensive impervious surfaces, the potential impact of increased infiltration is limited; thus, its marginal regulatory effect is weaker than that of vegetation cover and slope-driven flow. The substantial reduction in errors following parameter calibration indicates that the original parameter system was reasonable, with residual errors concentrated in specific surface types and steep-slope areas rather than reflecting instability in the overall model structure.

Intrinsic reasons for high runoff under current conditions

Under existing conditions, both mean annual runoff depth and runoff coefficient are relatively high. This is not solely due to insufficient green space, but rather to a structural mismatch between community landscape configuration and surface hydrological processes. High proportions of impervious surfaces reduce storage and infiltration capacity, causing rainfall to convert rapidly into surface runoff. Meanwhile, green patches exist but are small and scattered, preventing the formation of continuous buffers or interception zones. Spatially, stronger connectivity among hardened surfaces creates direct runoff pathways, placing greater runoff concentration pressure on local low-lying areas. In other words, the community does not lack regulatory units, but the connectivity between these units is insufficient to support synergistic runoff reduction. Individual green patches still provide ecological function, but when separated by roads, auxiliary paved surfaces, and scattered hardened areas, the overall retention capacity is significantly reduced. This explains why community-scale studies cannot rely solely on coverage proportion but must also consider landscape structural connectivity and patch configuration.

Mechanisms behind the significant improvement under the optimized scenario

The increase in runoff regulation efficiency under the optimized scenario is driven not only by the expansion of green-space area but also by the synergistic effects of spatial configuration. As green-space coverage rises, the capacity for rainfall interception, surface storage, and soil infiltration expands, initially reducing total runoff. When patch aggregation also increases, previously scattered small green patches form more continuous ecological blocks. This extends the spatial pathways for rainfall movement, slows surface flow, and allows local runoff reduction effects to propagate over a larger area [32]. In other words, increasing green-space area addresses whether regulatory space exists, while higher aggregation determines whether these spaces can interact to form functional networks. When both factors improve together, internal retention chains and buffer structures within the community become more complete, producing a marked enhancement in runoff reduction. These findings indicate that sponge city upgrades should not rely solely on expanding green space; spatial organization is equally crucial for achieving effective regulation. These findings are consistent with the conclusions of Study [7], which reported that the spatial configuration of green stormwater infrastructure significantly affects the supply–demand matching of flood regulation services. They also align with Study [13], which emphasized that the efficiency of nature-based solutions is constrained by spatial adaptability. However, this study further refines the analysis by downscaling it to the community unit level and quantitatively evaluating the combined effects of increased green coverage and enhanced patch aggregation on runoff reduction magnitude.

Reasons for differences in fragmentation, aggregation, and configuration patterns

Higher landscape fragmentation leads to greater runoff generation because green and permeable areas become segmented, disrupting continuous infiltration interfaces. In contrast, higher green space aggregation enables the formation of more integrated retention surfaces and transitional zones, thereby enhancing runoff reduction capacity. The superior performance of a dispersed green space layout does not contradict the benefits of an aggregated layout, as the two operate at different spatial levels. Aggregated configurations improve local structural integrity, while dispersed layouts expand interception coverage across multiple runoff initiation points. In high-density communities, broad spatial coverage is crucial for preventing rapid hydrological connectivity and runoff concentration. The findings of this study—namely, that increased landscape fragmentation intensifies runoff output and that higher aggregation enhances runoff reduction capacity—are consistent with the conclusions of Study [9], which emphasized that the effectiveness of green infrastructure in runoff regulation is strongly governed by spatial form. Compared with existing studies, this work further demonstrates that dispersed layouts achieve higher overall regulation efficiency within high-density community contexts. This difference suggests that runoff initiation points in communities are more spatially dispersed, and that single centralized configurations are insufficient to provide system-wide interception advantages.

Reasons why impervious surface connectivity becomes a key driver

Runoff increases significantly as impervious surface connectivity rises, reflecting a clear hydrological mechanism. Once hardened surfaces form continuous channels, rainfall rapidly converges along paths of least resistance, compressing surface storage and infiltration processes. In high-density communities, roads, plazas, and hardened zones surrounding buildings often form such fast-flow networks. Compared with total impervious surface proportion, connectivity better explains why some communities exhibit different runoff responses despite similar levels of surface hardening. The proportion indicates how much impervious surface exists, whereas connectivity determines how efficiently water is transmitted. In high-density urban areas, connectivity is often more directly linked to sudden runoff surges than surface coverage alone. Therefore, reducing impervious area is not sufficient; breaking continuous hardened chains, introducing permeable transition zones, and adding micro-buffer units are necessary to structurally mitigate rapid runoff amplification. This interpretation is consistent with Study [3], which identified hydrological connectivity as a central factor in flood regulation services. It also aligns with Study [32], which demonstrated that the source–sink relationship of urban green infrastructure–runoff systems is strongly influenced by spatial structure. The contribution of this study lies in explicitly identifying impervious surface connectivity as a key triggering factor for rapid runoff amplification at the community scale, and incorporating it into the landscape optimization analysis framework.

Implications for community landscape optimization

Runoff regulation at the community scale is not determined by a single indicator but is jointly shaped by landscape composition, spatial configuration, and micro-topography. High-resolution digital landscape representation improves the identification of spatial heterogeneity, while parameter calibration enhances the reliability of runoff simulation. The results indicate that regulation efficiency depends on whether spatial structure can delay surface runoff, enhance infiltration, and weaken the rapid connectivity of impervious surfaces. This shifts the analytical focus from simple area-based comparisons to structural coordination and provides a basis for subsequent planning recommendations.