A Comprehensive Modelling Framework for Identifying Green Infrastructure Layout in Urban Flood Management of the Yellow River Basin

Abstract

The Yellow River Basin faces severe challenges in water security and ecological protection: at the basin scale, complex hydrological processes and fragile ecosystems undermine the water security pattern; at the local scale, waterlogging risks have intensified in Zhengzhou—a core city in the lower reaches—impacting the city itself and also exerting negative effects on the basin’s water security. To address this, mapping the scientific layout of green infrastructure (GI) is urgent. However, existing studies on GI layout at the basin-urban scale have certain limitations: neglect of underlying surface spatial heterogeneity, insufficient integration of natural, hydrological and social factors’ synergies, and lack of research on large-scale basins and cities, especially ecologically sensitive areas with complex hydrological processes. To fill these gaps, this study proposes an integrated framework (SCS–GIS–MCDM) combining the SCS hydrological model, GIS spatial analysis, and multi-criteria decision making (MCDM). The SCS hydrological model is refined via localized parameter calibration for better accuracy; indicator weights are determined through the MCDM framework; and green infrastructure (GI) suitability maps are generated by integrating ArcGIS spatial analysis with fuzzy logic. Results show that (1) 6.8% of Zhengzhou is highly suitable for GI, mainly in riparian areas and the Yellow River alluvial plain; (2) sensitivity analysis confirms flooded areas and runoff corridors as key drivers; (3) spatial validation against government-issued ecological control zone plans demonstrates the model’s value in balancing flood safety and socio-economy. This framework provides a replicable application model for GI construction in cities along the Yellow River Basin, thereby supporting urban planners in making evidence-based decisions for sustainable blue–green space planning.

1. Introduction

The Yellow River, culturally revered as China’s “Mother River,” is the country’s second-longest river [1]. Ensuring water security and strengthening ecological protection in the Yellow River Basin are critical to regional sustainable development [2]. The water resources of the Yellow River face sustainability risks in both quantity and quality: hydrological studies indicate that, since the 1950s, river runoff has decreased due to human activities; meanwhile, rapid industrialization and urbanization have exacerbated water quality deterioration, with concentrations of heavy metals and total nitrogen gradually increasing from the upper to the lower reaches [3]. These issues have led to severe ecological imbalance and degradation [4,5].

As a core city in the Yellow River alluvial plain, Zhengzhou adjoins a critical “Hanging River” segment, with the Yellow River’s main channel 17 km from its downtown. This geographical attribute, combined with rapid urbanization, renders Zhengzhou highly flood-vulnerable: (1) a substantial portion of its area faces 100-year flood risks [6], linked to the basin’s overall hydrological vulnerability; (2) urban expansion has elevated underlying surface hardening rates, reducing natural water infiltration [7], and, when coupled with increased extreme rainfall frequency [8,9], this has exacerbated urban waterlogging—38 incidents were recorded during 2011–2018, 26 of which caused extensive impacts [10]. These local waterlogging events also pose cascading risks to the Yellow River Basin’s broader water security framework. Sole reliance on traditional grey infrastructure fails to balance rainwater-flood regulation, ecological restoration, and socio-economic development [11,12,13,14]. In contrast, green infrastructure (GI) simulates natural hydrological processes [15,16,17], enabling simultaneous runoff reduction, soil erosion control, and habitat restoration [18], thereby offering a systematic solution to basin–urban water security conflicts.

While GI’s multi-functional advantages are widely recognized, its effectiveness depends heavily on scientific spatial layout—misplaced GI facilities rarely deliver intended ecological benefits. Recent GI research in stormwater management and urban resilience has focused on three core areas: (1) GI facility design and application (e.g., structural optimization of green roofs or rain gardens [19,20,21,22]); (2) GI effect assessment (quantifying runoff reduction, water purification, and ecological improvement [17,23,24,25,26]); (3) GI–urban planning integration (exploring GI layout adaptability across scales [27,28,29]). This study specifically focuses on the third direction, with an emphasis on conducting research on the site selection and configuration of GI at the basin–urban scale.

Currently, the mainstream technical approaches for GI layout research revolve around the “GIS + MCDM” integration framework and hydrological simulation-driven methods. However, two limitations persist, making these approaches ill-suited to the specific needs of cities along the Yellow River in the Yellow River Basin. For one thing, most hydrological simulations employ the SCS–CN or SWMM models with default parameters [30,31,32,33]. These models operate on the assumption of basin homogenization, neglecting the underlying surface heterogeneity caused by the superposition of loess parent material, sandy soil, and urbanization. This oversight results in significant errors in runoff prediction [34,35,36]. For another, while existing GI studies have covered spatial planning at neighbourhood or urban central scales [23,37,38], and some have explored hydrological-process-driven GI optimization [39,40], research focusing on larger spatial units at the basin scale (e.g., large cities encompassing urban–rural transition zones or areas adjacent to important rivers) remains relatively limited. In particular, though there have been attempts to link GI planning with urban resilience [41,42], studies that integrate soil spatial heterogeneity, hydrological complexity, and socio-economic constraints for GI collaborative planning in ecologically sensitive regions like the Yellow River Basin are still insufficient [43].

In response to the needs of “water-city” co-ordinated governance in the Yellow River Basin and the urgency of flood control in Zhengzhou [3,6,44], this study focuses on the administrative area of Zhengzhou, with key coverage of ecologically sensitive areas along the Zhengzhou section of the main Yellow River and areas prone to waterlogging. It then proposes an integrated GI identification framework (SCS–GIS–MCDM) that combines the SCS hydrological model, GIS spatial analysis, and multi-criteria decision making (MCDM), with fuzzy logic specifically embedded in the framework’s suitability classification stage to address ambiguities in natural and socio-economic gradients. The rationale for integrating these three core techniques (SCS, MCDM, fuzzy logic), rather than alternative approaches, lies in their synergistic ability to address the unique challenges of GI planning in the Yellow River Basin. Compared to SWMM (which is more suited for small-scale urban neighbourhood simulations [33]), SCS model better adapts to the loess-sandy soil heterogeneity, reducing runoff prediction errors caused by default parameters. Unlike single-factor evaluation methods, MCDM systematically balances multi-objective trade-offs, preventing one-dimensional biases in site selection. Fuzzy logic enhances the robustness and applicability of the analysis by resolving the uncertainty in the definition of spatial boundaries.

The objectives of this study are as follows: (1) improve the runoff simulation accuracy of the SCS hydrological model in the study area via localized CN calibration; (2) directly couple hydrological outputs with BWM-based multi-criteria spatial suitability analysis; (3) conduct weight sensitivity analysis and spatial matching validation to provide a reusable method for GI priority delineation at the basin–urban scale.

While this integrated framework offers some reference value for existing research, in practical application, uncertainties from data characteristics and methodological attributes still need to be addressed. Specific constraints are as follows: First, the framework is highly dependent on the accuracy of basic datasets. Second, the dynamic changes in land cover pose challenges to the model’s applicability. Activities such as urban expansion and ecological restoration alter land surface cover, which not only necessitates frequent calibration of SCS parameters but may also shorten the timeliness of GI suitability conclusions. Third, differences in the preferences of decision-makers may lead to weight biases.

2. Study Area and Data

Zhengzhou (34°16′–34°58′ N, 112°42′–114°14′ E) is located on the southern bank of the middle and lower reaches of the Yellow River (Figure 1). As the capital of Henan Province and a core city in the Yellow River Basin, it covers a total area of 7566 km², with a permanent population of 13.086 million and an urbanization rate of 81% in 2024 [45]. The topography and geomorphology show significant stepped differentiation [46]. Zhengzhou has a history of frequent flood and waterlogging disasters, particularly the catastrophic heavy rainfall event on July 20 in 2021 [47,48,49]. During this event, Zhengzhou National Meteorological Station recorded a maximum daily rainfall of 624.1 mm, breaking the meteorological observation record in mainland China [50]; the disaster caused 380 deaths and a direct economic loss of 40.9 billion yuan in Zhengzhou, while also leading to a comprehensive paralysis of urban transportation and damage to infrastructure in some areas, such as water and power outages and communication disruptions. This incident exposed the vulnerability of the urban flood management system to extreme climates and also posed a potential threat to flood control safety in the lower reaches of the Yellow River.

The main data used in this study include remote sensing images, topographic data, soil data, meteorological data, hydrological data, and socio-economic data (

Table 1. Basic Information on Research Data.

Data Type	Data Name	Year(s)	Resolution/Scale	Data Preprocessing
Remote Sensing	Landsat 8	2017; 2020; 2021	30 m	Visual interpretation; Local calibration
Topographic	DEM	2018	30 m	Verification
Soil	Soil Classification Vector Map	-	1:200,000	Soil property table; Reclassification;
Meteorological	Extreme Heavy Rainfall Days	2016–2022	-	None
Hydrological	Water Level and Flood Discharges of Changzhuang Reservoir	(rainstorm days)	-	Calculate inflow volume
Socio-economic	Administrative Boundary Map Vector Map of Traffic Network	2022 2023	1:100,000	Data integration Format conversion

). Key details are as follows:

(1) Remote sensing and topographic data

Landsat 8 images (Path/Row: 124/36; dates: 27 March 2017, 19 March 2020, 22 March 2021; cloud-free, good quality) and 30 m resolution ASTER GDEM (DEM) were from the National Geospatial Data Cloud Platform (http://www.gscloud.cn).

DEM accuracy: Verified via 50 random points vs. 1:50,000 topographic maps (Zhengzhou Municipal Bureau of Natural Resources and Planning); elevation deviation ≤ 5 m (meets basin-scale flood simulation precision).

(2) Land use data

Generated by interpreting multi-phase Landsat 8 images (for hydrological model calibration) via SVM supervised classification + visual interpretation; covers five types (construction land, farmland, forest land, water bodies, unused land).

Accuracy assessment: Ground truth samples (field surveys + high-resolution imagery; n = 300/phase, covers all land use spatial heterogeneity) showed OA ≥ 86%, Kappa ≥ 0.86; per-class accuracy (producer’s/user’s): construction land (≥87%/≥84%), forest land (≥88%/≥86%), farmland (≥83%/≥81%), water bodies (≥91%/≥89%), unused land (≥75%/≥72%).

(3) Soil data

Sourced from the 1:200,000 Henan soil classification vector map (National Earth System Science Data Sharing Platform, http://www.geodata.cn). A soil property table (including saturated hydraulic conductivity, bulk density, and effective water-holding capacity) was constructed, soil types were reclassified, and both were verified via 0–30 cm field sampling in Zhengzhou.

(4) Hydrological and other spatial data

Hydrological data: Water levels/flood discharge at Changzhuang Reservoir (2016–2022, extreme rainfall periods; Zhengzhou Water Conservancy Bureau) for hydrological model calibration/validation.

Spatial data: Zhengzhou 1:100,000 administrative boundaries (2022) and traffic network (provincial/urban arterial roads, 2023 update) from Zhengzhou Municipal Bureau of Natural Resources and Planning.

3. Methods

The GI identification framework constructed in this study focuses on the dual goals of water security in the Yellow River Basin and flood resilience in Zhengzhou, which is specifically divided into three stages (Figure 2). (1) Identifying Zhengzhou’s flood security pattern: Based on the calibrated SCS-CN hydrological model, the inundation processes under rainfall scenarios with different return periods in Zhengzhou are simulated to analyse the key issues of the city’s hydrological characteristics and flood disaster mechanisms, and to extract runoff corridors and graded inundation areas. (2) Establishing a MCDM system for GI suitability: Nine evaluation indicators are selected from three dimensions—natural constraints, flood security (runoff corridors, inundation area), and socio-economy (land cost, transportation cost). The best–worst method (BWM) is used to determine the weights of each evaluation indicator. Vector data are uniformly converted into 30 m × 30 m raster format via the ArcGIS platform to achieve data standardization for geospatial multi-criteria analysis. (3) Optimizing GI spatial configuration oriented by natural water processes: Fuzzy membership functions are used to normalize the indicators. Combined with the weighted linear combination (WLC) and fuzzy overlay methods, a suitability grade map for GI construction is generated.

3.1. SCS Model and Calibration

The SCS (Soil Conservation Service) hydrological model has been widely applied in runoff simulation for medium-to-large scale river basins due to its concise structure and interpretable parameters [51,52,53]. The grid-based SCS model has significantly improved runoff prediction accuracy through spatially explicit modelling [54,55].

3.1.1. Rainfall–Runoff Relationship

The rainfall–runoff relationship expression of the SCS model is

$$\mathit{Q}=\left\{\begin{array}{cc}{\displaystyle \frac{{(\mathit{P}-\mathit{I}\mathit{a})}^{\mathbf{2}}}{\mathit{P}+\mathit{S}-\mathit{I}\mathit{a}}},\hfill & \hfill \mathit{P}\ge \mathit{I}\mathit{a}\\ \hfill \mathbf{0},& \hfill \mathit{P}<\mathit{I}\mathit{a}\end{array}\right.$$

In the formula, Q is the runoff (mm), P is the rainfall (mm), Ia is the initial abstraction (mm), and S is the maximum basin retention (mm). Its core parameter S is determined by the CN value (dimensionless, 0 ≤ CN ≤ 100):

$$\mathit{S}={\displaystyle \frac{\mathbf{25,400}}{\mathit{C}\mathit{N}}}-\mathbf{254}$$

CN (curve number) is a dimensionless parameter used to describe the relationship between rainfall and runoff. It integrates multiple factors influencing hydrological processes, including antecedent moisture condition (AMC), soil texture, and current land use status.

3.1.2. Local Calibration of CN Values

The underlying surface of Zhengzhou exhibits significant multi-scale spatial heterogeneity. Given this, localized calibration of CN values is necessary to reduce model parameterization errors, ensuring that storm runoff simulation results align with urban hydrological response characteristics. Changzhuang Reservoir is located in Zhongyuan District, western Zhengzhou (34°48′ N, 113°32′ E), situated on the Jiayu River, a tributary of the Jialu River. With an 82 km² catchment area, the Changzhuang Basin covers all the study area’s underlying surfaces (loess platforms, urban/farmland/forest land) and serves as a typical “tableland–urban” transitional basin in western Zhengzhou, capturing the urban–suburban–rural underlying surface gradient. Valid selected data from the region’s long-term hydrological records covers typical rainfall scenarios, supporting this study’s runoff derivation and parameter calibration.

To avoid the impact of temporal changes in land use on CN accuracy, this study first systematically examined the variation characteristics of the underlying surface pattern in the reservoir’s controlled catchment using Zhengzhou’s 2016–2022 land use planning documents, annual data from the Natural Resources Bureau, and publicly available lists of major construction projects from the government. Results indicated that, although the catchment’s land use underwent overall adjustments during the study period, it exhibited a “stage-specific stability” trait: no drastic changes affecting CN (such as the expansion of construction land or deforestation) occurred in the three periods of 2016–2018, 2019–2020, and 2021–2022. The differences between these periods served as the objective basis for dividing calibration stages. Based on this, the study divided the 2016–2022 period into three stable stages, each matched with a set of remotely sensed land use data: data from 2017 were used to calibrate rainfall events in 2016–2018, 2020 data for events in 2019–2020, and 2021 data for events in 2021–2022. Ultimately, 10 typical rainfall events were selected for calibration based on two core criteria: (1) diverse scenarios, covering 44.5–552.5 mm rainfall intensities (including the extreme event on 20 July 2021) and concentrated in Zhengzhou’s July–August flood season, matching the study’s flood risk focus; (2) complete observed data, with continuous reservoir water level, flood discharge, and rainfall records to ensure reliable “true” runoff calculation and to avoid calibration bias.

Changzhuang Reservoir’s measured “total basin inflow volume” was used as the true value to assess SCS model runoff simulation accuracy: increased reservoir storage (calculated via water level changes and verified “water level-storage curve”) was summed with total flood discharge to obtain inflow volume, which was then converted to runoff depth by basin area. The SCS model simulated runoff depth using IDW-generated CN spatial distribution (adjacent CN differences ≤ 10 to avoid unreal changes) and land use/soil data; CN values were adjusted via manual iterative zoning to maximize NSE and to reduce simulation deviation. Post-calibration, all 10 events showed consistent high performance: relative error narrowed from −34.4–40.8% to 1.2–9.8%, NSE rose from 0.33–0.52 to 0.81–0.88 (Appendix A 

Table A1. Calibration Results of 10 Rainfall Events in Changzhuang Reservoir Watershed.

Event No.	Date	Rainfall (P, mm)	Observed Q (mm)	Simulated Q (mm)	Relative Error (%)	NSE
1	9 July 2016	102.1	55	55.66	1.2	0.81
2	19 July 2016	76.5	33.2	33.83	1.9	0.82
3	2 August 2017	54.6	8.7	8.92	2.5	0.83
4	19 August 2018	52.5	8.6	8.88	3.3	0.84
5	1 August 2019	90.2	46	47.89	4.1	0.85
6	4 July 2020	56.4	9.3	9.75	4.8	0.85
7	7 August 2020	88.5	40	42.48	6.2	0.86
8	20 July 2021	552.5	450	481.05	6.9	0.87
9	29 August 2021	78.6	35	37.84	8.1	0.87
10	25 July 2022	44.5	7	7.69	9.8	0.88

), confirming model robustness for subsequent inundation simulation and GI suitability assessment. Local land use–soil CN values are in Appendix A 

Table A2. Localised CN Values for Land Use–Soil Combinations in Zhengzhou (AMC II).

Land Use	Soil Texture		CN Value
Construction land	Sandy soil	Coarse sand	68
		Fine sand	70
	Loam	Sandy loam	79
		Silt loam	82
	Clay loam	Clay loam	86
		Silty clay loam	88
	Clay soil	Heavy clay	92
		Light clay	90
Farmland	Sandy soil	Coarse sand	62
		Fine sand	65
	Loam	Sandy loam	72
		Silt loam	75
	Clay loam	Clay loam	80
		Silty clay loam	82
	Clay soil	Heavy clay	85
		Light clay	83
Forest land	Sandy soil	Coarse sand	36
		Fine sand	40
	Loam	Sandy loam	55
		Silt loam	60
	Clay loam	Clay loam	70
		Silty clay loam	72
	Clay soil	Heavy clay	77
		Light clay	75
Unused land	Sandy soil	Coarse sand	77
		Fine sand	80
	Loam	Sandy loam	84
		Silt loam	86
	Clay loam	Clay loam	88
		Silty clay loam	90
	Clay soil	Heavy clay	91
		Light clay	90
Water bodies	All soil types		100

.

3.2. Rainfall-Flood Simulation

3.2.1. Multi-Duration Design Rainfall Scenarios

From the daily rainfall series in Zhengzhou from 2000 to 2022, the maximum 1 h, 6 h, and 24 h rainfall amounts for each year were synchronously extracted to form a sample of multi-duration extreme events. The Pearson Type Ⅲ (P-Ⅲ) distribution was used to fit the rainfall frequency characteristics [56,57]. Combined with the hydrological parameters of the semi-humid region in the Yellow River Basin, multi-duration rainfall-intensity–frequency relationship curves were established. Finally, the design rainfall amounts for the 2-year return period (50% frequency), 10-year return period (10% frequency), and 50-year return period (2% frequency) were derived (

Table 2. Rainfall for Different Recurrence Periods in Zhengzhou.

Rainfall Frequency	Maximum 1 h Rainfall	Maximum 6 h Rainfall	Maximum 24 h Rainfall
50%	39.3 mm	58.8 mm	85.2 mm
10%	75.9 mm	120.4 mm	169.9 mm
2%	111.7 mm	180.9 mm	251.1 mm

), providing scenario inputs that align with the rainstorm characteristics of the Yellow River Basin for rainstorm and flood simulation.

Goodness-of-fit diagnostics confirmed that the P-Ⅲ distribution provided satisfactory representation (R² = 0.93; RMSE = 6.2 mm). Temporal rainfall profiles were constructed using the Chicago design storm method, widely applied in semi-humid basins, ensuring realistic intra-storm distribution of peak intensity.

3.2.2. Extract Runoff Corridors and Inundated Areas

The extraction of runoff corridors and inundated areas provides critical spatial support for the site selection and assessment of GI in Zhengzhou. From a practical perspective, it is essential to first clarify the applicability of core foundational data, as this is a prerequisite for ensuring the reliability of results. On the one hand, the inundation areas extracted based on the 2021 land use data can still meet the practical needs of GI suitability assessment for a certain period in the future. This is because, since 2021, Zhengzhou’s urbanization process has shifted from “incremental expansion” to “inventory optimization”, and the dynamic changes of land cover have entered a stable period with low fluctuations [58]. Meanwhile, the Zhengzhou Territorial Spatial Master Plan (2021–2035) [59] clearly puts forward the goal of “optimizing the blue–green space system” and imposes rigid constraints on the expansion of construction land. On the other hand, DEM data exhibits stability at the scale of rainstorm events: given the stable terrain and water retention capacity at the scale of rainstorm events, the static DEM can reflect the actual flood detention space; moreover, the error is reduced through the catchment-specific coupling of runoff volume, balancing efficiency and the accuracy of inundation representation.

Based on the aforementioned applicable foundational data, the specific method for extracting runoff corridors and inundated areas is implemented as follows. The flow direction (using the D8 algorithm) and flow accumulation were extracted from the depression-filled DEM, and a natural runoff corridors network was generated based on the spatial distribution characteristics of flow accumulation [60,61]. During the flow accumulation process, the simplified assumption of “downslope one-way routing” was adopted: water flow only moves along the maximum slope direction extracted from the DEM, and lateral overland flow across corridors is not considered temporarily. Since runoff in the study area is dominated by concentrated corridor confluence, the impact of dispersed overland flow on the confluence process is negligible, and this assumption will not significantly increase errors. No additional retention parameters were set for artificial flood detention areas, because the flood detention effect of Changzhuang Reservoir has been calibrated and quantified, while other small reservoirs and ponds have not been assigned separate parameters for the time being. Combined with the runoff volume (Q) of each catchment calculated by the calibrated SCS model, the runoff volume (V) was dynamically coupled with the 3D topographic features of the DEM using the spatial statistical function of GIS. The iterative approximation method was used to adjust the inundation height step by step, so that the relative error between the water storage volume (V’) of topographic depressions and the runoff volume was ≤2%. Finally, the continuous areas in the DEM lower than the inundation height were extracted as the inundation range.

3.3. Data Selection and Evaluation Criteria

The multi-criteria decision-making system for GI suitability analysis includes three criterion layers: natural conditions, social economy, and flood security (Table 3).

The natural conditions criterion layer selects five indicators: elevation, slope, aspect, soil texture, and vegetation coverage index (NDVI). These aim to quantify the constraint mechanism of terrain, soil, and ecological background on the hydrological functions of GI. Elevation and slope restrict the adaptability of GI’s hydrological functions and the feasibility of engineering implementation [62,63]; Zhengzhou (plains/loess tablelands) classifies areas with elevation <15 m and slope <5° as highly suitable for GI. Aspect affects rainwater infiltration efficiency through microclimate regulation [64]; Zhengzhou’s (Northern Hemisphere) shady/half-shady slopes aid infiltration more than sunny slopes. Soil texture and NDVI, respectively, characterize the permeability of the underlying surface and the vegetation interception capacity, which together form the ecological basis for rainwater and flood regulation [65,66]. Among them, sandy soil has the best permeability, and the higher the NDVI value, the stronger the interception capacity.

The social economy criterion layer includes two indicators: land use and distance to provincial highways and urban arterial roads. These focus on the construction feasibility and social acceptability of GI implementation [67,68]. Green spaces within 30 m of road can reduce disturbances from the traffic environment to residents and improve their comfort [69]; green belts within 100–200 m outside roads can facilitate residents’ daily use [70]; however, for green spaces more than 200 m away from roads, the development costs increase significantly, which, in turn, reduces construction feasibility [71].

The flood security criterion layer is based on two indicators: simulated inundation areas and runoff corridors. These respond to the disaster risks caused by frequent extreme rainfall events. Table 4 presents the criteria/factor classifications for the fuzzy membership relationship in the GI suitability model (e.g., “high flood security suitability” for runoff corridor buffer zone (0–100 m) and inundated area under 50% probability rainfall). The spatial analysis of ArcGIS was used for data processing and analysis, and all factors are presented in raster data format (Figure 3) with a consistent spatial resolution of 30 m × 30 m to ensure data comparability. To verify the validity of the simulated inundation area, the extreme rainfall event in Zhengzhou on 20 July 2021 was taken as a reference. Using measured inundation data from the Field Scientific Observation and Research Station for Eco-Hydrological Evolution in the Lower Yellow River Plain (Ministry of Water Resources), quantitative validation was further conducted: the model achieved an 79% hit rate (correctly capturing actual inundation areas) and a 17% false alarm rate, with the inundation areas in Zhengzhou’s Yellow River beach area extracted by the model showing good spatial consistency with those measured by the station.

Table 3. Criteria/Factor Classification (Fuzzy Membership) Used for GI Suitability Model [72,73].

Criteria	Sub-Criteria	Highly Suitable	Moderately Suitable	Less Suitable	Not Suitable	Range Source
Flood security	Runoff corridor buffer zone (m)	0–100	100–250	250–500	>500	Literature and GIS analysis
	Inundated area	Inundated area under 50% probability rainfall	Inundated area under 10% probability rainfall	Inundated area under 2% probability rainfall	Uninundated area	Literature and GIS analysis
Natural conditions	Elevation (m)	<150	150–350	350–850	>850	Literature and GIS analysis
	Slope (degree)	<5	5–15	15–25	>25	Literature and GIS analysis
	Aspect	North/ northeast/ northwest	East/west	Southeast/ southwest	South	Literature and GIS analysis
	NDVI	>0.5	0.3–0.5	0–0.3	<0	Literature and GIS analysis
	Soil texture	Sandy soil	Loam	Clay loam	Clay soil	Literature
Social economy	Land use	forest land/ water area	unused land	farmland	construction land	Literature and GIS analysis
	Distance to provincial road/ urban arterial road (m)	<30	30–100	100–200	>200	Literature and GIS analysis

Note: The normalized difference vegetation index (NDVI) effectively depicts the distribution of surface vegetation in the study area.

Table 3. Criteria/Factor Classification (Fuzzy Membership) Used for GI Suitability Model [72,73].

Criteria	Sub-Criteria	Highly Suitable	Moderately Suitable	Less Suitable	Not Suitable	Range Source
Flood security	Runoff corridor buffer zone (m)	0–100	100–250	250–500	>500	Literature and GIS analysis
	Inundated area	Inundated area under 50% probability rainfall	Inundated area under 10% probability rainfall	Inundated area under 2% probability rainfall	Uninundated area	Literature and GIS analysis
Natural conditions	Elevation (m)	<150	150–350	350–850	>850	Literature and GIS analysis
	Slope (degree)	<5	5–15	15–25	>25	Literature and GIS analysis
	Aspect	North/ northeast/ northwest	East/west	Southeast/ southwest	South	Literature and GIS analysis
	NDVI	>0.5	0.3–0.5	0–0.3	<0	Literature and GIS analysis
	Soil texture	Sandy soil	Loam	Clay loam	Clay soil	Literature
Social economy	Land use	forest land/ water area	unused land	farmland	construction land	Literature and GIS analysis
	Distance to provincial road/ urban arterial road (m)	<30	30–100	100–200	>200	Literature and GIS analysis

Note: The normalized difference vegetation index (NDVI) effectively depicts the distribution of surface vegetation in the study area.

3.4. Weighting Criteria

In this study, the best–worst method (BWM) is adopted to determine the relative weights of various evaluation criteria in the multi-criteria decision-making model for GI layout. This method systematically compares critical criteria [74,75,76], significantly reducing the number of pairwise comparisons required by the traditional analytic hierarchy process (AHP) and leveraging a consistency optimization model to minimize subjective bias. It is particularly suitable for complex spatial decision-making problems [77].

Five interdisciplinary experts (covering three core fields: ecological planning, hydrological modelling, and GIS technology) were invited to participate in the evaluation. Specifically, they included one hydrologist (12 years of experience, specializing in rainfall-runoff models and hydrological risk assessment), one GIS expert (10 years of experience, focusing on spatial analysis, land use classification, and statistical modelling), one ecological planner (15 years of experience, specializing in ecological corridors, green space system planning, and biodiversity assessment), one civil engineer (12 years of experience, engaged in urban drainage design and flood control infrastructure construction), and one environmental policy scholar (9 years of experience, focusing on ecosystem service assessment and policy integration for urban resilience planning). To ensure the reliability and relevance of experts’ evaluations, the selection strictly followed three core criteria: (1) field matching: all experts had research/practical backgrounds related to GI planning, hydrological modelling, or GIS spatial analysis; (2) regional relevance: all experts had experience in projects related to Zhengzhou and the Yellow River Basin, and were familiar with the uniqueness of the study area, ensuring that weight assignment aligned with local realities; (3) qualification threshold: all experts had at least 8 years of experience in their target fields, with solid theoretical foundations and practical experience.

The specific steps are as follows:

Step 1: Construction of the criterion system. As described in Section 3.3, 9 core criteria were selected to build a hierarchical evaluation system (Table 4).

Step 2: Identify the best (most important) and worst (least important) criteria.

Step 3: Best-to-others comparison vector (BO vector): define the relative importance of the best criterion (B) to each of the other criteria (j) using a 1–9 scale (Table 4), and construct a row vector:

$${\mathit{A}}_{\mathit{B}}=\left[{\mathit{a}}_{\mathit{B}\mathbf{1}},{\mathit{a}}_{\mathit{B}\mathbf{2}},\dots ,{\mathit{a}}_{\mathit{B}\mathit{j}}\right]$$

(1)

Step 4: Others-to-worst comparison vector (OW vector): define the priority of each of the other criteria (j) relative to the worst criterion (W), and construct a row vector:

$${\mathit{A}}_{\mathit{w}}=\left[{\mathit{a}}_{\mathbf{1}\mathit{w}},{\mathit{a}}_{\mathbf{2}\mathit{w}},\dots ,{\mathit{a}}_{\mathit{j}\mathit{w}}\right]$$

(2)

Step 5: Linear programming solution: calculate the optimal weight vector by minimizing the maximum absolute difference (Formulas (3)–(5)).

$$\mathbf{m}\mathbf{i}\mathbf{n} \mathit{\xi }$$

$$\mathbf{s}.\mathbf{t}.\left|{\displaystyle \frac{{\mathit{w}}_{\mathit{B}}}{{\mathit{w}}_{\mathit{j}}}}-{\mathit{a}}_{\mathit{B}\mathit{j}}\right|\le \mathit{\xi }, \forall \mathit{j}$$

(3)

$$\left|{\displaystyle \frac{{\mathit{w}}_{\mathit{j}}}{{\mathit{w}}_{\mathit{w}}}}-{\mathit{a}}_{\mathit{j}\mathit{w}}\right|\le \mathit{\xi }, \forall \mathit{j}$$

(4)

$$\sum {\mathit{w}}_{\mathit{j}}=\mathbf{1},{\mathit{w}}_{\mathit{j}}\ge \mathbf{0}$$

(5)

In this linear model, $\mathit{\xi }$ is an indicator for comparing consistency, and a value close to zero indicates a high level of consistency. In line with existing research and application norms for this method [75], when $\mathit{\xi }$ < 0.1, the experts’ judgements on the importance of indicators can be considered to have acceptable consistency, with no obvious logical contradictions. It should be noted, however, this consistency check mainly mitigates “judgemental logical contradictions” rather than eliminating subjective deviations entirely.

Table 4. Fundamental Pairwise Comparison Scale [77].

The Intensity of Importance on an Absolute Scale	Definition	Explanation
1	Equal importance	Two activities contribute equally to the objective
3	Moderate importance of one over another	Experience and judgement strongly favour one option over another
5	Essential or strong importance	Experience and judgement strongly favour one activity over another
7	Very high importance	Activity is strongly favoured, and its dominance demonstrated
9	Extreme importance	The evidence favouring one activity over another is of the highest possible order of affirmation

Table 4. Fundamental Pairwise Comparison Scale [77].

The Intensity of Importance on an Absolute Scale	Definition	Explanation
1	Equal importance	Two activities contribute equally to the objective
3	Moderate importance of one over another	Experience and judgement strongly favour one option over another
5	Essential or strong importance	Experience and judgement strongly favour one activity over another
7	Very high importance	Activity is strongly favoured, and its dominance demonstrated
9	Extreme importance	The evidence favouring one activity over another is of the highest possible order of affirmation

The indicator weights calculated via BWM serve as the core input parameters for subsequent GI suitability assessment. Uncertainties in weight assignment (e.g., reasonable deviations in experts’ judgements) may exert potential impacts on the final suitability zoning results. To quantify such impacts and verify the robustness of the results, the subsequent sections will evaluate the effects of weight fluctuations and synergistic perturbations on GI suitability outcomes through single-factor perturbation analysis and multi-factor perturbation analysis.

3.5. Fuzzy Logic

In the geospatial analysis of GI system modelling, when classifying based on attribute objects, challenges around ambiguous classification boundaries often arise due to the continuity of natural hydrological processes and the spatial heterogeneity of urban underlying surfaces. Fuzzy logic can model uncertainties in data through membership functions, including data errors, transitional characteristics of categories (such as the gradual relationship from “highly suitable areas” to “moderately suitable areas”), and the subjectivity of expert experience [78,79]. Traditional data analysis often ignores data uncertainty, while fuzzy logic provides a more realistic and natural expression framework for multi-criteria spatial decision making [80]. The application of fuzzy logic in GIS involves two steps: first, fuzzy membership relationships; second, fuzzy overlay. Fuzzy membership functions convert crisp values into a fuzzy space ranging between 0 and 1. Fuzzy overlay, on the other hand, can combine the layers converted from crisp to fuzzy [81]. Formula (6) assigns a membership degree to each spatial data point in GIS:

$$\begin{gathered} \mathit{\mu }(\mathit{x})={\displaystyle \frac{\mathit{f}\left(\mathit{x}\right)-\mathit{f}\left({\mathit{x}}_{\mathbf{1}}\right)}{\mathit{f}\left({\mathit{x}}_{\mathbf{2}}\right)-\mathit{f}\left({\mathit{x}}_{\mathbf{1}}\right)}} \\ \left\{\begin{array}{c}\mathit{\mu }\left(\mathit{x}\right)=\mathbf{0} \mathbf{i}\mathbf{f} \mathit{f}\left(\mathit{x}\right)\le \mathit{f}\left({\mathit{x}}_{\mathbf{1}}\right)\\ \mathit{\mu }\left(\mathit{x}\right)=\mathbf{1} \mathbf{i}\mathbf{f} \mathit{f}\left(\mathit{x}\right)\ge \mathit{f}\left({\mathit{x}}_{\mathbf{2}}\right)\end{array}\right. \end{gathered}$$

(6)

Here, $\mathit{\mu }\left(\mathit{x}\right)$ refers to the fuzzy membership function; $\mathit{f}\left(\mathit{x}\right)$ is the membership degree of the function; $\mathit{f}\left({\mathit{x}}_{\mathbf{1}}\right)$ is the given minimum value; and $\mathit{f}\left({\mathit{x}}_{\mathbf{2}}\right)$ is the maximum value.

This study employs the GAMMA operator to fuse the standardized multi-criteria membership degree layers, thereby achieving high-precision and high-stability spatial data integration.

$${\mathit{\mu }}_{\mathit{\gamma }}=\prod _{\mathit{i}=\mathbf{1}}^{\mathit{n}}{\mathit{\mu }}_{\mathit{i}}^{\mathbf{1}-\mathit{\gamma }}{\left(\mathbf{1}-\prod _{\mathit{i}=\mathbf{1}}^{\mathit{n}}\left(\mathbf{1}-{\mathit{\mu }}_{\mathit{\gamma }}\right)\right)}^{\mathit{\gamma }}$$

(7)

Here, ${\mathit{\mu }}_{\mathit{\gamma }}$ is the fuzzy membership degree fused by the GAMMA operator, $\mathit{\mu }$ is the membership degree of the i-th standardized criterion layer, $\mathit{\gamma }$ is the parameter of the GAMMA operator, and $\mathit{n}$ is the number of criterion layers involved in the fusion.

When conducting analysis in a GIS environment, there are various fuzzy integration functions, with common ones including AND, OR, product, sum, and the GAMMA operator. Through comparative research, it was found that, when the parameter γ in the GAMMA function is set to 0.90, it demonstrates unique advantages in suitability analysis and evaluation [77].

4. Results

4.1. Weights of Evaluation Criteria

In the weight calculation based on the BWM method, experts identified the “best indicator” and “worst indicator” through their judgements, and they constructed the BO and OW vectors using a 1–9 scale (

Table 5. BO and OW Vectors Based on 1–9 Scale.

	Elevation	Slope	Aspect	NDVI	Soil Texture	Inundated Area	Runoff Corridor	Land Use	Traffic Buffer
Best-to-Others (BO)	8	6	9	3	7	1	2	3	5
Others-to-Worst (OW)	2	3	1	7	5	8	8	7	6

). Through this, the standard weights for green infrastructure (GI) suitability are calculated (

Table 6. The Weights of Criteria Calculated Using BWM for GI Suitability.

Criteria	Sub-Criteria	Weights (%)
Natural conditions	Elevation	4.52%
	Slope	6.88%
	Aspect	2.91%
	NDVI	11.44%
	Soil texture	5.68%
Flood security	Inundated area	32.13%
	Runoff corridor	15.73%
Social economy	Land use	12.44%
	Traffic buffer	8.26%

). By using the BWM linear optimization model, the consistency measure ξ is calculated to be 0.0917, which is less than the critical value of 0.1. This indicates that there is no obvious logical contradiction in experts’ judgements on the relative importance of indicators, further ensuring the reliability of the weight calculation.

To assess expert subjectivity, we conducted a sensitivity analysis: 100 random ± 1 perturbations of BO/OW vectors (simulating minor judgement deviations) and recalculated BWM weights and GI suitability. Key results: (1) core high-weight indicators “Inundated area” and “NDVI” showed small fluctuations, with maximum deviations from original values of 4.5% and 3.8%, respectively; (2) the total area of highly suitable GI regions varied by up to 5.2% across simulations, with a CV of 0.048. This confirms slight expert bias has limited, acceptable effects on BWM weights and final GI suitability results.

Based on the weights of GI suitability criteria calculated by the best–worst method (BWM) (Table 6), inundation areas are of the highest importance, followed by runoff corridors. Avoiding flood risks is the core basis for GI planning, and the protection and restoration of natural hydrological channels are also crucial for rainwater and flood regulation. The high weight of land use reflects the need for spatial co-ordination between human activities and ecological land, requiring a balance between development intensity and the service scope of GI. The transportation network has a dual impact on GI. The relatively high weight of vegetation highlights the fundamental role of the ecological background in ensuring the sustainability of GI. Slope directly affects soil and water conservation capacity and construction suitability, and its high weight reflects its constraint on the spatial layout of GI. Other evaluation criteria are less critical, relatively speaking.

4.2. Suitability Analysis for GI

The analysis of the GI suitability evaluation results for Zhengzhou (Figure 4) is as follows: In terms of area proportion, the highly suitable areas (shown in blue) cover 512.7 km², accounting for 6.8% of the total area of Zhengzhou, and the moderately suitable areas (green) span 3089.9 km², accounting for 40.8%. Spatially, the highly suitable areas exhibit a pronounced pattern of clustering along rivers and lakes. The hydrological processes of these water systems help create favourable runoff regulation spaces, making such areas optimal sites for GI implementation.

To verify the impact of GAMMA operator parameters on evaluation results, three values (γ = 0.80, 0.90, 0.95) were tested using the same fuzzy membership layers (

Table 7. Sensitivity Analysis of GAMMA Parameter on GI Suitability Evaluation Results.

Suitability Class	Index Type	γ = 0.80	γ = 0.90 (Benchmark)	γ = 0.95
Highly Suitable	Area Proportion	12.5%	6.8%	5%
	Overlap with Benchmark	100%	-	74%
Moderately Suitable	Area Proportion	50.3%	40.8%	35.1%
	Overlap with Benchmark	86%	-	82%
Less Suitable	Area Proportion	35.5%	46.4%	50.6%
	Overlap with Benchmark	67%	-	93%
Not Suitable	Area Proportion	1.7%	6%	9.3%
	Overlap with Benchmark	28%	-	100%

Note: The area proportion is the ratio of each suitability zone’s area to the total area of the study region. Overlap is calculated as the proportion of the spatial intersection area of the corresponding suitability zone with the benchmark (γ = 0.90) zone to the area of the benchmark zone.

). Although the area proportions of each suitability level differ across γ values, the core spatial pattern remains highly stable: highly suitable areas overlap with the γ = 0.90 baseline by 100% (γ = 0.80) and 74% (γ = 0.95), while moderately suitable areas show over 80% overlap with the baseline. Core zones (e.g., the Yellow River beach area and riparian areas of Jialu River) do not shift spatially with γ, confirming the model’s robustness to reasonable γ fluctuations.

4.3. Sensitivity Analysis

4.3.1. Single-Factor Perturbation Sensitivity Analysis

Sensitivity analysis is an important method for verifying the robustness of the results of land suitability models and for identifying key driving factors [82,83]. This study adopted the one-factor-at-a-time (OAT) method to systematically evaluate the sensitivity response mechanism of the changes in the weights of each criterion on the spatial differentiation characteristics of suitability. The specific process is as follows. (1) Based on the standard weights obtained by the BWM method (Table 6), only one of the nine criteria is perturbed individually (increased by 20%, 40%, and 60%, respectively), and the weights of the remaining eight criteria are reduced proportionally; it is ensured that the total weight after adjustment remains ∑Wi = 1 (Wi is the weight of criterion i). A total of 27 weight combination schemes are generated. (2) An automated processing chain is built in the ArcGIS model builder platform, and the 27 weight combination schemes are dynamically coupled with the fuzzy overlay model (GAMMA function, γ = 0.90) to generate the suitability maps for the corresponding scenarios. (3) Taking the number of rasters of each suitability class under the original weights as the benchmark, the change rates of the number of rasters under the four suitability classes (“Highly suitable”, “Moderately suitable”, “Less suitable”, “Not suitable”), in different scenarios relative to the original results, are calculated and quantified, respectively.

The results (Figure 5) show that the “Inundated area” criterion (runs 16–18) exhibits large changes in the rate of grid cell count across all suitability classes. In particular, during the 18th run, the rate of change in grid cell count for the “Not suitable” class (red) exceeded 300%, which is significantly higher than that of other criteria. This indicates that changes in the weight of the “Inundated area” criterion can drastically alter the spatial extent of areas “Not suitable for GI implementation”. It also confirms that the “Inundated area” criterion is a key driving criterion, directly constraining the identification of GI-unsuitable areas. The “Runoff corridor” criterion has a significant impact on changes in grid cell count for the “Highly suitable” class (blue) and “Not suitable” class (model runs 19–21). Among these, the rate of change for the “Highly suitable” class exceeded 60% during the 21st run, with a magnitude far greater than that of natural criteria such as “Elevation” and “Slope”. This suggests that changes in the weight of the “Runoff corridor” criterion significantly influence both the highly suitable and unsuitable areas for GI implementation. Changes in the weight of the “Land use” criterion (runs 22–24) also have a considerable impact on the grid cell count of the “Highly suitable” and “Not suitable” classes. Although the magnitude is smaller than that of the “Inundated area” and “Runoff corridor” criteria, it is significantly higher than that of other criteria. As a representative of socio-economic criteria, changes in the weight of “Land use” reflect the dynamic and complex nature of socio-economic factors in regulating GI suitability. Criteria such as “Elevation” (run 1), “NDVI” (run 12), and “Soil texture” (run 15) each result in a rate of change in grid cell count exceeding 30% for the “Not suitable” class. These criteria serve as basic constraints for GI suitability; while their individual perturbation results in a lower rate of change compared to core criteria, they collectively influence changes in suitable areas through synergistic effects with the “Inundated area” and “Runoff corridor” criteria. For most other criteria, the rate of change in grid cell count across all suitability classes remains within ±30%, indicating that these criteria have a relatively weak impact.

Therefore, the suitability model proposed in this study is sensitive to criteria weights. Both the “Inundated area” criterion and the “Runoff corridor” criterion play crucial roles in GI suitability. Additionally, the sensitivity of topographic factors (elevation, slope, aspect) is relatively low. This confirms the spatial optimization logic of “hydrological processes dominating and topographic elements collaborating” in GI planning for the Yellow River Basin.

4.3.2. Multi-Factor Perturbation Sensitivity Analysis

To further capture the interaction effects among key factors and supplement the evaluation of the stability of GI suitability results under the synergistic perturbation of multiple factors, this study designed a multi-factor random perturbation experiment based on the core driving factors identified by OAT. The specific design and analysis are as follows. With reference to the sensitivity ranking of key factors in OAT and the reasonable error range of expert weight assignment, the perturbation parameters of three core factors (inundation area, runoff corridors, and land use) were determined (

Table 8. Parameter Design for Multi-Factor Disturbance Experiment.

Criteria	Sub-Criteria	Benchmark Weight Value	Disturbance Range
Flood security	Inundated area	32.13%	±10%
Flood security	Runoff corridor	15.73%	±10%
Socio-economic	Land use	12.44%	±8%

). Using the random number generation function (RAND function) in Excel, 50 sets of independent “inundation area weight–runoff corridor weight–land use weight” combinations were generated within the above perturbation range. Each set of combinations underwent “weight normalization” to ensure the sum of weights equals 1.

The 50 sets of weight combinations were input into the GI suitability fuzzy overlay model (with γ = 0.90 kept constant), respectively, to generate suitability rasters for corresponding scenarios. Taking “the coefficient of variation (CV), variance, independent factor contributions, and synergistic factor contributions related to each suitability class” as core indicators, the result stability under multi-factor perturbation was evaluated (

Table 9. Multi-Factor Perturbation on GI Suitability.

Suitability Class	CV	Variance	Independent Factor Contribution (%)	Synergistic Factor Contribution (%)
Highly suitable	0.165	0.0102	Runoff corridor (38%); Land use (32%)	Runoff corridor × Land use (28%)
Moderately suitable	0.103	0.0045	Inundated area (25%); Land use (19%)	Inundated area × Land use (9%)
Less suitable	0.087	0.0032	Runoff corridor (18%); Inundated area (14%)	Runoff corridor × Inundated area (6%)
Not suitable	0.213	0.0182	Inundated area (65%); Land use (12%)	Inundated area × Land use (25%)

).

Results show that unsuitable areas exhibit the highest coefficient of variation (CV = 0.213) and variance (0.0182), remaining the most sensitive under multi-factor synergistic perturbation. Variance partitioning analysis indicates that inundated areas independently contribute 65% of the total variance, with an additional 25% contributed by their interaction with land use. This is consistent with the findings of the one-factor-at-a-time (OAT) method, confirming that inundated areas dominate the identification of unsuitable areas. Highly suitable areas have a coefficient of variation of 0.165 and a variance of 0.0102, showing significant fluctuations. Interaction index analysis reveals a notable synergistic effect between runoff corridors and land use, which accounts for 28% of the variance: runoff corridors independently contribute 38%, land use contributes 32%, and the synergistic effect of the two accounts for 28% of the variance in highly suitable areas, verifying their synergistic importance. Moderately suitable areas have a coefficient of variation of approximately 0.103, with interaction effects contributing about 12% of the variance. These areas exhibit stability while still reflecting the impact of factor interactions. These results confirm that the key factors identified by the OAT method remain dominant in multi-factor scenarios, and factor interactions amplify the impact on extreme suitability levels.

4.4. Spatial Matching Verification

The Zhengzhou Gardens Bureau (https://ylj.zhengzhou.gov.cn/) has released the Zhengzhou Urban Green Space System Planning (2013–2030). The planning defines the scope of Zhengzhou’s ecological control zones, which consist of two components: ecological patches and ecological corridors. Ecological patches refer to concentrated green spaces with ecological service functions (such as parks and wetlands), while ecological corridors are linear ecological spaces that connect patches (such as green belts along rivers and road shelterbelts). The regional boundaries of ecological patches and ecological corridors are marked in the Planning Map of Ecological Patches and Large-Scale Ecological Corridors in Zhengzhou. From the perspective of GI theory, the ecological control zones (including ecological patches and corridors) delineated in the plan are precisely the core carriers of GI. Therefore, by comparing the spatial overlap between ecological patches, corridors, and the GI suitability evaluation map, the compatibility and rationality of the suitability model with existing planning in terms of ecological space layout can be quantitatively verified (Figure 6).

First, download Zhengzhou’s “Ecological Patches and Large-Scale Ecological Corridors Plan Map” from Zhengzhou Gardens Bureau’s website. Import it into GIS for georeferencing and projection transformation (unified to WGS-1984 UTM Zone 50 N), then resample to 30 m × 30 m resolution to match the GI suitability evaluation map in spatial resolution and co-ordinate system. Second, convert the plan map to a binary raster (1 = ecological control zones, 0 = others) via RGB threshold segmentation. Finally, conduct spatial overlay analysis of the binary raster and suitability results, and construct a confusion matrix by cross-tabulating ecological control zone areas in each suitability level (

Table 10. Confusion Matrix of Spatial Matching. (Unit: km²).

	Ecological Control Zone	Non-Ecological Control Zone	Total
Highly suitable	273.6	239.1	512.7
Moderately suitable	1922.5	1167.4	3089.9
Less suitable	511.7	3000.6	3512.3
Not suitable	76.7	374.4	451.1
Total	2784.5	4781.5	7566

). Clarify the ideal matching relationship between ecological control zones and suitability levels, define a quadratic weighting matrix (

Table 11. Quadratic Weighted Matrix.

	Ecological Control Zone (Planning Objective: Adapt to High Suitability)	Non-Ecological Control Zone (Planning Objective: Adapt to Low Suitability)
Highly suitable	1	1/9
Moderately suitable	1	1/4
Less suitable	1/4	1
Not suitable	1/9	1

Note: The quadratic weights are set based on the “square of grade differences”. The larger the grade difference, the smaller the weight and the lower the matching degree.

, weights based on “square of level differences: larger differences = smaller weights”), and calculate the weighted Kappa coefficient to quantify spatial consistency between the plan and model.

The weighted Kappa coefficient was 0.463, indicating a moderate level of spatial consistency between the planned ecological control zones and the GI suitability model. It should be noted that the possibility of “random coincidence” must still be excluded from the above consistency results. For this purpose, this study constructed a null model (random suitability model) for comparative analysis, with the steps as follows. Based on the original GI suitability evaluation results, while keeping the area of each suitability level unchanged, the Randomly Reclassify Raster tool in ArcGIS was used to randomly assign 30 m × 30 m raster cells to four suitability levels, generating 100 random suitability maps. For these 100 randomized GI suitability evaluation maps, the previous analysis process was repeated one by one, and finally 100 weighted Kappa values for the randomized scenarios were obtained. The weighted Kappa value of the SCS–GIS–MCDM model was compared with the distribution of the 100 random Kappa values (Figure 7).

The results of the null model verification show that the weighted Kappa values of the 100 randomized scenarios range from 0.02 to 0.35, with an average of 0.16. As shown in Figure 7, the weighted Kappa value of the SCS–GIS–MCDM model in this study far exceeds the upper limit of the random value distribution, which proves that the spatial consistency between ecological control zones and GI suitability is not a random coincidence. Notably, even after excluding random coincidence, the current moderate spatial consistency does not allow direct equating of GI-suitable areas with actual GI implementation areas, mainly due to limitations like static land use data and unincorporated land costs.

From the perspective of spatial matching details, 239.1 km² of highly suitable areas are not included in the protection scope of ecological control zones, accounting for 46.6% of the total area of highly suitable areas; 76.7 km² of ecological control zones fall into GI unsuitable areas, accounting for 17% of the total area of unsuitable areas. To further clarify the spatial agglomeration characteristics of mismatched areas between ecological control zones and GI suitability, Figure 8 presents the district-level decomposition of mismatched areas.

In Figure 8, Zhongmu County and Huiji District have relatively high proportions of GI highly suitable areas not included in ecological control zones (shown by green bars), and this deviation can be verified by field observations. Data from the Field Scientific Observation and Research Station for Eco-Hydrological Evolution of the Floodplain in the Lower Yellow River show that a 17.5 km² area in the core section of the Yellow River floodplain—an area inherently eligible for GI highly suitable zoning due to its status as a key river ecological corridor (with open green space and hydrological connectivity that support runoff regulation)—was not incorporated into the ecological control zone. The delineation of traditional ecological control zones does not integrate key hydrological and soil property factors such as rainwater runoff depth and soil permeability, which may be one of the core reasons for the mismatch between their spatial scope and actual ecological needs.

As also seen in Figure 8, suburban counties including Xinmi City, Xinzheng City, Dengfeng City, Zhongmu County, and Gongyi City have significantly higher proportions of ecological control zones falling into GI unsuitable areas (shown by orange bars). Against the backdrop of urban renewal, land use in suburban areas undergoes more dynamic changes. However, the static data (e.g., land use, soil texture) relied on by the model in this study cannot fully reflect the dynamic changes in GI suitability in suburban counties caused by urban expansion and human activities. Meanwhile, the model insufficiently considers socio-economic factors such as land cost and green space accessibility. These two aspects together result in deviations from the actual GI suitability requirements.

5. Discussion

As a key node in the ecological security pattern of the middle and lower reaches of the Yellow River Basin, Zhengzhou has seen its urbanization rate exceed 81%. In this process, the surge in underlying surface imperviousness, coupled with extreme rainstorm events, has made it difficult for traditional grey infrastructure to address the “rainstorm-city” contradiction [84]. Against this backdrop, the scientific planning and layout of GI have become a critical path to resolve the conflict between the basin’s ecological vulnerability and urban development needs [12]. In recent years, Zhengzhou’s decision-makers have incorporated blue–green space system planning into the core framework of urban development strategies [85], reflecting the city’s urgent demand for evidence-based blue–green space planning.

From the quantitative results of Zhengzhou’s GI suitability evaluation, highly suitable areas for GI account for 6.8% of Zhengzhou’s total area, while moderately suitable areas account for 40.8%. This indicates that nearly half of Zhengzhou has good potential for GI layout. Spatially, highly suitable areas exhibit a significant clustering pattern along rivers and lakes. It is recommended to restore the green attributes of these areas and prioritize ecological protection projects and the construction of green spaces within them. Moderately suitable areas are primarily concentrated in suburban ecologically resource-rich zones, such as the western mountainous regions, eastern river systems, and southern woodlands. It is strongly advised to enhance the quality of GI in these areas while ensuring the protection of existing ecological resources. The practical solutions for GI construction proposed in this study align with the goal of “constructing a blue–green network” outlined in the Zhengzhou Territorial Spatial Ecological Restoration Plan (2021–2035) [86].

Figure 9 presents the proportion of GI suitability grades across Zhengzhou’s six municipal districts, one county, and five county-level cities. The spatial differentiation of suitability reflects the combined effects of natural conditions and human activities: Huiji District and Zhongmu County exhibit the highest proportions of highly suitable areas, which is not only attributed to their proximity to the Yellow River Ecological Belt and Zhengzhou-Kaifeng Ecological Corridor but also to their dominant soil texture, facilitating rainwater infiltration and runoff regulation. In contrast, municipal districts such as Jinshui District, Guancheng Hui District, and Erqi District have a generally high proportion of low-suitability areas, reflecting the current situation where ecological space in urban built-up areas is under pressure. For such regions, the core idea of GI construction should focus on “stock optimisation” to enhance the ecological resilience of built-up areas.

In conclusion, this study provides a technical reference for the blue–green space planning of the Yellow River Basin. Based on natural hydrological laws and combined with socio-economic needs, the ideas and methods proposed in this study for the “water security crisis” faced by cities in the Yellow River Basin, such as Zhengzhou, are expected to support the region in achieving multi-dimensional sustainable development in ecology, economy, and society [87]. However, this study has three limitations requiring further improvement. First, socio-economic data lack sufficient spatial granularity—especially refined data on GI implementation costs and land ownership—limiting in-depth project feasibility analysis and weakening the practical guiding value of evaluation results for actual GI planning. Second, the model prioritizes hydrological processes but inadequately explains the corresponding ecological space mechanism. Third, it relies on static 2021 land use data and excludes rainfall spatiotemporal heterogeneity, reducing long-term applicability and scenario adaptability.

6. Conclusions and Policy Implications

The site selection of GI should not be blind or succumb to uncontrolled urban sprawl. Methods based on hydrological process simulation are conducive to alleviating urban waterlogging and achieving sustainable development. This study integrated the SCS–GIS–MCDM system to identify the distribution characteristics of highly suitable GI areas in Zhengzhou, and it confirmed that “Inundated area” and “Runoff corridor” are the key driving factors affecting GI suitability. The constructed reusable GI suitability assessment process provides technical references for GI planning in similar regions; meanwhile, the localised calibration of hydrological model parameters and multi-method integration have optimized the application effect of GIS in eco-hydrological analysis. The main conclusions of this study are as follows:

(1)

6.8% of the land in Zhengzhou is highly suitable for developing GI, while 6% of the land is not suitable. These proportions are not fully comparable to existing studies, mainly due to differing scales and indicator frameworks.

(2)

Sensitivity analysis shows that elements of hydrological process simulation are key influencing factors, while elements such as elevation and slope aspect have weak sensitivity.

(3)

Confusion matrix comparison between ecological control zones (from green space system planning) and GI suitability zones indicates moderate spatial compatibility, but deviations persist due to static data limitations, insufficient resolution, and modelling assumptions.

Policy implications are as follows:

(1)

Guide spatial planning based on GI suitability zoning and formulate ecological management and control systems. Designate Huiji District and Zhongmu County as “Priority Control Units for Blue–Green Spaces”, incorporate them into the “Yellow River Basin Ecological Protection Plan”, clarify the spatial boundaries for GI construction, and prohibit urban expansion from encroaching on ecologically sensitive areas.

(2)

Incorporate the SCS–GIS–MCDM system into blue–green space planning, and promote the quantitative implementation of the three-dimensional constraints of “water security–natural background–socio-economy”.

(3)

Promote the upgrading of China’s river basin governance from the single goal of flood control to synergistic development. With GI planning as a link, strive to build a paradigm for sustainable urban and community development.

The above conclusions further confirm the applicability of the SCS–GIS–MCDM system in GI suitability modelling, while it is worth noting that the study still has limitations as discussed earlier—including constraints from data resolution, potential subjectivity in expert-driven weight determination, and moderate model validation consistency—which provide directions for future research optimization.

Future research can advance in three aspects. First, conduct cross-scale GI planning synergy research using high-resolution remote sensing data (e.g., Sentinel-2, 10 m resolution) to build a coupled model covering community micro-scales and watershed macro-scales, addressing the disconnection between small-scale suitability and large-scale watershed governance. Second, improve the multi-decision assessment model by incorporating accessible socio-economic indicators (GDP from local statistical yearbooks, population density from the National Population Census Database, green space accessibility from urban planning datasets) to distinguish urban–rural social needs and to enhance the model’s explanatory power and regional adaptability. Third, focus on transforming “suitability to implementation priority” (a core step to boost actionability) via two steps: (1) supplementing data—district-level socio-economic stats (e.g., per capita income, partially overlapping with the second aspect but repurposed for cost proxies) and new regional green space construction cost standards from housing departments; (2) developing a proof-of-concept framework (Figure 10) to quantify priority via a composite index integrating suitability, receptor proximity, and cost.