Method and system for evaluating nonlinear influence of greenbelt landscape pattern on rainfall flood regulation service

Abstract

The invention discloses a method and a system for evaluating the nonlinear influence of a greenbelt landscape pattern on a rainfall flood regulation service. The method comprises the following steps: step 1, constructing a basic database; 2, processing data variables; 3, constructing a nonlinear model: taking a green land landscape pattern factor as a core explanation variable, taking a rainfall flood regulation service index as a response variable, introducing a natural environment factor and a social economic factor as control variables, training the random forest model, and constructing a rainfall flood regulation service prediction model; 4, key factors are identified, wherein the top-ranked factors are regarded as key factors playing a role in rainfall flood adjustment service; step 5, analyzing a key threshold: performing visual analysis on a nonlinear relationship between the key factor and the rainfall flood adjustment service by adopting a partial dependency graph algorithm; and by observing positive and negative influence trends and change rates of the key factors on the rainfall flood regulation service indexes in different value intervals, identifying potential key thresholds.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of ecological environment, and particularly relates to a method and system for evaluating nonlinear influence of green space landscape pattern on rainwater flood regulation service. BACKGROUND

[0002] Under the dual effects of climate change and urbanization, urban flood disasters are experiencing multidimensional evolution from frequency, intensity to spatiotemporal pattern, and their risk situation is becoming more complex and severe. As a key natural element for improving rainwater ecological resilience, green space can effectively regulate the processes of rainwater infiltration, retention, storage and collection, thereby providing important rainwater regulation services. However, in the process of rapid urbanization, high-intensity development and construction activities have led to the reduction and fragmentation of green space, thereby disrupting the original natural hydrological cycle of the city and weakening the rainwater regulation function of the urban green space system. Therefore, in-depth research on the optimization path of green space spatial pattern under the guidance of rainwater resilience enhancement has important theoretical and practical value for enhancing rainwater regulation service function, reducing urban flood risk and ensuring the safety of human settlements.

[0003] Green space landscape pattern reflects the combination and configuration of green space elements of different areas and shape characteristics in space, and the differences in its spatial components and configurations will produce different rainwater regulation effects, which are the key factors affecting rainwater flow and its spatial distribution. Current research on the optimization of green space landscape pattern for rainwater management either based on multiple spatial scales or from the dynamic dimension of space and time, mostly uses correlation analysis or linear regression methods to analyze the relationship between the two, providing scientific basis for rainwater ecological regulation practice, but still has certain limitations. Existing research mainly evaluates the influence of green space landscape pattern factors on rainwater regulation service based on linear assumption, and lacks systematic analysis of the nonlinear correlation mechanism and threshold effect between the two, resulting in limited model applicability and insufficient prediction accuracy when quantifying their relationship. Existing research has shown that the influence of green space landscape pattern on hydrology often presents significant nonlinear characteristics, i.e. when a certain landscape index reaches a certain threshold, the rainwater regulation function may respond with a jump or tend to be saturated. It is worth noting that current green space spatial pattern optimization practice mostly suggests a one-size-fits-all approach of "increasing or decreasing", without specifying how and how much to increase, and the lack of threshold control mode may cause a gap between "what should be" and "what is".

[0004] In summary, the existing research has the following defects:

[0005] 1. Insufficient analysis of nonlinear influence: existing research generally relies on linear assumption in the modeling process, ignoring the nonlinear correlation characteristics of ecological hydrological processes, and it is difficult to accurately reveal the complex influence of landscape pattern change on rainwater regulation service.

[0006] 2. Lack of threshold effect analysis: Existing research has failed to fully reveal the key thresholds of green space landscape pattern affecting rainwater regulation services, making it difficult to provide refined green space regulation recommendations.

[0007] Therefore, it is necessary to further explore the non-linear impact of green space landscape pattern on rainwater regulation services, identify key factors and threshold effects, and thus provide more refined guidance strategies and more effective ecological practice paths for urban rainwater management. SUMMARY

[0008] To solve the above technical problems existing in the prior art, the present application provides an evaluation method and system for the non-linear impact of green space landscape pattern on rainwater regulation services. The present application explores the quantitative relationship between green space landscape pattern and rainwater regulation services by constructing an interpretable machine learning model, identifies key influencing factors and visualizes non-linear relationships, and explores their key thresholds.

[0009] The technical solution adopted by the present application is:

[0010] The first aspect of the present application relates to an evaluation method for the non-linear impact of green space landscape pattern on rainwater regulation services, characterized by comprising the following steps:

[0011] Step 1: Building a basic database

[0012] The GIS software is used to unify the coordinates, correct the accuracy and project the data, to ensure the high consistency of different source data in space;

[0013] Step 2: Process data variables

[0014] Step 21, considering the hydrogeological conditions, water system distribution, land use and urban administrative boundary factors, the water management unit is divided;

[0015] Step 22, based on the SCS-CN model, the rainwater regulation service of each water management unit is quantified;

[0016] Step 23, referring to existing research, green space patches are extracted from each water management unit, and a green space landscape pattern index system is constructed from three dimensions of scale, shape and structure;

[0017] Step 3: Building a non-linear model

[0018] Taking the green space landscape pattern factor as the core explanatory variable and the rainwater regulation service index as the response variable, while introducing natural environmental factors and social economic factors as control variables, the random forest model is trained to construct a rainwater regulation service prediction model;

[0019] Step 4: Identify key factors

[0020] The green landscape pattern factors are ranked according to the importance of the rainwater regulation service prediction model output, and the top-ranked factors are regarded as key factors that play a key role in rainwater regulation services;

[0021] Step 5: Analyzing key thresholds

[0022] By using the partial dependence plot algorithm, the nonlinear relationship between the key factor and the rainwater regulation service is visualized and analyzed; by observing the positive and negative influence trend and change rate of the key factor on the rainwater regulation service index in different value intervals, the potential key threshold is identified.

[0023] Among them, the green landscape pattern factors are selected from three dimensions of scale, shape and structure. The scale characteristics include patch density (PD), maximum patch area proportion (LPI) and average patch size (AREA_MN); the shape characteristics include landscape shape index (LSI), average shape index (SHAPE_MN) and edge density (ED); the structure characteristics include fragmentation (DIVISION), connectivity (CONNECT) and cohesion (COHESION); the above factors are calculated by the landscape pattern analysis software fragstats. In addition, natural environmental factors (vegetation index, slope), social and economic factors (construction land proportion, population density) are selected as control variables.

[0024] Further, in step S1, the basic data includes DEM elevation data, soil data, normalized vegetation index data, administrative boundary data, population density data, drainage partition data and land use data.

[0025] Further, in step S1, the land use data is based on the "geospatial data cloud" platform to select the Landsat8 remote sensing image data map of Hangzhou in May 2023, and ENVI5.1 software is used for atmospheric correction and projection definition; referring to the LUCC classification system, the maximum likelihood classification method and the combination of field investigation and visual interpretation are used to divide the land in the study area into 6 types according to the use type, i.e. construction land, forest land, grassland, water area, farmland and unused land.

[0026] Further, in step S2, the rainwater regulation service space quantification is mainly based on the SCS_CN model, and the specific calculation formula is as follows:

[0027] C = Δv x (0.001 x P h x A) -1 x 100% (1)

[0028]

[0029] Where C represents the surface runoff regulation rate, i.e., the stormwater regulation service evaluation result; v represents the surface runoff storage capacity (mm); P h A represents the rainfall (mm); A represents the area of ​​the study unit (m²). 2 );Q i Q represents the surface runoff (mm) of a 100% impermeable surface; n A represents the actual surface runoff (mm) of the underlying surface; i pixel area (m 2 The higher the surface runoff regulation rate C, the higher the stormwater regulation service provided by the green space.

[0030] Furthermore, the SCS-CN hydrological model was used to simulate surface runoff Q under different rainfall scenarios. The specific calculation formula is as follows:

[0031]

[0032] I a =λ×S (5)

[0033] Where Q is surface runoff; P is precipitation; S is the maximum potential retention rate of precipitation; CN is a dimensionless parameter characterizing surface runoff capacity, ranging from 0 to 100; I a λ represents the initial infiltration rate of precipitation; λ is the soil infiltration coefficient, usually taken as an empirical value of 0.2.

[0034] Furthermore, based on the area proportion of each land use within the research water management unit, the comprehensive CN value of the unit is calculated using a weighted method, as shown in the following formula:

[0035]

[0036] Among them, CN1, CN2, CN3, CN4, CN5, and CN6 are the CN values ​​corresponding to construction land, grassland, bare land, forest land, cultivated land, and water area; a1, a2, a3, a4, a5, and a6 are the area proportions of the six land use categories within the unit.

[0037] Furthermore, in step S3, a structured dataset constructed based on multiple water management units is input into the model for training. The structured dataset contains response variables, explanatory variables, and control variables.

[0038] During the model training phase, stratified random sampling is used to divide the dataset into a training set and a test set in a 7:3 ratio, where the training set is used for model learning and the test set is used for independent performance evaluation.

[0039] In the parameter optimization phase of the model, based on the Scikit-learn library, the grid search algorithm GridSearchCV is used in combination with 10-fold cross-validation to optimize the model hyperparameters, and the key parameters n_estimators and max_depth are adjusted;

[0040] In the model verification phase, the model performance is evaluated on the training set and the test set respectively, and the determination coefficient R 2 And the mean absolute error MAE index is used to measure the fitting ability and prediction accuracy of the model.

[0041] Further, in step S4, by extracting and sorting the importance scores of each feature variable, those factors that contribute more to the model prediction are identified, so as to reveal the key factors of green landscape pattern affecting the rainwater regulation service; wherein the fragmentation DIVISION, the maximum patch area ratio LPI, the average patch size AREA_MN, the cohesion COHESION, the edge density ED and the average shape index SHAPE_MN are the key factors.

[0042] Further, in step S5, the PDPs (Partial Dependence Plots) algorithm is used for visual analysis of the key factors; by drawing the relationship curve between the different values of the feature variables and the average of the model prediction values, the threshold effect of the rainwater regulation service is identified: 1) when the curve presents an obvious inflection point or mutation, it indicates that there is a threshold effect, and the feature value corresponding to the inflection point is the critical threshold; 2) the change of the curve slope reflects the sensitivity change of the feature influence.

[0043] The second aspect of the application relates to an evaluation system for the nonlinear influence of green landscape pattern on rainwater regulation service, characterized by comprising:

[0044] A data acquisition and processing module is used to obtain a structured data set in a target research area; wherein the structured data set includes a rainwater regulation service index, a green landscape pattern index, natural environment data and social and economic data;

[0045] A model construction and training module is used to construct a nonlinear model, taking the rainwater regulation service index as the response variable, the green landscape pattern index as the explanatory variable, and the natural environment data and social and economic data as the control variable; the random forest model is trained to construct a rainwater regulation service prediction model, and the rainwater regulation service prediction model is visually explained according to the PDPs (Partial Dependence Plots) algorithm;

[0046] A result analysis and interpretation module is configured to sort green land landscape pattern factors of the target research region according to the result of the influence of the green land landscape pattern on the rain flood regulation service, and identify a threshold value, to obtain a key factor and a key threshold value of the green land landscape pattern influencing the rain flood regulation service, and the analysis result can be directly used for scientific guidance of optimized construction and practice of the urban green land landscape pattern.

[0047] The technical concept of the present application is that: based on the random forest algorithm, an association model between the green land landscape pattern and the rain flood regulation service is constructed, and by drawing a partial dependence plot between the feature variable and the rain flood regulation service, the key threshold value of each green land landscape pattern factor is quantitatively analyzed, and the nonlinear influence mechanism of the rain flood regulation service is determined.

[0048] The innovation of the present application lies in:

[0049] (1) Modeling method based on random forest algorithm: the random forest algorithm is used to establish a nonlinear response model between the green land landscape pattern factor and the rain flood regulation service index, which breaks through the limitation of traditional linear modeling on the representation of complex ecological service mechanism.

[0050] (2) Green land landscape pattern key threshold value identification method: by analyzing the model output result, the influence inflection point or jump interval of the green land landscape pattern factor on the rain flood regulation service in different value intervals is identified, and the nonlinear action mechanism is revealed.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] 1. Accurate identification of key control factors

[0053] The present application can clearly identify and quantify the key factors influencing the rain flood regulation service by sorting the importance of the green land landscape pattern factors, and provides a quantitative decision basis for the optimization of rain flood regulation.

[0054] 2. Threshold effect quantitative analysis

[0055] In the present application, the application of the random forest model PDPs algorithm makes the nonlinear relationship and threshold effect between the green land landscape pattern and the rain flood regulation service intuitively presented, which provides a theoretical basis for further understanding and optimizing the green land configuration, so as to realize more efficient use of green land space. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The flowchart of the method of the present application is shown in the figure;

[0057] Figure 2 The flowchart of the embodiment of the present application is shown in the figure;

[0058] Figure 3 The key factor identification diagram of the embodiment of the present application is shown in the figure;

[0059] Figure 4 a: the response curve of fragmentation index and the stormwater regulation service - with the increase of fragmentation, the stormwater regulation service as a whole shows a downward trend, and when the fragmentation is less than 0.64, the stormwater regulation service as a whole is at a higher level.

[0060] Figure 4 b: the response curve of the maximum patch area and the stormwater regulation service - with the increase of the maximum patch area ratio, the stormwater regulation service as a whole shows an upward trend, and when the maximum patch area ratio is greater than 65.83, the stormwater regulation service is at a higher level.

[0061] Figure 4 c: the response curve of edge density and the stormwater regulation service - with the increase of edge density, the stormwater regulation service as a whole shows an upward trend, and when the edge density is greater than 166.98, the stormwater regulation service is in a rapid growth state.

[0062] Figure 4 d: the response curve of cohesion index and the stormwater regulation service - as a whole, it shows a trend of first gentle and then sharp rise, and when the cohesion is greater than 97.36, the stormwater regulation service capacity is significantly improved.

[0063] Figure 4 e: the response curve of average patch area and the stormwater regulation service - as a whole, it shows a trend of first steep rise and then stable, and when the average patch area is greater than 11.81, the stormwater regulation service is at a higher level.

[0064] Figure 4 f: the response curve of average shape index and the stormwater regulation service - as a whole, it shows a trend of first decrease and then slow rise, and when the average shape value is greater than 1.29, the stormwater regulation service is at a higher level.

[0065] Figure 5 The structural schematic diagram of the evaluation system of the present application. DETAILED DESCRIPTION

[0066] The specific embodiments of the embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application.

[0067] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0068] The present application will be described in detail below with reference to the accompanying drawings and in combination with exemplary embodiments.

[0069] Embodiment 1

[0070] The application discloses an evaluation method for nonlinear influence of green space landscape pattern on rain flood regulation service.

[0071] Step 1: Constructing a basic database

[0072] The GIS software is used for coordinate unification, precision correction and projection conversion of the basic data, so as to ensure high consistency of different source data in space.

[0073] Step 2: Processing data variables

[0074] Step 21: The water management unit is demarcated by comprehensively considering hydrogeological conditions, water system distribution, land utilization and urban administrative boundary factors;

[0075] Step 22: The rain flood regulation service of each water management unit is quantified based on the SCS-CN model;

[0076] Step 23: According to existing research, green space patches are extracted from each water management unit, and a green space landscape pattern index system is constructed from three dimensions of scale, shape and structure;

[0077] Step 3: Constructing a nonlinear model

[0078] The random forest model is trained by taking the green space landscape pattern factor as a core explanatory variable, the rain flood regulation service index as a response variable, and simultaneously introducing natural environment factors and social economic factors as control variables, so as to construct a rain flood regulation service prediction model;

[0079] Step 4: Identifying key factors

[0080] According to the importance of the green space landscape pattern factor output by the rain flood regulation service prediction model, the variables are sorted, and the factors ranking in the front are regarded as key factors playing a key role in the rain flood regulation service;

[0081] Step 5: Analyzing key threshold values

[0082] The partial dependence plot algorithm is used to visually analyze the nonlinear relationship between the key factor and the rain flood regulation service; by observing the positive and negative influence trend and change rate of the key factor on the rain flood regulation service index in different value intervals, potential key threshold values are identified.

[0083] Among them, the green space landscape pattern factors are selected from three dimensions of size, shape and structure. The size characteristics include patch density (PD), largest patch index (LPI) and mean patch size (AREA_MN); the shape characteristics include landscape shape index (LSI), mean shape index (SHAPE_MN) and edge density (ED); the structure characteristics include division (DIVISION), connectivity (CONNECT) and cohesion (COHESION); the above factors are calculated by the landscape pattern analysis software fragstats. In addition, natural environmental factors (vegetation index, slope), social and economic factors (construction land proportion, population density) are selected as control variables.

[0084] In the embodiment, in step S1, the basic data includes DEM elevation data, soil data, normalized vegetation index data, administrative boundary data, population density data, drainage partition data and land use data.

[0085] In the embodiment, in step S1, the land use data is based on the "geospatial data cloud" platform to select Landsat8 remote sensing image data of Hangzhou in May 2023, and ENVI5.1 software is used for atmospheric correction and projection definition; referring to the LUCC classification system, the maximum likelihood classification and the combination of field investigation and visual interpretation are used, and the land in the study area is divided into 6 types according to the use type, i.e. construction land, forest land, grassland, water area, cultivated land and unused land.

[0086] In the embodiment, in step S2, the rainwater regulation service space quantification is mainly based on the SCS_CN model, and the specific calculation formula is as follows:

[0087] C = Δv x (0.001 x P h x A) -1 x 100% (1)

[0088]

[0089] Among them, C is the surface runoff regulation rate, i.e. the rainwater regulation service evaluation result; v is the surface runoff storage capacity (mm); P h is the rainfall (mm); A is the area of the study unit (m 2 ); Q i is the surface runoff of the proportion of 100% impermeable surface (mm); Q n is the actual surface runoff of the underlying surface (mm); A i is the pixel area (m 2 ); the higher the surface runoff regulation rate C is, the higher the rainwater regulation service provided by the green space is.

[0090] In this embodiment, the SCS-CN hydrological model is used to simulate the surface runoff Q under different rainfall scenarios, and the specific calculation formula is as follows:

[0091]

[0092] I a = λ × S (5)

[0093] Where Q is the surface runoff; P is the precipitation; S is the maximum potential retention rate of precipitation; CN is a dimensionless parameter representing the surface runoff capacity, with a value range of 0-100; I a is the initial infiltration of precipitation; λ is the soil infiltration coefficient, usually taking the empirical value 0.2;

[0094] In this embodiment, based on the area proportion of each land use in the water management unit, the unit comprehensive CN value is calculated by the weighted method, and the specific formula is as follows:

[0095]

[0096] Where CN1, CN2, CN3, CN4, CN5, CN6 are the CN values of construction land, grassland, bare land, forest land, farmland and water area respectively; a1, a2, a3, a4, a5, a6 are the area proportions of the six types of land use in the unit.

[0097] In this embodiment, in step S3, the structured dataset constructed based on multiple water management units is input into the model for training, and the structured dataset contains response variables, explanatory variables and control variables;

[0098] In the model training stage, the data set is divided into training set and test set in the proportion of 7:3 by stratified random sampling, wherein the training set is used for model learning, and the test set is used for independent performance evaluation;

[0099] In the parameter optimization stage of the model, based on the Scikit-learn library, the grid search algorithm GridSearchCV combined with 10-fold cross-validation is used to optimize the model hyperparameters, and the key parameters n_estimators and max_depth are adjusted;

[0100] In the model validation stage, the model performance is evaluated on the training set and test set respectively, and the determination coefficient R 2 and the mean absolute error MAE index are calculated to measure the fitting ability and prediction accuracy of the model.

[0101] In the embodiment, in step S4, by extracting and sorting the importance scores of each feature variable, factors that contribute more to the model prediction are identified, thereby revealing the key factors of green space landscape pattern affecting the rainwater regulation service; wherein the fragmentation DIVISION, the maximum patch area proportion LPI, the average patch size AREA_MN, the cohesion COHESION, the edge density ED and the average shape index SHAPE_MN are the key factors.

[0102] In the embodiment, in step S5, the PDPs (Partial Dependence Plots) algorithm is used for visual analysis of the key factors; by drawing the relationship curve between the different values of the feature variables and the average of the model prediction values, the threshold effect of the rainwater regulation service is identified: 1) when the curve presents an obvious inflection point or mutation, it indicates that there is a threshold effect, and the feature value corresponding to the inflection point is the critical threshold; 2) the change of the curve slope reflects the change of the sensitivity of the feature influence.

[0103] Embodiment 2

[0104] The second aspect of the application relates to an evaluation system for the nonlinear influence of green space landscape pattern on rainwater regulation service, characterized in that it comprises:

[0105] A data acquisition and processing module is used to obtain a structured data set in the target research area; wherein the structured data set comprises a rainwater regulation service index, a green space landscape pattern index, natural environment data and social and economic data;

[0106] A model construction and training module is used to construct a nonlinear model, taking the rainwater regulation service index as the response variable, the green space landscape pattern index as the explanatory variable, and the natural environment data and social and economic data as the control variable; the random forest model is trained to construct a rainwater regulation service prediction model, and the rainwater regulation service prediction model is visually explained according to the PDPs (Partial Dependence Plots) algorithm;

[0107] A result analysis and interpretation module is used to sort and identify the importance of the green space landscape pattern factors in the target research area according to the influence of the green space landscape pattern on the rainwater regulation service, to obtain the key factors and key thresholds of the green space landscape pattern affecting the rainwater regulation service, and the analysis results can be directly used for scientific guidance of the optimization construction and practice of urban green space landscape pattern.

[0108] Embodiment 3

[0109] For the purpose, technical solutions and advantages of the present application to be clearer and more apparent, the present application is further described in detail below in combination with the drawings and examples, and Hangzhou City Center Eight District is selected as the research area in the examples. Please refer to Figure 2 A method for evaluating the nonlinear influence of green space landscape pattern on rainwater flood regulation service includes the following steps:

[0110] Step S1, construction of basic database

[0111] The basic database includes the following data: DEM elevation data, soil data, normalized vegetation index data, administrative boundary data, population density data, drainage partition data, and land use data. Among them, the land use data is based on the "Geospatial Data Cloud" platform to select Landsat8 remote sensing image data of Hangzhou City in May 2023, and ENVI5.1 software is used for atmospheric correction and projection definition. Further referring to the LUCC classification system, the research area is divided into 6 types according to the use type, i.e. construction land, forest land, grassland, water area, farmland and unused land, by using maximum likelihood classification and combining field investigation and visual interpretation.

[0112] Step S2, data variable processing

[0113] Considering the factors of water system, hydrogeological conditions, land use types and urban management implementation of Hangzhou City, combined with the rainwater drainage partition, catchment area and street administrative management unit of Hangzhou City, 110 water management units are divided. The specific division idea is as follows: if a street contains two or more drainage partition characteristics, it should be divided into multiple research units; the drainage partition and catchment area boundaries are used as the control unit boundaries in the mountainous and hilly areas with obvious catchment characteristics; if a drainage partition contains two or more catchment areas, it should be divided into multiple research units.

[0114] Among them, the spatial quantification of rainwater regulation service is mainly based on the SCS_CN model, and the specific calculation formula is as follows:

[0115] C = Δv x (0.001 x P h x A) -1 x 100% (1)

[0116]

[0117] In the formula, C is the surface runoff regulation rate, i.e. the evaluation result of rainwater regulation service; v is the surface runoff storage capacity (mm); P h is the rainfall (mm); A is the area of the research unit (m 2 ); Q i is the surface runoff of the proportion of 100% impermeable surface (mm); Q n is the actual surface runoff of the underlying surface (mm); Ai The pixel area (m 2 ) is the area of a single pixel. The higher the runoff adjustment rate, the higher the rainwater regulation service provided by the green space.

[0118] The SCS-CN model developed by the US Soil Conservation Service is widely used in urban runoff simulation and calculation, mainly through the curve number (CN) to represent the ability of rainfall to runoff under different land use types, soil texture and vegetation cover conditions. The SCS-CN hydrological model was used to simulate the surface runoff (Q) under different rainfall scenarios, and the specific calculation formula is as follows:

[0119]

[0120] I a = λ × S (4)

[0121] In the formula, Q is the surface runoff; P is the precipitation; S is the potential maximum retention rate of precipitation; CN is a dimensionless parameter representing the surface runoff capacity, with a value range of 0-100; I a is the initial infiltration of precipitation; λ is the soil infiltration coefficient, usually taking the empirical value 0.2.

[0122] The CN value, as a key parameter of the SCS-CN model, plays an important role in reasonably estimating the runoff capacity of urban areas. Considering factors such as regional land use, soil type, vegetation cover and wetness degree, and referring to the American engineering manual, Chinese empirical values and the actual situation of Hangzhou City, the CN values corresponding to each land use are determined. Based on the area proportion of each land use in the unit, the comprehensive CN value of the unit is calculated by the weighted method. The formula is as follows:

[0123]

[0124] In the formula, CN1, CN2, CN3, CN4, CN5, CN6 are the CN values corresponding to construction land, grassland, bare land, forest land, farmland and water area; a1, a2, a3, a4, a5, a6 are the area proportions of the six types of land use in the unit. The study selected 50a-1 rainfall to evaluate the rainwater regulation service of the eight central districts of Hangzhou City. According to the “Hangzhou City Rainstorm Intensity Calculation Standard (2020 Edition)”, the rainfall under the 50a-1 scenario is 122.6mm.

[0125] Among them, the green landscape pattern driving factors include three dimensions of scale, shape and structure. The scale characteristics include patch density (PD), largest patch index (LPI), mean patch size (AREA_MN); the shape characteristics include landscape shape index (LSI), mean shape index (SHAPE_MN), edge density (ED); the structure characteristics include division (DIVISION), connectivity (CONNECT), cohesion (COHESION); the above indexes are calculated by the landscape pattern analysis software fragstats. In addition, natural environmental factors (vegetation index, slope), social and economic factors (construction land proportion, population density) are selected as control variables.

[0126] Step S3: Nonlinear model construction

[0127] The structured dataset constructed based on 110 water management units is input into the model for training, which contains rainwater regulation services (response variable), green landscape pattern factors (explanation variable) and related factors such as natural environment and social economy (control variable). In the model training stage, the data set is divided into training set (n=77) and test set (n=33) in the ratio of 7:3 by stratified random sampling, of which the training set is used for model learning and the test set is used for independent performance evaluation. In the parameter optimization stage of the model, mainly based on Scikit-learn library, grid search algorithm (GridSearchCV) combined with 10-fold cross-validation is used to optimize the model hyperparameters, focusing on adjusting key parameters such as n_estimators and max_depth. In the model verification stage, the model performance is evaluated on the training set and test set respectively, and the determination coefficient (R 2 ) and mean absolute error (MAE) and other indicators are calculated to measure the model fitting ability and prediction accuracy.

[0128] Step S4: Key factor identification

[0129] Further use the variable importance ranking results output by the model to analyze the influence degree of different green landscape pattern factors on rainwater regulation services, please refer to Figure 3 . By extracting and ranking the importance scores of each feature variable, those factors that contribute more to the model prediction can be identified, thus revealing the main control factors of green landscape pattern affecting rainwater regulation services. Among them, the division (DIVISION), largest patch index (LPI), mean patch size (AREA_MN), cohesion (COHESION), edge density (ED), and mean shape index (SHAPE_MN) are the main control factors.

[0130] Step S5: Key threshold analysis

[0131] The PDPs (Partial Dependence Plots) algorithm is used for visual analysis of the above-mentioned main control factors. For details, please refer to Figure 4 The PDPs can quantitatively show the marginal effect of a single landscape pattern index on the model prediction results when other variables remain unchanged. By drawing the relationship curve between the different values of the characteristic variables and the average of the model prediction values, the threshold effect of the rain and flood regulation service is identified.

[0132] The analysis of the curve focuses on two points: 1) when the curve presents a clear inflection point or mutation, it indicates that there is a threshold effect, and the characteristic value corresponding to the inflection point is the critical threshold; 2) the change of the curve slope reflects the change of the sensitivity of the characteristics. The key threshold of the fragmentation (DIVISION) is 0.64, the key threshold of the largest patch index (LPI) is 65.83, the key threshold of the mean patch size (AREA_MN) is 11.81, the key threshold of the cohesion (COHESION) is 97.36, the key threshold of the edge density (ED) is 166.98, and the key threshold of the mean shape index (SHAPE_MN) is 1.29.

[0133] Although the embodiments of the present application have been shown and described above, it can be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for evaluating the non-linear effect of green space landscape pattern on stormwater regulation service, characterized in that, The method comprises the following steps: Step 1: constructing a basic database The GIS software is used for coordinate unification, precision correction and projection conversion of the basic data, so as to ensure the high consistency of different source data in space; Step 2: processing data variables Step 21, considering the hydrogeological conditions, water system distribution, land use and urban administrative boundary factors, the water management unit is delimited; Step 22, based on the SCS-CN model, the rainwater regulation service of each water management unit is quantified; Step 23, referring to existing research, the green space patches are extracted from each water management unit, and a green space landscape pattern index system is constructed from three dimensions of scale, shape and structure; Step 3: constructing a nonlinear model Taking the green space landscape pattern factor as the core explanatory variable and the rainwater regulation service index as the response variable, while introducing the natural environment factor and the social and economic factor as the control variable, the random forest model is trained to construct a rainwater regulation service prediction model; Step 4: identifying key factors According to the importance of the green space landscape pattern factor output by the rainwater regulation service prediction model, the variables are sorted, and the top-ranked factors are regarded as the key factors that play a key role in the rainwater regulation service; Step 5: analyzing key thresholds By using the partial dependence plot algorithm, the non-linear relationship between the key factor and the rainwater regulation service is visualized; by observing the positive and negative influence trend and the change rate of the rainwater regulation service index in different value intervals of the key factor, the potential key threshold is identified.

2. The method of claim 1, wherein the method is used to evaluate the non-linear effects of green space landscape pattern on stormwater management services. In step S1, the basic data includes DEM elevation data, soil data, normalized vegetation index data, administrative boundary data, population density data, drainage partition data and land use data.

3. The method of claim 2, wherein the method is used to evaluate the non-linear effects of green space landscape pattern on stormwater management services. In step S1, the land use data is based on the "Geospatial Data Cloud" platform to select the Landsat8 remote sensing image data map of Hangzhou in May 2023, and ENVI5.1 software is used for atmospheric correction and projection definition; referring to the LUCC classification system, the maximum likelihood classification method and the combination of field research and visual interpretation are used to divide the land in the study area into 6 types, namely construction land, forest land, grassland, water area, cultivated land and unused land.

4. The method of claim 1, wherein, In step S2, the rainwater regulation service space quantification is mainly based on the SCS-CN model, and the specific calculation formula is as follows: C = Δv x (0.001 x P h x A) -1 x 100% (1) Where C represents the surface runoff regulation rate, i.e., the stormwater regulation service evaluation result; v represents the surface runoff storage capacity (mm); P h A represents the rainfall (mm); A represents the area of ​​the study unit (m²). 2 );Q i Q represents the surface runoff (mm) of a 100% impermeable surface; n A represents the actual surface runoff (mm) of the underlying surface; i pixel area (m 2 The higher the surface runoff regulation rate C, the higher the stormwater regulation service provided by the green space.

5. The method for assessing the non-linear effects of greenfield landscape pattern on stormwater regulation services according to claim 4, wherein, The SCS-CN hydrological model is used to simulate the surface runoff Q under different rainfall scenarios, and the specific calculation formula is as follows: I a = λ x S (5) Where Q is the surface runoff, P is the precipitation, S is the maximum potential retention rate of precipitation, CN is a dimensionless parameter representing the surface runoff capacity, and I a is the initial infiltration of precipitation, and λ is the soil infiltration coefficient, usually taking an empirical value of 0.

2.

6. The method for assessing the non-linear effect of green space landscape pattern on stormwater regulation service according to claim 5, wherein, Based on the area proportion of each land use in the water management unit, the unit comprehensive CN value is calculated by the weighted method, and the specific formula is as follows: Wherein, CN1, CN2, CN3, CN4, CN5, CN6 are the CN values of construction land, grassland, bare land, forest land, cultivated land and water area; a1, a2, a3, a4, a5, a6 are the area proportions of the six types of land use in the unit.

7. The method of claim 1, wherein, In step S3, the structured data set constructed based on multiple water management units is input into the model for training, and the structured data set contains response variables, explanatory variables and control variables; In the model training stage, the data set is divided into training set and test set by stratified random sampling in the proportion of 7:3, wherein the training set is used for model learning, and the test set is used for independent performance evaluation; In the parameter optimization stage of the model, based on the Scikit-learn library, the grid search algorithm GridSearchCV is combined with 10-fold cross-validation to optimize the model hyperparameters, and the key parameters n_estimators and max_depth are adjusted; In the model validation phase, the model performance is evaluated on the training set and test set respectively, and the determination coefficient R is calculated 2 With the mean absolute error MAE index, to measure the model fitting ability and prediction accuracy.

8. The method of claim 1, wherein, In step S4, by extracting and sorting the importance scores of each feature variable, those factors that contribute more to the model prediction are identified, so as to reveal the key factors of green landscape pattern affecting rainwater regulation service; wherein the fragmentation DIVISION, the maximum patch area ratio LPI, the average patch size AREA_MN, the cohesion COHESION, the edge density ED and the average shape index SHAPE_MN are the key factors.

9. The method of claim 1, wherein, In step S5, the PDPs algorithm is used for visual analysis of the key factors; by drawing the relationship curve between the different values of the feature variables and the average of the model prediction values, the threshold effect of rainwater regulation service is identified: 1) when the curve presents an obvious inflection point or mutation, it indicates that there is a threshold effect, and the feature value corresponding to the inflection point is the critical threshold; 2) the change of the curve slope reflects the sensitivity change of the feature influence.

10. An evaluation system for non-linear effects of greenfield landscape patterns on stormwater regulation services, characterized in that, Comprise: A data acquisition and processing module for obtaining a structured data set in a target study area; wherein the structured data set includes rainwater regulation service index, green landscape pattern index, natural environment data and social economic data; A model construction and training module for constructing a nonlinear model, taking the rainwater regulation service index as the response variable, the green landscape pattern index as the explanatory variable, and the natural environment data and social economic data as the control variable; training the random forest model to construct a rainwater regulation service prediction model, and visualizing the rainwater regulation service prediction model according to the PDPs algorithm; A result analysis and interpretation module for sorting and threshold identification of the green landscape pattern factor importance of the target study area according to the results of the influence of the green landscape pattern on the rainwater regulation service, obtaining the key factors and key thresholds of the green landscape pattern affecting the rainwater regulation service, and the analysis results can be directly used for scientific guidance of the optimization construction and practice of urban green landscape pattern.

Images