A shap-based urban rainstorm waterlogging influence factor quantification analysis method

Abstract

The application relates to the technical field of urban renewal planning research, and discloses a city stormwaterlogging influence factor quantitative analysis method based on SHAP. First, DEM data, high-resolution land use data and waterlogging point spatial distribution data in a research area are collected; second, the collected data are preprocessed through format conversion, cutting, projection and the like; third, terrain factors are extracted and landscape pattern indexes are calculated based on the collected data; fourth, all the obtained influence factor data are standardized, and a simple random sampling method is used to divide the data into training samples and verification samples; the application combines feature importance sorting, tolerance and a variance inflation factor, retains the most important influence factors, solves information cross-redundancy between the influence factors, and makes the analysis result of city waterlogging and the influence factors more reasonable and easier to explain.

Description

Technical Field

[0001] This invention relates to the field of urban renewal planning research technology, specifically a quantitative analysis method for urban stormwater flooding influencing factors based on SHAP. Background Technology

[0002] Cities are a symbol of human civilization. According to data released by the United Nations in its "2018 World Urbanization Trends" report, the global average urbanization rate was 55% as of 2018, and is projected to reach 68% by 2050. Developed countries in Europe and America have already exceeded 80%. my country's urbanization rate has reached 59.58%, and is projected to exceed 70% by 2035, significantly higher than the world average. At the same time, with the expansion of urbanization and the impact of extreme weather, urban flooding has become a major factor affecting residents' lives and social production. According to statistics from the Ministry of Housing and Urban-Rural Development, 213 out of 351 cities in my country have experienced urban flooding disasters, 39% of cities experience flooding more than three times a year, and 57 cities have experienced flooding lasting more than 12 hours, causing significant social and economic losses. First-tier cities such as Beijing, Shanghai, Guangzhou, and Shenzhen are particularly severely affected.

[0003] Therefore, exploring the causes and mechanisms of urban flooding caused by rainstorms and its coping strategies has become a research hotspot and is an important foundational study for preventing, controlling, and reducing disaster risks. Currently, the methods commonly used domestically and internationally to analyze the influencing factors of urban flooding include Pearson correlation analysis, partial correlation analysis, multiple stepwise regression, and geographically weighted regression. However, the relationship between urban flooding and its influencing factors is often non-linear. When the relationship between independent and dependent variables becomes more complex, these methods are often difficult to apply, and they also struggle to quantify the relative contribution of each influencing factor to urban flooding. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a quantitative analysis method for urban stormwater flooding influencing factors based on SHAP. This method can effectively analyze the correlation between urban flooding and various influencing factors, and quantify the relative contribution of each factor to urban flooding.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A quantitative analysis method for urban stormwater flooding influencing factors based on SHAP includes the following steps:

[0009] (A) Collect DEM data, high-resolution land use data and spatial distribution data of flood points within the study area;

[0010] (B) Perform preprocessing on the collected data, such as format conversion, cropping, and projection;

[0011] (C) Extract terrain factors based on DEM data;

[0012] (D) Calculate the landscape pattern index based on land use data;

[0013] (E) Standardize the data and use a simple random sampling method to divide it into training samples and validation samples;

[0014] (F) A model of the relationship between urban flooding and various influencing factors was constructed based on the XGboost algorithm; the extracted topography, land use and landscape pattern were used as independent variables of the model, and the spatial distribution of flooding points was used as the dependent variable of the model to train the XGboost model.

[0015] (G) Based on SHAP, the trained model is interpreted, the importance of the influencing factors is ranked, and the optimal combination of influencing factors is determined by tolerance and variance inflation factor.

[0016] (H) Based on the determined optimal combination, the relationship model between urban flooding and each influencing factor is reconstructed again based on the XGboost model, and the relationship between urban flooding and each influencing factor is quantitatively analyzed by combining the SHAP method.

[0017] Preferably, in step (A), the study area dataset requires the following steps:

[0018] 1) The acquisition time of DEM data, land use data, and spatial distribution data of waterlogging points should be kept as consistent or close as possible;

[0019] 2) The spatial resolution of DEM data and land use data should be controlled within 30m×30m as much as possible.

[0020] 3) The DEM data used should be adjusted to remove the deviations caused by tree height and building height in order to ensure higher quality terrain information.

[0021] Preferably, in step (B), the preprocessing of the collected data, including format conversion, cropping, and projection, mainly includes:

[0022] 1) Unify DEM data and land use data into .tif format;

[0023] 2) The DEM data and land use data were cropped using vector data of the study area;

[0024] 3) Unify the DEM data and land use data into the same coordinate system.

[0025] Preferably, in step (C), five topographic factors are selected: elevation, slope, surface relief, surface roughness, and runoff accumulation.

[0026] Extracting terrain factors based on DEM data mainly includes the following steps:

[0027] 1) Extract altitude: The pixel value of each raster in the DEM data is the altitude.

[0028] 2) Extract the slope, the calculation formula is: tanα = elevation difference / horizontal distance; where α represents the slope;

[0029] 3) Extract the surface relief. Within a specific area, the difference between the elevation of the highest point and the elevation of the lowest point is the surface relief.

[0030] 4) Extract surface roughness. The method for calculating surface roughness is: the ratio of the Earth's surface area to its projected area within a specific region.

[0031] 5) Extract the cumulative flow. The cumulative flow is calculated according to the natural law of water flowing from high to low, and then the amount of water flowing through each spatial unit is calculated based on the numerical matrix of water flow direction in the study area. The numerical matrix of water flow direction uses the D8 algorithm.

[0032] Preferably, in step (D), seven landscape pattern indices are selected: patch number index, patch shape index, patch aggregation index, patch fragmentation index, largest patch index, spread index, and Shannon diversity index.

[0033] Preferably, in step (E), the data standardization formula used is:

[0034]

[0035] In the formula, x represents the original value, min represents the maximum value in the dataset, max represents the minimum value in the dataset, and X represents the transformed value.

[0036] Preferably, in step (G), the method for determining the optimal combination of influence factors is as follows:

[0037] 1) Based on the SHAP values, sort the impact factors from highest to lowest importance;

[0038] 2) Starting with the two most important features, take them as a feature subset and calculate the tolerance and variance inflation factor. When the tolerance is less than 0.1 or the variance inflation factor is greater than 10, it indicates that there is information redundancy between the factors. At this time, remove the least important factor from the subset.

[0039] 3) Based on the subset from the previous step, use an exhaustive approach to add subsequent factors to the subset one by one according to their importance, and calculate the tolerance and variance inflation factor. After the iteration is complete, the impact factors retained in the subset are the optimal factor combination.

[0040] (III) Beneficial Effects

[0041] Compared with existing technologies, this invention provides a quantitative analysis method for urban stormwater flooding influencing factors based on SHAP, which has the following beneficial effects:

[0042] Compared with traditional regression analysis methods, this invention can quantify the nonlinear relationship between urban flooding and various influencing factors, as well as quantify the relative contribution of each influencing factor to urban flooding. At the same time, considering that information redundancy among influencing factors can affect the accuracy of the analysis results, this invention combines feature importance ranking, tolerance, and variance inflation factor. While retaining the most important influencing factors, it solves the problem of information cross-redundancy among influencing factors, making the analysis results of urban flooding and various influencing factors more reasonable and easier to interpret. Attached Figure Description

[0043] Figure 1 This is a flowchart of the present invention;

[0044] Figure 2 This is a ranking chart of impact factors based on SHAP.

[0045] Figure 3 A diagram for determining the optimal combination of influence factors;

[0046] Figure 4 To quantify the relative contributions of each influencing factor and their relationship with urban flooding based on SHAP; Detailed Implementation

[0047] The following description fully illustrates specific embodiments of the invention to enable those skilled in the art to practice them. The example of this invention uses Shenzhen City, Guangdong Province as the study area, with a land area located at 22°27–22°52′N, 113°46′–114°37′E, and a total area of ​​approximately 2020 km².

[0048] Reference Figure 1-4 A quantitative analysis method for urban stormwater flooding influencing factors based on SHAP is presented, with the following specific steps:

[0049] The first step is to prepare the dataset. This includes the following aspects:

[0050] 1) Download the FABDEM (Forest And Buildings removed Copernicus DEM) dataset, which has a spatial resolution of 30m×30m. This dataset removes the bias caused by trees and buildings from COPDEM30 (Copernicus GLO-30).

[0051] 2) Download land use data from 2015, with a spatial resolution of 2m×2m. Land use includes impervious surfaces, green spaces, bare land, agricultural land, and water bodies.

[0052] 3) A total of 394 data points were collected, including data points in areas prone to flooding and data points not prone to flooding.

[0053] The second step involves preprocessing the collected data, including format conversion, cropping, and projection. The DEM and land use data formats are unified to .tif. Vector data from the study area is used to crop the DEM and land use data, and the DEM and land use data are unified to the same coordinate system, with the projection coordinate system being WGS_1984_UTM_Zone_50N.

[0054] The third step, based on the DEM data, is to extract five factors: elevation, slope, topographic relief, runoff accumulation, and surface roughness. The steps are as follows:

[0055] 1) Extracting altitude. The pixel value of each raster in the DEM data represents the altitude;

[0056] 2) Extract the slope. The calculation formula is: tanα = elevation difference / horizontal distance. Where α represents the slope.

[0057] 3) Extracting surface relief. Within a specific area, the difference between the elevation of the highest point and the elevation of the lowest point is the surface relief.

[0058] 4) Extracting surface roughness. The method for calculating surface roughness is: the ratio of the Earth's surface area to its projected area within a specific region.

[0059] 5) Extracting the cumulative runoff volume. The cumulative runoff volume is calculated according to the natural law of water flowing from high to low altitudes, and then the volume of water flowing through each spatial unit is calculated based on the numerical matrix of water flow direction in the study area. The numerical matrix of water flow direction uses the D8 algorithm.

[0060] Step 4: Based on high-resolution land use data, landscape pattern indices were calculated. The landscape pattern indices include two levels: type level and landscape level. The type level is used to calculate the landscape index for different land use types, while the landscape level is used to calculate the landscape index for the overall land use type. The seven selected landscape pattern indices are shown in Table 1.

[0061] Table 1 Landscape Pattern Index

[0062]

[0063] The fifth step is to perform Max-Min standardization on the data and then use a simple random sampling method to divide it into 75% training samples and 25% validation samples.

[0064] The specific steps of standardizing the data and dividing it into training and validation samples using simple random sampling include:

[0065] 1) Standardization is performed using Min-Max, and the formula is:

[0066]

[0067] In the formula, x represents the original value, min represents the maximum value in the dataset, max represents the minimum value in the dataset, and X represents the transformed value.

[0068] 2) The samples are divided into a training set and a validation set using a simple random sampling method. The training set is used for model training, and the validation set is used for model validation.

[0069] Step 6: Construct a model of the relationship between urban flooding and various influencing factors based on the XGboost (eXtreme Gradient Boosting) algorithm. The extracted topography, land use, and landscape pattern are used as independent variables in the model, and the spatial distribution of flooded areas is used as the dependent variable. The XGboost model is then trained. Here, 1 and 0 represent flooded areas and non-flooded areas, respectively.

[0070] Step 7: Based on SHAP, interpret the trained model, rank the importance of the influencing factors, and determine the optimal combination of influencing factors using tolerance and variance inflation factor. When the tolerance is less than 0.1 or the variance inflation factor is greater than 10, it indicates that there is information redundancy among the factors, and the least important factor in the set is removed.

[0071] Determining the optimal combination of influence factors specifically includes:

[0072] 1) Based on the SHAP values, sort the impact factors from highest to lowest importance;

[0073] 2) Starting with the two most important features, take them as a feature subset and calculate the tolerance and variance inflation factor. When the tolerance is less than 0.1 or the variance inflation factor is greater than 10, it indicates that there is information redundancy between the factors. At this time, remove the least important factor from the subset.

[0074] 3) Based on the subset from the previous step, use an exhaustive approach to add subsequent factors to the subset one by one according to their importance, and calculate the tolerance and variance inflation factor. After the iteration is complete, the impact factors retained in the subset are the optimal factor combination.

[0075] Step 8: Based on the determined optimal combination, reconstruct the relationship model between urban flooding and various influencing factors based on the XGboost model, and combine the SHAP method to quantitatively analyze the relationship between urban flooding and various influencing factors.

Claims

1. A quantitative analysis method for urban stormwater flooding influencing factors based on SHAP, characterized in that: Includes the following steps: (A) Collect DEM data, high-resolution land use data, and spatial distribution data of flood points within the study area; (B) Perform format conversion, cropping, and projection preprocessing on the collected data; (C) Extract terrain factors based on DEM data; (D) Calculate the landscape pattern index based on land use data; (E) Standardize the data and use a simple random sampling method to divide it into training samples and validation samples; (F) A model of the relationship between urban flooding and various influencing factors was constructed based on the XGboost algorithm; the extracted topography, land use and landscape pattern were used as independent variables of the model, and the spatial distribution of flooding points was used as the dependent variable of the model to train the XGboost model. (G) Based on SHAP, the trained model is interpreted, the importance of the impact factors is generated and ranked, and the optimal combination of impact factors is determined by tolerance and variance inflation factor. (H) Based on the determined optimal combination, the relationship model between urban flooding and each influencing factor is reconstructed again based on the XGboost model, and the relationship between urban flooding and each influencing factor is quantitatively analyzed by combining the SHAP method. In step (G), the method for determining the optimal combination of influence factors is as follows: 1) Based on the SHAP values, sort the impact factors from highest to lowest importance; 2) Starting with the two most important features, take them as a feature subset and calculate the tolerance and variance inflation factor. When the tolerance is less than 0.1 or the variance inflation factor is greater than 10, it indicates that there is information redundancy between the factors. At this time, remove the least important factor from the subset. 3) Based on the subset from the previous step, use an exhaustive method to sort the factors according to their importance, add subsequent factors to the subset one by one, and calculate the tolerance and variance inflation factor; when the iteration is complete, the factors retained in the subset are the optimal factor combination.

2. The method for quantitative analysis of urban stormwater flooding influencing factors based on SHAP according to claim 1, characterized in that: In step (A), the collection of the dataset covering the study area requires the following steps: 1) The acquisition time of DEM data, land use data, and spatial distribution data of waterlogging points should be kept as consistent or close as possible; 2) The spatial resolution of DEM data and land use data should be controlled within 30m×30m as much as possible; 3) The DEM data used should be adjusted to remove the deviations caused by tree height and building height in order to ensure higher quality terrain information.

3. The method for quantitative analysis of urban stormwater flooding influencing factors based on SHAP according to claim 1, characterized in that: In step (B), the format conversion, cropping, and projection preprocessing of the collected data mainly include: 1) Unify the DEM data and land use data into the .tif format; 2) The DEM data and land use data were cropped using vector data of the study area; 3) Unify the DEM data and land use data into the same coordinate system.

4. The method for quantitative analysis of urban stormwater flooding influencing factors based on SHAP according to claim 1, characterized in that: In step (C), five topographic factors were selected: elevation, slope, surface relief, surface roughness, and runoff accumulation. Extracting terrain factors based on DEM data mainly includes the following steps: 1) Extract altitude: The pixel value of each raster in the DEM data is the altitude. 2) Extract the slope, the calculation formula is: tanα = elevation difference / horizontal distance; where α represents the slope; 3) Extract the surface relief. Within a specific area, the difference between the elevation of the highest point and the elevation of the lowest point is the surface relief. 4) Extract surface roughness. The method for calculating surface roughness is: the ratio of the surface area to its projected area within a specific region. 5) Extract the cumulative flow. The cumulative flow is calculated according to the natural law of water flowing from high to low. Then, the amount of water flowing through each spatial unit is calculated based on the digital matrix of water flow direction in the study area. The digital matrix of water flow direction adopts the D8 algorithm.

5. The method for quantitative analysis of urban stormwater flooding influencing factors based on SHAP according to claim 1, characterized in that: In step (D), seven landscape pattern indices were selected: patch number index, patch shape index, patch aggregation index, patch fragmentation index, largest patch index, spread index, and Shannon diversity index.

6. The method for quantitative analysis of urban stormwater flooding influencing factors based on SHAP according to claim 1, characterized in that: In step (E), the data standardization formula used is: ； In the formula, x The value represents the original value, min represents the maximum value in the dataset, max represents the minimum value in the dataset, and X represents the transformed value.

Images