Urban-oriented composite water pollution simulation and evaluation method and system

Abstract

The invention provides a simulation and evaluation method and system for urban composite water pollution, and the method comprises the steps: taking a rainstorm management model SWMM as a basic platform, dividing a sub-basin and a spatial patch, and building a spatial association relationship among the spatial patch, the sub-basin, a sewer network and an inspection well; a water power module, a runoff pollution module and a combined system overflow CSO pollution module are constructed respectively, coupling among the modules is established to obtain a coupling model, output of the water power module drives load calculation of the runoff pollution module and the CSO pollution module, pollution loads of the runoff pollution module and the CSO pollution module are summarized, and combined pollution total output is obtained; acquiring climate prediction data and urbanization prediction data of a future scene, performing downscaling and standardization processing on the climate prediction data and the urbanization prediction data respectively, generating input data matched with the scale of the coupling model, inputting the input data into the coupling model, and performing simulation to obtain pollution results under different future scenes; according to the method, the simulation precision, prediction reliability, prevention and control targeting and planning adaptability of urban combined pollution are improved.

Description

Technical Field

[0001] This invention relates to the field of urban water environment simulation and pollution control technology, and in particular to a method and system for simulating and assessing complex urban water pollution. Background Technology

[0002] Over the past few decades, climate change and human activities have exacerbated water pollution, leading to water quality deterioration and posing a serious threat to human society. Notably, dense human activities in urban areas have significantly exacerbated water pollution, reduced water quality, intensified the shortage of clean water resources, threatened human well-being, exacerbated regional inequality, and ultimately undermined progress toward achieving the Sustainable Development Goals.

[0003] While traditional stormwater management models are widely used in urban hydrological simulation, they suffer from systemic deficiencies when dealing with the combined pollution problems of surface runoff and combined sewer overflows. The hydrodynamic module and the pollution module often operate independently or are simply correlated, failing to achieve deep coupling between the surface runoff generation, runoff pollution, and CSO pollution modules. This makes it impossible to accurately reflect the collaborative migration process and mutual influence of the two types of pollution, and it is also impossible to quantify the contribution of climate and urbanization factors to pollution. It is difficult to identify the main pollution controlling factors, resulting in a lack of targeting in the formulation of prevention and control measures and reducing the accuracy of simulation assessments. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for simulating and assessing complex urban water pollution, aiming to solve the problems of low coupling between the hydrodynamic module and the pollution module in the prior art and the inability to quantify the contribution of climate and urbanization factors to pollution.

[0005] The technical solution of this invention is implemented as follows: In a first aspect, this invention provides a method for simulating and assessing complex urban water pollution, comprising the following steps: S1, based on the SWMM storm management model, divides the land into sub-basins and spatial patches based on surface elevation data, and establishes the spatial relationship between spatial patches, sub-basins, sewer networks and inspection wells. S2, based on spatial correlation, a hydrodynamic module, a runoff pollution module, and a combined sewer overflow (CSO) pollution module are constructed respectively, and a coupling model is established between the modules. The output of the hydrodynamic module drives the load calculation of the runoff pollution module and the CSO pollution module, and the pollution loads of the two are summarized to obtain the total output of composite pollution. S3: Obtain climate and urbanization prediction data for future scenarios, downscale and standardize them respectively, generate input data that matches the scale of the coupled model, and input it into the coupled model to simulate pollution results under different future scenarios. S4, based on the urbanization attributes and pollution results of spatial patches, performs attribute classification and spatial correlation visualization to identify high-risk pollution areas; and based on the input variables and output results of the coupled model, uses a random forest model combined with SHAP interpretability analysis to quantify the contribution of climate factors and urbanization factors to pollution results. S5, based on spatial patches, conducts patch-based deployment of low-impact development (LID) measures, and uses a coupled model to evaluate the pollution reduction effects of LID measures under normal and extreme events.

[0006] Based on the above technical solutions, preferably, step S1 includes the following sub-steps: S11: Obtain the original surface elevation data at a preset resolution, and perform terrain correction, noise reduction and depression filling to obtain standard surface elevation data. S12, based on the deterministic eight-neighborhood D8 algorithm, performs confluence direction analysis on standard surface elevation data to generate a global confluence direction matrix; S13, based on the confluence direction matrix and the location of manholes in the sewer network, aggregates all runoff into the same manhole to form a sub-basin; S14, preset the minimum catchment area threshold, divide each sub-basin into multiple spatial patches, and establish the spatial relationship between spatial patches, sub-basins, sewer network and inspection wells.

[0007] Based on the above technical solutions, preferably, the construction of the hydrodynamic module in step S2 includes the following sub-steps: S211, Obtain the input data of the coupled model, including historical hydrological observation data, underlying surface attribute data and sewer network parameters. The historical hydrological observation data includes rainfall data and flow data. The underlying surface attribute data includes the impermeability and soil type of each spatial patch. The sewer network parameters include the structural parameters of pipes and manholes, as well as the geometric parameters of the overflow weir. S212 uses spatial patches as the basic computational unit and employs a stormwater management model to simulate the processes of rainfall interception, soil infiltration, surface runoff, and pipe confluence. Specifically, rainfall interception and soil infiltration both utilize the Horton infiltration model to calculate the rainfall interception amount from vegetation and building roofs. Surface runoff is calculated using the kinematic wave equation to determine the surface runoff rate and volume before confluence with the outlet of the corresponding sub-basin. Pipe confluence is simulated using the Saint-Venant equations to model one-dimensional unsteady flow within sewer pipes, calculating the flow rate, water level, and overflow at the overflow weir. S213 outputs hydrodynamic simulation results, including surface runoff rate, sub-basin outlet flow, pipeline flow and water level, and overflow weir flow for each spatial patch.

[0008] Based on the above technical solutions, preferably, the construction of the runoff pollution module in step S2 includes the following sub-steps: S221 uses spatial patches as the calculation unit to simulate the accumulation, flushing and emission processes of runoff pollution; S222, In the simulation of pollutant accumulation, an exponential accumulation model is used to calculate the cumulative load of pollutants on spatial patches during the drought period. The expression for the exponential accumulation model is: ; In the formula, B i pollutants i Actual accumulated load on spatial patches b 1 i,j pollutants i In land cover type j The maximum accumulation amount on b2; j Land cover type j The accumulation rate is d; d is the accumulation time. S223, In the simulation of pollutant scour, based on the surface runoff rate data output by the hydrodynamic module, a scour model is used to calculate the scour load of pollutants during the rainfall period. The expression of the scour model is as follows: ; In the formula, It is a pollutant i In land cover type j The scouring load on the surface; It is a type of land cover j The scouring coefficient on; It is a type of land cover j The scouring index on the surface; q t It is a point in time. t Runoff flow; S224, in the pollutant emission simulation, the emission load is obtained by subtracting the adsorption amount of the underlying surface from the scouring load, which serves as the final output of the runoff pollution module.

[0009] Based on the above technical solutions, preferably, the construction of the combined overflow CSO pollution module described in S2 includes the following sub-steps: S231, Based on population data of spatial patches, calculate the pollutant load and corresponding pollutant concentration carried by domestic sewage in each patch, expressed as: ; ; In the formula, L i,j Spatial plaques jThe load of pollutants i carried by domestic sewage. Pop j Spatial plaques j The number of people served l i pollutants i per capita load, C i,j Spatial plaques j Pollutants in domestic sewage i concentration, dw Per capita domestic sewage volume; S232 simulates the migration and mixing of pollutants from domestic sewage and runoff pollution modules within the pipeline to determine the concentration of each pollutant in the inspection well. S233, based on the overflow flow of the overflow weir output by the hydrodynamic module, identifies CSO events. When the overflow flow of the overflow weir is greater than 0, it is determined that a combined overflow CSO event has occurred. A preset dry time interval is used to distinguish continuous combined overflow CSO events. The pollutant load of CSO emissions is calculated by multiplying the pollutant concentration in the inspection well by the overflow flow. S234, the emission load output by the runoff pollution module and the pollutant load emitted by CSO are combined to obtain the total output of compound pollution.

[0010] Based on the above technical solutions, preferably, step S3 includes the following sub-steps: S31, acquire historical climate data and future climate prediction data from multiple climate models, the climate data including temperature data and precipitation data; S32, based on historical climate data, performs bias correction on future predicted climate data, downscales the corrected data to the target resolution using spatial interpolation, calculates statistical values ​​of multiple climate model datasets to remove outliers, and generates climate scenario input data for future periods. S33, acquire historical urbanization scenario data and future predicted urbanization scenario data, wherein the urbanization scenario data includes land cover rate and population density data; S34 employs a spatial downscaling method to reduce future land cover and population density prediction data to a target resolution that matches the spatial patch scale, in order to generate input data for future urbanization scenarios. S35, the processed climate input data and urbanization input data are combined according to different time dimensions and scenario dimensions to generate multiple sets of future scenario input datasets, and input into the coupled model to simulate the spatiotemporal changes of compound pollution under different scenarios; the time dimension includes near future time period and far future time period, and the scenario dimension includes low greenhouse emissions and high greenhouse emissions.

[0011] Based on the above technical solutions, preferably, step S4, which involves classifying attributes and visualizing spatial correlations based on the urbanization attributes and pollution results of spatial patches to identify high-risk pollution areas, includes the following sub-steps: S41, based on land cover data and population data of spatial patches, divide each spatial patch into land cover intensity level and population density level as urbanization attribute level; S42, based on the simulation results output by the coupled model, the pollutant load level and overflow frequency level are divided for each river segment as the pollution result level; S43. Associate and label the land cover intensity level and population density level of spatial patches with the corresponding river segments through the hydraulic connectivity between the spatial patches and the river segments; S44. Cross-statistically analyze the urbanization attribute level and pollution result level associated with the river section to generate a visualization result representing the spatial distribution of pollution risk under different urbanization intensity levels, so as to identify high-risk river sections.

[0012] Based on the above technical solutions, preferably, in step S4, the input variables and output results based on the coupled model are used, and a random forest model combined with SHAP interpretability analysis is employed to quantify the contribution of climate factors and urbanization factors to the pollution results, including the following sub-steps: S45. Select climate factors and urbanization factors from the input variables of the coupled model, and select pollution result indicators from the output results to construct a driving analysis dataset. The climate factors and urbanization factors include average temperature, annual rainfall, maximum rainfall, longest drought period, rainfall frequency, drought frequency, associated population and impervious area. The pollution result indicators include overflow frequency, emission load output by the runoff pollution module and pollutant load emitted by CSO. S46. A prediction model is constructed based on the random forest model. The training dataset is input into the random forest model for training. A mapping relationship from input variables to pollution result indicators is established, and the coefficient of determination is used to evaluate the model performance to obtain a well-trained prediction model. S47, using the K-means algorithm, a predetermined number of representative samples are selected from the training dataset as the benchmark for calculating the SHAP value. The SHAP value of each input variable with respect to the output data is calculated, expressed as: ; In the formula, The baseline value is M, where M is the number of input variables. For the first i The SHAP values ​​of the input variables; S48. Based on the ranking of SHAP values, identify the climate factors and urbanization factors that contribute the most to the pollution results, and analyze the nonlinear relationship between key variables and pollution results through dependency graph analysis to quantify the overall contribution weight of climate factors and urbanization factors to the pollution results.

[0013] Based on the above technical solutions, preferably, step S5 includes the following sub-steps: S51, based on the actual area and proportion of building roofs, roads and green spaces in each spatial patch, matches and deploys corresponding low impact development (LID) measures for each land use type. Among them, green roofs (GR) are deployed on building roofs, permeable pavement (PP) is deployed on roads, and bioretention ponds (BR) are deployed on green spaces. The area of ​​each LID measure accounts for 15% of the total area of ​​the land use type to which it is applied, and the service life of the LID measures is set at 25 years. S52, based on the coupled model, simulate the compound pollution process before the deployment of LID measures, and calculate the direct reduction of LID on runoff pollution and the indirect reduction of LID on CSO pollution by comparison. The formula for calculating the direct reduction of runoff pollution by LID is as follows: ; In the formula, Pollutants caused by runoff i Remove the load, For land cover type j LID practice on pollutants during flushing i The removal coefficient, Land cover type j The area; For at a certain point in time t 1. Land Cover Type j Pollutants in upstream surface runoff i The concentration; For at a certain point in time t 1. Land Cover Type j Upstream runoff volume; In terms of land cover type j LID practices for pollutant capture during runoff i The interception coefficient; The formula for calculating the indirect reduction of CSO pollution by LID is as follows: ; In the formula, pollutants i CSO removes load; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The concentration of pollutants; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The amount of sewage; S53. Based on direct and indirect reductions, calculate the performance indicators of LID measures under long-term hydrological simulation, including the annual average runoff reduction rate, the annual average runoff pollution reduction rate, and the annual average CSO pollution reduction rate. Then, divide the rainstorm events during the simulation period into multiple severity levels according to the total rainfall and intensity, and calculate the CSO frequency reduction rate and pollution load reduction rate under each level of event to evaluate the adaptability of LID under extreme climate conditions.

[0014] Secondly, the present invention also provides a simulation and assessment system for urban complex water pollution, implemented using a method for simulating and assessing urban complex water pollution. The system includes: The spatial infrastructure construction module is used to divide the watershed and spatial patch based on the SWMM storm management model as the basic platform, and to establish the spatial relationship between the spatial patch, the watershed, the sewer network and the inspection well. The model coupling construction module is used to construct a hydrodynamic module, a runoff pollution module, and a combined sewer overflow (CSO) pollution module based on spatial correlation, and to establish the coupling between the modules to obtain a coupled model. The output of the hydrodynamic module drives the load calculation of the runoff pollution module and the CSO pollution module, and the pollution loads of the two are summed to obtain the total output of composite pollution. The scenario simulation module is used to acquire climate prediction data and urbanization prediction data for future scenarios, and then downscale and standardize them to generate input data that matches the scale of the coupled model. This input data is then fed into the coupled model to simulate the pollution results under different future scenarios. The pollution risk analysis quantification module is used to perform attribute classification and spatial correlation visualization based on the urbanization attributes and pollution results of spatial patches in order to identify high-risk pollution areas; and based on the input variables and output results of the coupled model, it uses a random forest model combined with SHAP interpretability analysis to quantify the contribution of climate factors and urbanization factors to pollution results. The evaluation module is used for patchy deployment of low-impact development (LID) measures based on spatial patches, and uses a coupled model to evaluate the pollution reduction effectiveness of LID measures under normal and extreme events.

[0015] The method and system for simulating and assessing complex urban water pollution of the present invention have the following advantages over existing technologies: (1) By establishing a multi-level association system of spatial patches-sub-basins-pipeline, the deep coupling of multiple modules of dynamics-runoff pollution-CSO pollution is realized. The key parameters output by the hydrodynamic module, such as surface runoff rate, pipeline flow rate and overflow weir flow rate, directly drive the scouring load calculation of the runoff pollution module and the event identification and load accounting of the CSO pollution module, forming a dynamic data interaction mechanism. This ensures the collaborative simulation of the two types of pollutants in the entire process of generation-migration-emission. Finally, through the integration and summarization of pollution loads, a complete characterization of the composite pollution system is realized, improving the model's simulation accuracy of the collaborative migration process of multiple pollutants and their mutual influence in the actual urban water environment. (2) By combining multi-model data integration, bias correction and spatial downscaling of climate and urbanization data, a high-precision input matching the scale of the basic coupled model is generated, which improves the accuracy and matching degree of future scenario input data, thereby improving the accuracy and reliability of future pollution trend prediction. (3) By constructing a driving analysis dataset through multi-scenario data of the coupled model, and using random forest and SHAP interpretability analysis technology, the contribution of climate and urbanization factors was accurately quantified. It can not only identify the dominant pollution factors, but also reveal the nonlinear relationship between variables and pollution. Finally, it clarifies the specific contribution weight of each factor, thereby improving the scientificity and accuracy of environmental decision-making. (4) By establishing a quantitative model of direct and indirect reduction, the control mechanism of LID on compound pollution was revealed; the constructed multi-dimensional assessment system for regular and extreme events provides an LID adaptive assessment tool for responding to extreme rainfall events under climate change, and improves the accuracy of urban stormwater infrastructure planning. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the urban complex water pollution simulation and assessment method of the present invention; Figure 2 This defines the scope and location of the research area for the urban complex water pollution simulation and assessment method of this invention. Figure 2 (a) is a schematic diagram of surface elevation data. Figure 2 (b) is a schematic diagram of land cover distribution. Figure 2 (c) is a schematic diagram of the sewer system network. Figure 2 (d) is a diagram showing population density; Figure 3 This is a schematic diagram of the spatial pattern of rivers and canals with different CSO frequencies and pollutant loads under the baseline and future scenarios of the urban complex water pollution simulation and assessment method of the present invention. Figure 4 This is a diagram showing the SHAP values ​​of multiple driving factors for pollution-related indicators and the dependence of key climate and socioeconomic factors on the urban complex water pollution simulation and assessment method of the present invention. Figure 5 This is a schematic diagram of the LID performance under short-term extreme events for the urban complex water pollution simulation and assessment method of the present invention. Figure 6 This is a schematic diagram of the ID performance in the long-term hydrological process of the urban complex water pollution simulation and assessment method of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1-3 As shown, the present invention provides a method for simulating and assessing complex urban water pollution, characterized by comprising the following steps: S1 is based on the SWMM storm management model as its platform. It divides the land into sub-basins and spatial patches based on surface elevation data, and establishes spatial relationships between spatial patches, sub-basins, sewer networks and inspection wells.

[0020] Step S1 includes the following sub-steps: S11: Obtain the original surface elevation data at a preset resolution, and perform terrain correction, noise reduction, and depression filling to obtain standard surface elevation data.

[0021] It should be noted that, in a preferred embodiment of this invention, the original surface elevation data of the study area is obtained using surface elevation data with a resolution of 30m. This data is sourced from the ASTER open-source database. The original data is preprocessed in ArcGIS or a similar geographic information system platform to ensure the accuracy of the hydrological analysis. This preprocessing includes terrain correction to eliminate geometric distortions caused by sensor attitude and terrain undulations during satellite remote sensing imaging; denoising processing, using filtering algorithms to smooth local abnormal elevation points, eliminating data noise and avoiding the generation of false confluence paths; and depression filling processing to fill in small depressions on the surface, ensuring continuous surface runoff confluence and preventing analysis failure due to interrupted confluence paths.

[0022] S12, based on the deterministic eight-neighborhood D8 algorithm, performs confluence direction analysis on standard surface elevation data to generate a global confluence direction matrix; It should be noted that, based on the preprocessed standard surface elevation data, the deterministic eight-neighborhood D8 algorithm is used for confluence direction analysis. Specifically, the flow direction of each 30m×30m grid cell is calculated to the direction with the steepest slope among its eight neighboring cells. Through this process, a confluence direction matrix for the entire study area is generated. This matrix accurately records the water flow direction of each grid cell, laying the foundation for subsequent confluence accumulation and sub-basin division.

[0023] S13, based on the confluence direction matrix and the location of manholes in the sewer network, aggregates all runoff into the same manhole to form a sub-basin; It should be noted that the GIS vector data of the sewer network in the study area was imported, preferably provided by the municipal drainage management department, to ensure the accuracy of the pipe network; the location of the manholes was set as the confluence endpoint; based on the confluence direction matrix, the watershed division tool in the GIS platform was used to finally merge all surface runoff into the continuous raster area of ​​the same manhole, aggregating them into a sub-watershed; this method ensures that the divided sub-watersheds are completely matched with the actual drainage service area, realizing the spatial unification of natural hydrological processes and artificial drainage facilities.

[0024] S14, preset the minimum catchment area threshold, divide each sub-basin into multiple spatial patches, and establish the spatial relationship between spatial patches, sub-basins, sewer network and inspection wells.

[0025] It should be noted that, to meet the needs of subsequent urbanization attribute labeling and LID refined deployment, each sub-basin is further divided into smaller spatial units, namely spatial patches. A minimum runoff area threshold of 0.01 km² is set to balance computational efficiency and spatial accuracy. Continuous grids within each sub-basin that share the same secondary runoff path, i.e., those ultimately flowing into a pipe segment within the same sub-basin, are aggregated into a single spatial patch, resulting in several spatial patches, each with an area of ​​0.01-0.05 km². Through spatial overlay analysis, a four-level spatial relationship is established: spatial patch-sub-basin-sewage pipe-manhole. Each spatial patch clearly belongs to a sub-basin, and each sub-basin is connected to the sewer network through a unique manhole. Sewer pipes connect to manholes, forming a transport network. Ultimately, this relationship ensures that the entire path of runoff and pollutants generated from the smallest unit of the surface, from runoff to transport and discharge, is clearly defined and tracked without ambiguity.

[0026] This step establishes a unified spatial foundation, which solves the problem of chaotic spatial scale in existing technologies and provides a unified and high-precision spatial framework for the deep coupling of hydrodynamic and pollution modules and the precise deployment of LID measures.

[0027] S2. Based on spatial correlation, a hydrodynamic module, a runoff pollution module, and a combined sewer overflow (CSO) pollution module are constructed respectively, and a coupling model is established between the modules. The output of the hydrodynamic module drives the load calculation of the runoff pollution module and the CSO pollution module. The pollution loads of the two are summed to obtain the total output of composite pollution.

[0028] Based on the high-precision space established in S1, this step constructs hydrodynamic, runoff pollution, and CSO pollution modules, and achieves deep coupling among the three through data flow driving, overcoming the shortcomings of independent operation and loose correlation of modules in traditional models.

[0029] The construction of the hydrodynamic module described in step S2 includes the following sub-steps: S211, Obtain the input data of the coupled model, including historical hydrological observation data, underlying surface attribute data and sewer network parameters. The historical hydrological observation data includes rainfall data and flow data. The underlying surface attribute data includes the impermeability and soil type of each spatial patch. The sewer network parameters include the structural parameters of pipes and manholes, as well as the geometric parameters of the overflow weir. It should be noted that this module integrates multi-source, high-precision input data to ensure the physical basis of the simulation is realistic and reliable. In addition to the aforementioned spatial foundation data, it also includes historical hydrological observation data, underlying surface attribute data, and sewer network parameters. The historical hydrological observation data is monitoring data using high spatiotemporal resolution; for example, in one embodiment, rainfall records with a 5-minute step since 1996 and flow data measured at key overflow weirs were acquired, ensuring that the model can accurately respond to short-term heavy rainfall events. The underlying surface attribute data is assigned values ​​to each patch based on spatial patches; the impermeability of each patch is calculated using 10-meter resolution land cover data, and combined with soil type distribution data, accurate underlying surface characteristic parameters are provided for hydrological process simulation. The sewer network parameters use real pipe network data that has been digitized and verified by GIS. In one specific embodiment, the data includes the diameter, length, and slope of approximately 7,700 pipes, the elevation of 7,000 manholes, and the geometric parameters such as the height and width of over 100 overflow weirs.

[0030] S212 uses spatial patches as the basic computational unit and employs a stormwater management model to simulate the processes of rainfall interception, soil infiltration, surface runoff, and pipe confluence. Specifically, rainfall interception and soil infiltration both utilize the Horton infiltration model to calculate the rainfall interception amount from vegetation and building roofs. Surface runoff is calculated using the kinematic wave equation to determine the surface runoff rate and volume before confluence with the outlet of the corresponding sub-basin. Pipe confluence is simulated using the Saint-Venant equations to model one-dimensional unsteady flow within sewer pipes, calculating the flow rate, water level, and overflow at the overflow weir. It should be noted that by using spatial patches as the basic calculation unit, the complete physical process of "rainfall-runoff-confluence" is simulated, achieving a refined simulation from the ground surface to the pipe network; Rainfall interception and soil infiltration were simulated using the Horton infiltration model. This model can dynamically simulate the physical process of high infiltration rate in the early stage of rainfall, and the gradual decrease in infiltration rate over time until it stabilizes. It can accurately calculate the amount of rainfall interception by vegetation canopy and building roof and the amount of soil infiltration. Surface runoff is calculated using the kinematic wave equation, which, while ensuring computational efficiency, can accurately simulate the generation, collection, and evolution of surface runoff, outputting the surface runoff rate and runoff volume of each spatial patch at different time points; the runoff from all patches eventually flows into the outlet of their respective sub-basins.

[0031] The pipeline confluence is simulated using Saint-Venant's equations, which describe one-dimensional unsteady flow. These equations can accurately calculate the changes in flow rate, water level, and flow velocity in the pipeline over time, and thus calculate the overflow at each overflow weir. This overflow is the core criterion parameter for subsequently determining whether a CSO event has occurred.

[0032] S213 outputs hydrodynamic simulation results, including surface runoff rate, sub-basin outlet flow, pipeline flow and water level, and overflow weir flow for each spatial patch.

[0033] In this embodiment, high-resolution spatial patches are used as unified computing units, integrating the physical equations of the entire process from the ground surface to the pipeline network. This provides dynamic and accurate input for the subsequent runoff pollution module's scour load calculation and the CSO pollution module's event identification and load calculation, realizing deep coupling between modules.

[0034] The construction of the runoff pollution module described in step S2 includes the following sub-steps: It should be noted that the runoff pollution module simulates the entire process of "accumulation-flushing-emission" of pollutants on spatial patches and is deeply coupled with the hydrodynamic module to achieve accurate quantification of pollution load.

[0035] S221 uses spatial patches as the calculation unit to simulate the accumulation, flushing and emission processes of runoff pollution; It should be noted that a complete runoff pollution simulation framework was established using spatial patches as the basic calculation unit. This framework simulates the accumulation of pollutants during drought, the scouring during rainfall, and the final discharge process, realizing the whole chain tracking from pollutant generation to entry into water bodies.

[0036] S222, In the simulation of pollutant accumulation, an exponential accumulation model is used to calculate the cumulative load of pollutants on spatial patches during a dry period. The calculation expression of the exponential accumulation model is as follows: ; In the formula, It is a pollutant i In land cover type j Actual accumulated load (kg / ha) on the surface. It is a pollutant i In land cover type j The maximum accumulation amount (kg / ha) was obtained by calibration based on historical sampling data. It is a type of land cover j The accumulation rate on the surface characterizes how quickly pollutants accumulate. d It is the accumulated time, that is, the length of the drought period between two consecutive rainfall events; It should be noted that during droughts, the accumulation process of pollutants generated by human activities (such as transportation, industrial production, and daily life) on the impermeable surface of spatial patches is simulated using an exponential accumulation model to calculate the pollutant load. This model can accurately reflect the nonlinear variation characteristics of pollutant accumulation rate over time. This model covers a variety of typical pollutants, including suspended solids (SS), chemical oxygen demand (COD), total nitrogen (TN), and total phosphorus (TP), and can comprehensively characterize the characteristics of urban runoff pollution.

[0037] S223, In the simulation of pollutant scour, based on the surface runoff rate data output by the hydrodynamic module, a scour model is used to calculate the scour load of pollutants during the rainfall period. The expression of the scour model is as follows: ; In the formula, It is a pollutant i In land cover type j The scouring load on the surface (kg / ha); It is a type of land cover j The scouring coefficient (1 / mm) on the surface represents the scouring efficiency. It is a type of land cover j The scouring index characterizes the degree of nonlinear response of the scouring process to the runoff rate; q t It is a point in time. t The runoff flow rate originates from the output of the hydrodynamic module.

[0038] It should be noted that during rainfall, the simulation of accumulated pollutants being washed away by surface runoff and entering water bodies is directly driven by the surface runoff rate data output by the hydrodynamic module, reflecting the deep coupling between modules; a power function scour model is used to calculate the scour load, which can accurately reflect the nonlinear relationship between scour load and runoff rate.

[0039] S224, in the pollutant emission simulation, the emission load output by the runoff pollution module and the pollutant load emitted by CSO are combined to obtain the total output of composite pollution.

[0040] It should be noted that the simulation depicts the process by which flushed pollutants ultimately enter the aquatic environment. The flushed pollutants are carried by surface runoff into the sub-basin outlet, and then enter the sewer system or are directly discharged into the receiving water body. The pollutant discharge load is calculated based on the flushing load, taking into account the adsorption effect of the underlying surface. The formula is: Discharge Load = Flushing Load - Underlying Surface Adsorption Amount. The underlying surface adsorption amount is calculated based on the soil type of the spatial patch and a corresponding adsorption coefficient is set. The calculated discharge load is used as the final output of the runoff pollution module.

[0041] This embodiment uses the real-time runoff data output by the hydrodynamic module as the driving force to realize the dynamic coupling simulation of "hydrodynamics-runoff pollution". Through the exponential accumulation and power function scour model, the entire process of pollutants from generation, accumulation to migration is accurately quantified, providing a reliable runoff pollution load input for the simulation of complex pollution.

[0042] The construction of the combined overflow CSO pollution module described in S2 includes the following sub-steps: It should be noted that the CSO pollution module simulates the generation and migration of pollutants in domestic sewage, identifies overflow events and calculates pollution loads based on hydrodynamic conditions, and finally combines them with the runoff pollution module to form a complete composite pollution output.

[0043] S231, Based on population data of spatial patches, calculate the pollutant load and corresponding pollutant concentration carried by domestic sewage in each patch, expressed as: ; ; In the formula, L i,j Spatial plaques j The load (g / day) of pollutants i carried by domestic sewage. Pop j Spatial plaques j Number of people served l i pollutants i The average load per person per day (g / person·day). C i,j Spatial plaques j Pollutants in domestic sewage i Concentration (mg / L) dw The amount of domestic sewage per person per day (L / person·day).

[0044] It should be noted that the population served by each spatial patch was calculated using population density data with a resolution of 100 meters. Combined with the domestic sewage volume (120L / person·day) and pollutant load (SS: 6g / person·day, COD: 40g / person·day, TN: 8g / person·day, TP: 1.2g / person·day) specified in authoritative guidelines, the pollutant load and concentration of domestic sewage in each patch were calculated.

[0045] S232 simulates the migration and mixing of pollutants from domestic sewage and runoff pollution modules within the pipeline to determine the concentration of each pollutant in the inspection well. S233, based on the overflow flow of the overflow weir output by the hydrodynamic module, identifies CSO events. When the overflow flow of the overflow weir is greater than 0, it is determined that a combined overflow CSO event has occurred. A preset dry time interval is used to distinguish continuous combined overflow CSO events. The pollutant load of CSO emissions is calculated by multiplying the pollutant concentration in the inspection well by the overflow flow. Specifically, CSO events are identified based on the overflow flow rate of the overflow weir output by the hydrodynamic module. When the overflow flow rate of the overflow weir is greater than 0, a combined sewer overflow (CSO) event is determined to have occurred. A preset dry time interval is used to distinguish continuous combined sewer overflow (CSO) events. If there is no overflow for 4 consecutive hours, the current event is considered to have terminated, and the next overflow is considered a new event. The pollutant load of CSO emissions is calculated by multiplying the pollutant concentration in the inspection well by the overflow flow rate. The pollutant load of a single event and the cumulative pollutant emission load are obtained.

[0046] S234 combines the scouring load output from the runoff pollution module with the pollutant load emitted by the CSO to obtain the total output of composite pollution.

[0047] This embodiment starts with calculating the pollution source intensity based on detailed spatial population data, then moves to the mixing and migration of runoff pollution within pipelines, and finally to the precise overflow event identification and load calculation driven by the hydrodynamic module. Ultimately, by summarizing the runoff pollution data, it simulates the entire process of complex pollution in urban water bodies, providing effective data support for subsequent analysis.

[0048] S3 acquires climate and urbanization prediction data for future scenarios, downscales and standardizes them respectively, generates input data that matches the scale of the coupled model, and inputs it into the coupled model to simulate pollution results under different future scenarios.

[0049] This step addresses the model input mismatch problem caused by directly using low-resolution future forecast data in existing technologies. By systematically downscaling, bias correction, and integrating climate and urbanization data, a high-precision input matching the scale of the basic coupled model is generated, ensuring the reliability and accuracy of future scenario simulations.

[0050] Step S3 includes the following sub-steps: S31, acquire historical climate data and future climate prediction data from multiple climate models, the climate data including temperature data and precipitation data; It should be noted that, to ensure the comprehensiveness of climate predictions and quantify uncertainties, a multi-model ensemble strategy was adopted. Future climate data were derived from the public output of the Euro-CORDEX program (European Regional Climate Downscaling Experiment), integrating simulation results from three global circulation models (GCMs) and four regional climate models (RCMs). The three global circulation models (GCMs) include: HadGEM2-ES (UK Met Office), CNRM-CM5 (French Met Office), and MPI-ESM-MR (Max Planck Institute, Germany). The RCMs include: CCLM4-8-17 (German Met Office), RCA4 (Swedish Institute for Meteorology and Hydrology), HIRHAM5 (Danish Meteorological Institute), and WRF (National Center for Atmospheric Research, USA).

[0051] It should be noted that the original data is stored in NetCDF format and contains long-term time series of daily rainfall (mm) and daily average temperature (°C); however, its original spatial resolution is greater than 250 km, which cannot be directly used for fine simulation at the city or watershed scale, and must be downscaled.

[0052] S32 corrects the bias of future climate prediction data based on historical climate data, downscales the corrected data to the target resolution using spatial interpolation, calculates the statistical values ​​of multiple climate model datasets to remove outliers, and generates climate scenario input data for future periods.

[0053] It should be noted that bias correction and spatial downscaling are used to transform low-resolution climate data into high-resolution driving data suitable for this model; in order to eliminate systematic errors in the climate model, bias correction is performed using the linear scaling method.

[0054] Rainfall correction formula: ; In the formula: P corr To correct for the rainfall, P sim The original simulated rainfall for the climate model. P obs,avg This represents the historical average rainfall. P sim,avg This represents the average simulated rainfall over the historical period of the model. Temperature correction formula: ; In the formula: T corr For the corrected temperature, T sim The original simulated temperature for the climate model. T sim,avgThe average temperature simulated over the historical period of the model. T obs,avg This represents the average temperature observed over a historical period. The corrected model historical simulation data was compared with the observation data again to ensure that the deviation rate was reduced to less than 5%, effectively eliminating systematic errors.

[0055] Spatial downscaling employs ordinary kriging interpolation. This method constructs a semi-variogram and uses corrected data from known grids to estimate climate variables for unknown grids. Compared to other interpolation methods, it better considers the spatial continuity of climate data (such as spatial gradient changes in rainfall) and avoids abrupt outliers after interpolation. The corrected data is downscaled to a 1-kilometer resolution using ordinary kriging interpolation. Bias-corrected NetCDF data is imported into ArcGIS software. Daily rainfall and daily average temperature are used as interpolation variables. The semi-variogram is set to a spherical model to adapt to the spatial correlation characteristics of the climate data, with a search radius of 20 km. Daily-scale climate variables are calculated for each 1-kilometer grid, generating 1-kilometer × 1-kilometer regular grid climate data covering the study area. The 1-kilometer grid data is spatially overlaid with the sub-basin boundaries of the coupled model to ensure that average climate data from at least one grid is available for each sub-basin, achieving a complete scale match between climate data and sub-basin simulation.

[0056] To suppress extreme outliers in a single model and quantify climate uncertainty, the downscaling results of seven models were ensembled and statistically analyzed. The median was used as the main input for future climate scenarios because it is insensitive to outliers. At the same time, 95% confidence intervals were calculated to characterize the uncertainty between the predictions of different models. Finally, a future climate dataset with 1km resolution, daily scale, median, and 95% confidence intervals was formed.

[0057] Specific steps: The daily scale data at 1km resolution of the 7 models are summarized by grid-time-variable dimension to form an Excel data table. For example, the 7 model output values ​​corresponding to "grid ID=1, date=2030-01-01, variable=rainfall" are denoted as P_1-P_7 respectively. For each grid-time-variable combination of 7 model values, the median was calculated using Excel's MEDIAN function. The median can effectively suppress extreme outliers in individual models and ensure the stability of the input data. Calculating the 95% confidence interval involves the following steps: Step 1: Use Excel's STDEV.S function to calculate the sample standard deviation (S) of the 7 model values ​​to reflect the degree of data dispersion; Step 2: Use the T.INV.2T function to calculate the critical value of the t-distribution at the 95% confidence level; Step 3: Calculate the standard error SE = Sn; where n is the number of climate models. Step 4: Calculate the upper and lower limits of the 95% confidence interval; lower limit = median - 2.447 × SE, upper limit = median + 2.447 × SE; Results output: The final result is a future climate dataset with 1km resolution, daily scale, median, and 95% confidence interval. The median serves as the main climate input for the basic coupled model, and the 95% confidence interval is used for subsequent uncertainty analysis.

[0058] S33, acquire historical urbanization scenario data and future predicted urbanization scenario data, wherein the urbanization scenario data includes land cover rate and population density data; It should be noted that high-resolution open-source data were used directly. Historical land cover data came from the Zanaga database (10-meter resolution), and population density data came from the Liu et al. dataset (100-meter resolution). The data were used as model inputs in the form of actual values ​​(e.g., impermeability 35%, population density 2200 people / km²). Future predicted land cover data came from the Wang et al. (2022) dataset (original resolution 1km), and population density data came from the Chen et al. (2020) dataset (original resolution 1km).

[0059] S34 employs a spatial downscaling method to reduce future land cover and population density prediction data to a target resolution that matches the spatial patch scale, in order to generate input data for future urbanization scenarios. It should be noted that, in order to address the mismatch between the future data resolution and the spatial patch scale of the model, a spatial downscaling method is adopted. The downscaling method is an existing technology and will not be elaborated on here. It is described in the paper "Investigating flood exposure induced socioeconomic risk and mitigation strategy under climate change and urbanization at a city scale" proposed by Wenyu Yang in 2023. The land cover and population density prediction data are downscaled from 1 km resolution to 100 m resolution. The downscaled data is updated in the coupled model with actual numerical values ​​(such as the impermeability of a future patch of 42% and the population density of 2500 people / km²) to update the underlying surface impermeability, pollutant accumulation / scour parameters, and the amount of pollutants generated by domestic sewage, so as to ensure the accuracy of the input of future urbanization scenarios.

[0060] S35, the processed climate input data and urbanization input data are combined according to different time dimensions and scenario dimensions to generate multiple sets of future scenario input datasets, and input into the coupled model to simulate the spatiotemporal changes of compound pollution under different scenarios; the time dimension includes near future time period and far future time period, and the scenario dimension includes low greenhouse emissions and high greenhouse emissions.

[0061] It should be noted that the processed actual future climate values ​​(1km resolution) are combined with the actual urbanization values ​​(100m resolution), including near-future time periods (2026-2050) and far-future time periods (2051-2100). The scenario dimensions include SSP1-2.6 (low greenhouse gas emissions) and SSP5-8.5 (high greenhouse gas emissions), generating four types of future scenario input datasets, including near-future time periods SSP1-2.6, near-future time periods SSP5-8.5, far-future time periods SSP1-2.6, and far-future time periods SSP5-8.5. These datasets are then input into the coupled model to systematically simulate the spatiotemporal dynamic changes of compound pollution under different development paths.

[0062] In this implementation, for climate prediction data, a combination of multi-model integration, bias correction, and spatial interpolation was used to downscale the original data with a resolution greater than 250km to a high-resolution grid of 1km, and to achieve accurate spatial matching with sub-basins. For urbanization data, a spatial similarity downscaling method was used to refine the land cover and population density prediction data at a resolution of 1km to a patch level of 100m. This improved the accuracy and matching degree of future scenario input data, thereby improving the accuracy and reliability of future pollution trend prediction.

[0063] S4, based on the urbanization attributes and pollution results of spatial patches, performs attribute classification and spatial correlation visualization to identify high-risk pollution areas; and based on the input variables and output results of the coupled model, uses a random forest model combined with SHAP interpretability analysis to quantify the contribution of climate factors and urbanization factors to pollution results.

[0064] Step S4, which involves classifying and visualizing the spatial correlations of urban attributes and pollution results based on spatial patches to identify high-risk pollution areas, includes the following sub-steps: It should be noted that this step transforms the complex model simulation results into intuitive and actionable spatial decision-making information, enabling the accurate identification and location of high-risk pollution areas through the system's attribute classification and spatial correlation.

[0065] S41. Based on the land cover data and population data of spatial patches, land cover intensity level and population density level are divided for each spatial patch. The land cover intensity is divided into three levels: low, medium and high according to the impermeability of the spatial patch, and the population density level is divided into three levels: low, medium and high according to the population density of the spatial patch. It should be noted that, based on the spatial patches divided by S1 and the high-resolution actual urbanization values ​​processed by S3, the core attributes of each spatial patch are graded and labeled to form patch attribute labels; according to the actual impermeability of the spatial patches, the land cover intensity is divided into three levels: the low level of land cover intensity is impermeability <20%, the medium level is impermeability between 20% and 60%, and the high level is impermeability >60%; according to the population density of the spatial patches, the population density is divided into three levels: low, medium, and high. The low level of population density is population density <1500 people / km², the medium level is population density between 1500 and 2500 people / km², and the high level is population density >2500 people / km².

[0066] S42, based on the simulation results output by the coupled model, classifies the pollutant load level and overflow frequency level for each river segment; It should be noted that key pollution indicators are extracted from the output of the coupled model and graded and labeled for each river segment to form a river segment pollution attribute label. Based on the annual total pollution discharge load of the river segment, the levels are divided into low pollutant load <10 tons / year; medium pollutant load 10-100 tons / year; and high pollutant load >100 tons / year. Based on the annual CSO event occurrence frequency of the river segment, the levels are divided into overflow frequency <5 times / year; medium CSO overflow frequency 5-50 times / year; and high CSO overflow frequency >50 times / year. S43. Associate and label the land cover intensity level and population density level of the spatial patches with the corresponding river segments through the hydraulic connection between the spatial patches and the river segments; It should be noted that, based on the spatial patch-sub-basin-sewage network relationship established in step S1, the land cover intensity level and population density level attributes of the spatial patches are transferred and labeled to the corresponding receiving river sections through hydraulic connection path tracing, thus establishing an urbanization attribute labeling system at the river section scale.

[0067] S44. Cross-statistically analyze the land cover intensity level and population density level associated with the river section with the pollutant load level and overflow frequency level to generate a visualization result representing the spatial distribution of pollution risk under different urbanization intensity levels, so as to identify high-risk river sections.

[0068] It should be noted that cross-statistical analysis and spatial visualization of the urbanization and pollution attributes of river sections were performed. Specifically, four types of correlation analysis revealed the spatial correspondence between urbanization intensity and pollution risk: CSO frequency distribution in river sections with different impermeability levels; pollution load distribution in river sections with different impermeability levels; CSO frequency distribution in river sections with different population density levels; and pollution load distribution in river sections with different population density levels. Figure 3 As shown, thematic maps were created in the ArcGIS platform using a bivariate color system. The left map (a1-a5) uses a blue gradient to represent CSO frequency classification, with dark blue to light blue corresponding to high to low levels. The right map (b1-b5) uses a red gradient to represent pollution load classification, with dark red to light red corresponding to high to low levels. The statistical chart in the middle uses a combined bar chart to show the distribution of pollution indicators under different urbanization attribute levels. BU_H, BU_M, and BU_L represent high, medium, and low impermeable land cover levels, respectively, and PD_H, PD_M, and PD_L represent high, medium, and low population density levels, respectively.

[0069] This method realizes the spatial explicit correlation analysis between urbanization characteristics and pollution response, and can quickly identify key river sections with high development intensity and high pollution risk. It provides intuitive and reliable spatial decision support for the formulation of pollution prevention and control measures, and realizes the effective transformation from pollution simulation to risk management.

[0070] In step S4, based on the input variables and output results of the coupled model, a random forest model combined with SHAP interpretability analysis is used to quantify the contribution of climate and urbanization factors to the pollution results, including the following sub-steps: This step addresses the limitation of traditional simulation methods, which can only output pollution results but cannot analyze the driving mechanisms. By introducing interpretable machine learning methods, it quantitatively reveals the extent and manner in which climate and urbanization factors affect compound pollution.

[0071] S45, Select input variables and output data from the coupled model. The input variables include average temperature, annual rainfall, maximum rainfall, longest drought period, rainfall frequency, drought frequency, associated population, and impervious area. The pollution result indicators include overflow frequency, emission load output by the runoff pollution module, and pollutant load emitted by CSO.

[0072] It should be noted that input and output data were extracted from a large number of simulation scenarios in the basic coupled model to construct a standardized dataset for driving factor analysis; eight key feature variables were selected to comprehensively cover the two dimensions of climate and urbanization. The climate factor dimension includes average temperature, annual rainfall, maximum rainfall, longest drought period, rainfall frequency, and drought frequency; the urbanization factor dimension includes associated population and impervious area; the three types of core pollution results simulated by the model are used as output indicators, including overflow frequency, emission load output by the runoff pollution module, and pollutant load emitted by CSO.

[0073] S46. A prediction model is constructed based on the random forest model. The training dataset is input into the random forest model for training, and the model performance is evaluated using the coefficient of determination to obtain a trained prediction model, which is used to output the overflow frequency, runoff pollution load and CSO pollution load.

[0074] It should be noted that the random forest regression model was built using the ranger package in R, with 1000 decision trees to control computational complexity while ensuring prediction accuracy. A training dataset of over 12,000 samples was constructed by integrating simulated samples from historical periods and all future scenarios. Multicollinearity was tested, and the variance inflation factor of all variables was less than 3, indicating good independence among features and no redundant variables. Model performance was evaluated using the coefficient of determination R², expressed as: ; in, y i,o For the observed values, y i,s For predicted values, The values ​​are the average of the observed values. After testing, the determination coefficients R² of the overflow frequency, runoff pollution load and CSO pollution load are 0.89, 0.85 and 0.91, respectively, all of which are higher than the judgment threshold of 0.80 and meet the quantitative requirements.

[0075] S47, using the K-means algorithm, a predetermined number of representative samples are selected from the training dataset as the benchmark for calculating the SHAP value. The SHAP value of each input variable with respect to the output data is calculated, expressed as: ; In the formula, The baseline value is M, where M is the number of input variables. For the first i The SHAP values ​​of the input variables are given, where a positive SHAP value indicates that pollution is promoted, and a negative SHAP value indicates that pollution is inhibited. The larger the absolute value, the higher the contribution.

[0076] S48. Based on the ranking of SHAP values, identify the climate and urbanization factors that contribute the most to the pollution results. Analyze the nonlinear relationship between key variables and pollution results using dependency graphs, and quantify the overall contribution weight of climate and urbanization factors to the pollution results.

[0077] It should be noted that the input variables were ranked according to the average absolute value of the SHAP values ​​to identify the climate and urbanization factors that contribute the most to various types of pollution. A dependency plot was used to reveal the complex nonlinear relationship between key variables and pollution outcomes. For example, the analysis found that when annual rainfall exceeds 800 mm, the CSO frequency shows an accelerating upward trend with increasing rainfall. The SHAP values ​​of climate and urbanization factors were aggregated to calculate their overall contribution weights to pollution outcomes. Empirical analysis shows that, in the context of this study, urbanization factors contribute 65% to the total compound pollution, while climate factors contribute 35%.

[0078] like Figure 4 This study presents the importance ranking (SHAP value) of multiple driving factors for pollution-related indicators (a1–e1), as well as their dependence on key climate (a2–e2) and socioeconomic (a3–e3) factors, thereby revealing the driving mechanisms of climate change and socioeconomic change on complex water pollution. Five pollution-related indicators are involved: CSO frequency, CSO-induced COD load, CSO-induced TN load, runoff-induced COD load, and runoff-induced TN load.

[0079] Among them, CSO frequency ( Figure 5 As shown in Figure a1, three climatic factors—MaxDry (duration of the longest dry weather), MeanTemp (mean temperature within the sub-basin), and AnnualRainfall (total annual rainfall within the sub-basin)—were identified as key drivers, indicating that both ordinary and extreme hydrological processes jointly regulate CSO behavior. Meanwhile, Impervious Area (the area of ​​impervious cover within the sub-basin) significantly impacted CSO frequency, highlighting the important role of urban sprawl in CSO events. Figure 5 The dependency plot in (a2) shows the typical nonlinear response of CSO frequency to the longest drought period. The SHAP value decreases when the drought duration is close to 9 days, while it increases when the drought duration exceeds 10 days, indicating that ordinary drought events lead to a reduction in CSO, while prolonged drought exacerbates CSO risk. Impervious area (...) was observed. Figure 5 (a3) shows a steep upward trend in dependence on SHAP values, especially in the medium to high range, indicating a significant increase in CSO frequency in highly urbanized areas. Figure 4As shown in b1–e1, two socioeconomic factors, namely impervious area and connected population (where ConnectedPop represents the population within the sub-basin), rank among the top three in all models, indicating that urbanization is a key driver of pollutant load. For climate factors, in addition to annual rainfall, extreme wet and dry weather events (MaxRainfall representing maximum rainfall and longest drought period) also show high importance. Notably, the key climate factors differ between CSO-induced pollutants and runoff-induced pollutants. For CSO-induced pollutants, maximum rainfall (b2–c2) is the key climate driver. Within the range of 0–20,000 cubic meters, there is a significant positive correlation between SHAP values ​​and maximum rainfall, indicating that extreme rainfall is the main factor contributing to pollution peaks. However, for runoff-induced pollutants, the longest drought period shows greater importance (d2–e2). When the longest drought period was 5–9 days, the SHAP value showed a decreasing trend, while when the longest drought period exceeded 10 days, the SHAP value showed an increasing trend. This indicates that ordinary drought periods mitigate runoff-induced pollution to some extent, while severe droughts exacerbate the pollution risk. Meanwhile, when the connected population exceeded 5000, the SHAP value increased rapidly (b3–e3), indicating that densely populated areas are prone to severe CSO and runoff-induced pollution.

[0080] In this embodiment, a driving analysis dataset covering eight key climate and urbanization factors was constructed based on multi-scenario data output by the coupled model. By training a high-precision random forest model and calculating SHAP values, the contribution of factors such as average temperature, annual rainfall, impervious area, and population density was quantified. This not only identifies the dominant factors affecting various types of pollution but also reveals the nonlinear relationship between key variables and pollution outcomes through SHAP dependency graphs. Ultimately, the specific contribution weights of climate and urbanization factors to pollution formation are quantified. This enables pollution control measures to shift from homogeneous governance to targeted regulation, improving the accuracy and scientific nature of environmental decision-making.

[0081] S5, based on spatial patches, conducts patch-based deployment of low-impact development (LID) measures, and uses a coupled model to evaluate the pollution reduction effects of LID measures under normal and extreme events.

[0082] This embodiment addresses the problems of extensive deployment and limited evaluation dimensions in existing LID assessments by establishing a refined deployment method that is deeply integrated with urban land use and constructing a comprehensive evaluation system covering both routine and extreme events, providing data support for the optimized layout and adaptive management of LID measures.

[0083] Step S5 includes the following sub-steps: S51, based on the actual area and proportion of building roofs, roads and green spaces within each spatial patch, matches and deploys corresponding LID measures for various land uses. Among them, green roofs (GR) are deployed on building roofs, permeable pavement (PP) is deployed on roads, and bioretention ponds (BR) are deployed on green spaces. The area of ​​each LID measure accounts for 15% of the total area of ​​the land type to which it is applied. The service life of the LID measures is set to 25 years, and the deployment location is dynamically updated based on land cover data from different simulation cycles. It should be noted that the area of ​​each LID measure accounts for 15% of the total area of ​​the land type to which it is applied, which can achieve the best balance between pollution control effect and economic cost; the service life of LID measures is set at 25 years, and the deployment location is dynamically updated according to land cover data of different simulation periods to ensure that LID deployment keeps pace with urban development and changes.

[0084] S52, based on the coupling model, simulate the combined pollution process before and after the deployment of LID measures, and calculate the direct and indirect reduction of pollutants by LID by comparison; The formula for calculating the direct reduction of runoff pollution by LID is as follows: ; In the formula, Pollutants caused by runoff i Remove the load, For land cover type j LID practice on pollutants during flushing i The removal coefficient, Land cover type j The area; For at a certain point in time t 1. Land Cover Type j Pollutants in upstream surface runoff i The concentration; For at a certain point in time t 1. Land Cover Type j Upstream runoff volume; In terms of land cover type j LID practices for pollutant capture during runoff i The interception coefficient; The formula for calculating the indirect reduction of CSO pollution by LID is as follows: ; In the formula, pollutants i CSO removes load; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The concentration of pollutants; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The amount of sewage; S53. Based on the direct and indirect reduction of pollutants by LID, calculate the long-term performance indicators of LID measures under conventional hydrological conditions, including the annual average runoff reduction rate, the annual average runoff pollution reduction rate, and the annual average CSO pollution reduction rate. Then, classify the rainstorm events during the simulation period into four levels according to the total amount and intensity of rainfall: extreme events, high-severity events, medium-severity events, and low-severity events. Calculate the CSO frequency reduction rate and pollution load reduction rate under each level of event to evaluate the adaptability of LID under extreme climate conditions.

[0085] It should be noted that the expression for the annual average runoff reduction rate is: R q =( Q b,n - Q a,n ) / Q b,n ×100%; In the formula, Q b,n This represents the average annual runoff before LID. Q a,n This represents the average annual runoff after LID (Low-Intensity Discharge); the simulation results are 28%-35%.

[0086] The expression for the annual average runoff pollution reduction rate is: R n =( L b,n - L a,n ) / L b,n ×100%; In the formula, L b,n This indicates the emission load output by the LID pre-runoff pollution module. L a,n This represents the emission load output by the runoff pollution module after LID (Low Identification and Discharge); the simulation results are 36%-42%.

[0087] The expression for the annual CSO pollution reduction rate is: R c =( L b,c - L a,c ) / L b,c ×100%; In the formula, L b,c This indicates the pollutant load of CSO emissions before LID (Limited-Intake Distribution). L a,c This represents the pollutant load emitted by the CSO after LID; the simulation results are 32%-38%.

[0088] The simulation process ranked all rainstorm events during the simulation period based on total rainfall and rainfall intensity, classifying them into four levels: extreme events (top 5%, 24-hour rainfall >150mm), high-severity events (5%-25%, 100-150mm), moderate-severity events (25%-75%, 50-100mm), and low-severity events (75%-100%, <50mm). The CSO frequency reduction rate (e.g., 18%-22% reduction for extreme events, 45%-55% reduction for low-severity events) and pollution load reduction rate were calculated for each level of event to assess the adaptability of LID under extreme climate conditions.

[0089] like Figure 5 As shown, Low Impact Development (LID) practices mitigate complex water pollution by reducing CSO events and removing pollutants. To assess the performance of LID practices in reducing CSO, the CSO frequency (total number of CSOs) at all discharge points during each storm event was calculated, and events were categorized by frequency into high (>50 times), medium (5–50 times), and low (<5 times), as follows. Figure 5 As shown in Figures a1–e1, LID practices effectively reduced high- and intermediate-level events, while having a limited impact on low-level events. In the baseline scenario, the median CSO frequencies for high- and intermediate-level events were 51 and 28, respectively. Implemented LID practices reduced these medians to 47 and 22, respectively, representing an overall reduction of 15%–40%. Under the SSP1-2.6 scenario, climate change and urbanization increased the severity and uncertainty of high-frequency events, with the median rising to 73 in the near future and 87 in the far future, and the 95% confidence interval (CI) expanding from 27–69 in the baseline period to a maximum of 63–118. Implemented LID practices reduced the median to a minimum of 65 (CI: 44–84), representing an overall reduction of 14%–32%. Under the SSP5-8.5 scenario, the severity of high-frequency events was similar to that of SSP1-2.6, but with lower uncertainty. The median occurrences for the near and long-term future periods were 73 and 86, respectively, with CIs of 52–93 and 67–102. However, under SSP5–8.5 scenarios, LID performance gradually declined, and CSO reduction rates dropped to 7%–10%.

[0090] Extreme events contribute significantly to complex water pollution. For example... Figure 5As shown in Figures a2–e5, the emission loads (total pollutant loads) for each storm event are arranged in descending order and categorized into four intensity levels: extreme (top 5%), high (5%–25%), medium (25%–75%), and low (75%–100%). SS (suspended solids) emissions from extreme events varied significantly over time (356–813 kg), contributing a high percentage (19%–41%) to the total load across all events in each scenario. COD (chemical oxygen demand) emissions exhibited a similar pattern, with extreme events contributing 20%–41% (438–744 kg) to the total load. In contrast, TN (total nitrogen) and TP (total phosphorus) emissions from extreme events remained at lower levels (20–38 kg and 4–7 kg, respectively).

[0091] Under the influence of climate change and urbanization, the pollutant removal performance of LID (Low Impact Development) practices has significantly declined. For example... Figure 5 As shown in Figures a2–e5, red numbers represent the overall reduction rate for extreme rainfall events, while green numbers represent the overall reduction rate for ordinary rainfall events (high, medium, and low intensities). In the baseline scenario, LID practices achieve high pollutant removal rates for ordinary events (75%–91%) and moderate removal rates for extreme events (74%–86%), indicating effective control even for pulsed water pollution. However, the removal rate for ordinary rainfall events declines over time, reaching 26%–52% in SSP5–8.5 scenarios, almost approaching the removal rate for extreme events (21%–43443%). On the other hand, in the far-future period of SSP1–2.6, although the removal rate for low-intensity events remains above 70%, the removal rate for extreme events declines significantly to 14%–22%. Such a large difference in reduction rates across different rainfall events suggests a gradual decline in the performance of LID practices in reducing pulsed water pollution, potentially impacting their long-term resilience.

[0092] like Figure 6 As shown, this reflects the fluctuations in the performance of LID practices over the long term due to climate change and urbanization. Figure 6Figures a1–a4 show the annual loads of pollutants emitted under baseline and future scenarios. Light bars represent pollutants generated by runoff, and dark bars represent pollutants generated by CSO. Bars above zero represent pollutants generated, bars below zero represent pollutants removed by LID, and black dots represent the net load of pollutants. Regarding SS (suspended solids), the total annual generation in the baseline scenario is 1378 tons, of which runoff pollutants account for 88% (1214 tons) and CSO pollutants account for 12% (164 tons). Implemented LID practices effectively removed runoff pollutants (932 tons) and CSO pollutants (125 tons), resulting in a net emission of 321 tons. In the far-future period of SSP1-2.6, SS generation increases to a maximum load of 4652 tons, of which runoff pollutants and CSO pollutants account for 96% and 4%, respectively. Despite generating the highest pollutant load, 3,724 tons were removed by LID practices (3,602 tons of runoff pollutants and 122 tons of CSO pollutants), resulting in a moderate net emission level (928 tons). In contrast, although the annual load decreased to 2,224 tons in the long-term future of SSP5-8.5 (1,994 tons of runoff pollutants and 230 tons of CSO pollutants), net emissions still reached the highest level (1,202 tons) due to the decline in LID performance (removals of 851 tons and 171 tons). COD (Chemical Oxygen Demand) showed a similar trend, with loads ranging from 1,116 tons (80% runoff pollutants) to 3,840 tons (94% runoff pollutants). Notably, in the long-term future of SSP1-2.6, both the load removed by LID and the net emissions were higher than those of SS (1,638 tons and 2,202 tons, respectively). The loads of TN (total nitrogen) and TP (total phosphorus) remained at low levels, producing 48–171 tons and 9–34 tons per year, respectively, while the contribution of runoff-induced TN and TP (68–93%) was lower than that of SS and COD (80–97%).

[0093] LID practices have shown varying degrees of effectiveness in removing pollutants generated by runoff and CSO. For example... Figure 6As shown in Figures b1–b4, in the baseline scenario, LID practices removed approximately 69% of runoff-induced SS, with a 95% confidence interval (CI) of 51–85%. Under the influence of climate change and urbanization, the removal rate of runoff-induced SS fluctuated, rising to 79% (CI: 44–97%) in the long-term future of SSP1–2.6, but declining to 43–50% (CI: 39–72%) in other scenarios. The removal rates of COD and TN showed a significant downward trend, decreasing from 81% and 79% at baseline to 19–38% and 23–34% in the future, respectively. Regarding TP, although the removal rate was relatively low at baseline (69%), it showed moderate fluctuations in the future, ranging from 25% to 36%. In contrast, the removal rate of CSO-induced pollutants remained at a high level. Figure 6 (C1–C4)). At baseline, LID practices removed 68% of CSO-induced SS (95% CI: 50–82%), while future removal rates fluctuated between 61–74% (CI: 39–98%). TN and TP showed a similar trend to SS, with a baseline removal rate of 69% and future ranges between 62–74%. Notably, the removal rate of CSO-induced COD increased over time, from 69% (CI: 51–83%) to a maximum of 93% (CI: 47–99%).

[0094] The results indicate that climate change and urbanization have led to a decline in the effectiveness of LID practices in controlling runoff-induced pollution, with removal rates generally decreasing as greenhouse gas (GHG) emissions increase. In contrast, LID practices have shown greater resilience in controlling CSO-induced pollution, maintaining relatively stable removal rates despite a larger 95% CI range.

[0095] This embodiment reveals the control mechanism of LID on compound pollution by establishing a quantitative model of direct and indirect reduction. The constructed multi-dimensional assessment system for regular and extreme events overcomes the limitation of traditional methods that only focus on average performance, provides an LID adaptability assessment tool for responding to extreme rainfall events under climate change, and improves the accuracy of urban stormwater infrastructure planning.

[0096] Secondly, the present invention also provides a simulation and assessment system for urban complex water pollution, implemented using a method for simulating and assessing urban complex water pollution. The system includes: The spatial infrastructure construction module is used to divide the watershed and spatial patch based on the SWMM storm management model as the basic platform, and to establish the spatial relationship between the spatial patch, the watershed, the sewer network and the inspection well. The model coupling construction module is used to construct a hydrodynamic module, a runoff pollution module, and a combined sewer overflow (CSO) pollution module based on spatial correlation, and to establish the coupling between the modules to obtain a coupled model. The output of the hydrodynamic module drives the load calculation of the runoff pollution module and the CSO pollution module, and the pollution loads of the two are summed to obtain the total output of composite pollution. The scenario simulation module is used to acquire climate prediction data and urbanization prediction data for future scenarios, and then downscale and standardize them to generate input data that matches the scale of the coupled model. This input data is then fed into the coupled model to simulate the pollution results under different future scenarios. The pollution risk analysis quantification module is used to perform attribute classification and spatial correlation visualization based on the urbanization attributes and pollution results of spatial patches in order to identify high-risk pollution areas; and based on the input variables and output results of the coupled model, it uses a random forest model combined with SHAP interpretability analysis to quantify the contribution of climate factors and urbanization factors to pollution results. The evaluation module is used for patchy deployment of low-impact development (LID) measures based on spatial patches, and uses a coupled model to evaluate the pollution reduction effectiveness of LID measures under normal and extreme events.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for urban composite water pollution simulation and evaluation, characterized in that, The method comprises the following steps: S1, based on the stormwater management model SWMM as a platform, sub-basins and spatial patches are divided based on surface elevation data, and spatial correlation between spatial patches, sub-basins, sewer networks and inspection wells is established; S2, based on the spatial correlation, a hydrodynamic module, a runoff pollution module and a combined sewer overflow CSO pollution module are respectively constructed, and a coupling model is established by coupling between the modules, the output of the hydrodynamic module drives the load calculation of the runoff pollution module and the CSO pollution module, the pollution loads of the two are summarized to obtain the total output of the composite pollution; S3, obtain climate prediction data and urbanization prediction data of future scenarios, respectively, and perform downscaling and standardization processing to generate input data matching the scale of the coupling model, and input into the coupling model to simulate the pollution results under different future scenarios; S4, based on the urbanization attributes and pollution results of the spatial patches, attribute classification and spatial correlation visualization are performed to identify high-risk pollution areas; Based on the input variables and output results of the coupling model, a random forest model is used in combination with SHAP explainability analysis to quantify the contribution of climate factors and urbanization factors to the pollution results; S5, based on the spatial patches, the patch deployment of low-impact development LID measures is performed, and the coupling model is used to evaluate the pollution reduction effect of LID measures under normal and extreme events.

2. The urban composite water pollution simulation and assessment method of claim 1, wherein: Step S1 includes the following sub-steps: S11, obtain the original surface elevation data of a predetermined resolution, and perform terrain correction, denoising and filling processing to obtain standard surface elevation data; S12, based on the deterministic eight-neighborhood D8 algorithm, perform convergence direction analysis on the standard surface elevation data to generate a global convergence direction matrix; S13, based on the convergence direction matrix and the inspection well position of the sewer network, all pipes that finally converge runoff into the same inspection well are aggregated to form sub-basins; S14, a minimum convergence area threshold is preset, each sub-basin is divided into multiple spatial patches, and the spatial correlation between spatial patches, sub-basins, sewer networks and inspection wells is established.

3. The urban composite water pollution simulation and assessment method of claim 2, wherein, The hydrodynamic module in step S2 includes the following sub-steps: S211, obtain the input data of the coupling model, including historical hydrological observation data, underlying surface attribute data and sewer network parameters, the historical hydrological observation data includes rainfall data and flow data, the underlying surface attribute data includes the impervious rate and soil type of each spatial patch, and the sewer network parameters include the structural parameters of the pipe and inspection well, and the geometric parameters of the overflow weir; S212, take the spatial patch as the basic calculation unit, and use the stormwater management model to simulate the processes from rainfall interception, soil infiltration, surface runoff and pipe convergence; wherein, the rainfall interception and soil infiltration both use the Horton infiltration model to calculate the rainfall interception of vegetation and building roofs; the surface runoff uses the motion wave equation to calculate the surface runoff rate and runoff volume, and then converges into the outlet of the corresponding sub-basin; the pipe convergence uses the Saint-Venant equation set to simulate one-dimensional unsteady flow in the sewer pipe, and calculates the flow, water level in the pipe and the overflow amount at the overflow weir; S213, output the water power simulation results, including the surface runoff rate of each spatial patch, the sub-basin outlet flow, the pipeline flow and water level, and the overflow flow of the overflow weir.

4. The urban composite water pollution simulation and assessment method of claim 3, wherein: The runoff pollution module in step S2 comprises the following sub-steps: S221, simulate the accumulation, flushing and discharge processes of the runoff pollution in the spatial patch as a calculation unit; S222, in the pollution accumulation simulation, calculate the accumulation load of the pollutants on the spatial patch in the dry period by using an exponential accumulation model, and the expression of the exponential accumulation model is: ； wherein B i is a pollutant i the actual accumulation load on the spatial patch, b 1 i,j is a pollutant i the maximum accumulation amount on the land cover type j ; b2 j is the accumulation rate on the land cover type j ; d is the accumulation time; S223, in the pollution flushing simulation, calculate the flushing load of the pollutants in the rainfall period by using a flushing model based on the surface runoff rate data output by the water power module, and the expression of the flushing model is: ； wherein is a pollutant i is the erosion load on a land cover type j ; is the erosion coefficient on a land cover type j ; is the erosion index on a land cover type j ; q t is the runoff flow at a point in time t ; S224, in the pollution discharge simulation, subtract the adsorption amount of the underlying surface from the flushing load to obtain the discharge load, which is the final output of the runoff pollution module.

5. The urban composite water pollution simulation and assessment method of claim 4, wherein: The CSO pollution module in step S2 comprises the following sub-steps: S231, calculate the pollutant load carried by the domestic sewage in each patch and the corresponding pollutant concentration based on the population data of the spatial patch, and the expression is: ； ； wherein L i,j Spatial patch j The load of pollutants i carried by the domestic sewage, Pop j Spatial patch j The number of service population, l i The load of pollutants i Per capita, C i,j Spatial patch j The concentration of pollutants in the domestic sewage i dw The amount of domestic sewage per capita;​ S232, simulate the migration and mixing of the domestic sewage pollutants and the pollutants in the pipeline output by the runoff pollution module to determine the concentration of each pollutant in the inspection well; S233, identify the CSO event based on the overflow flow of the overflow weir output by the water power module, determine that a combined sewer overflow CSO event occurs when the overflow flow of the overflow weir is greater than 0, and distinguish the continuous combined sewer overflow CSO events by using a preset dry time interval, and calculate the pollutant load discharged by the CSO by multiplying the pollutant concentration in the inspection well by the overflow flow; S234, aggregate the discharge load output by the runoff pollution module and the pollutant load discharged by the CSO to obtain the total output of the combined pollution.

6. The urban composite water pollution simulation and assessment method of claim 1, wherein: The step S3 comprises the following sub-steps: S31, obtain historical climate data and future predicted climate data of a plurality of climate models, and the climate data comprises temperature data and rainfall data; S32, correct the deviation of the future predicted climate data based on the historical climate data, downscale the corrected data to a target resolution by using a spatial interpolation method, calculate the statistical values of a plurality of climate model data sets to eliminate abnormal values, and generate climate scenario input data in a future period; S33, obtain historical urbanization scenario data and future predicted urbanization scenario data, and the urbanization scenario data comprises land coverage rate and population density data; S34, downscale the future land coverage and population density prediction data to a target resolution matched with the spatial patch scale by using a spatial downscaling method to generate urbanization scenario input data in a future period; S35, combine the processed climate input data and the urbanization input data according to different time dimensions and scenario dimensions to generate a plurality of future scenario input data sets, and input the future scenario input data sets into the coupled model to simulate the spatiotemporal variation of the combined pollution under different scenarios; the time dimensions comprise a near-future time period and a far-future time period, and the scenario dimensions comprise low greenhouse gas emission and high greenhouse gas emission.

7. The urban composite water pollution simulation and assessment method of claim 6, wherein: The attribute grading and spatial correlation visualization based on the urbanization attribute and pollution result of the spatial patch in step S4 can identify high-risk areas of pollution, including the following sub-steps: S41, based on the spatial patch land cover data and population data, land cover intensity level and population density level are divided for each spatial patch as urbanization attribute level; S42, based on the simulation results output by the coupling model, pollutant load level and overflow frequency level are divided for each river section as pollution result level; S43, the land cover intensity level and population density level of the spatial patch are correlated and labeled to the corresponding river section through the hydraulic connection relationship between the spatial patch and the river section; S44, the urbanization attribute level and pollution result level correlated to the river section are cross-counted to generate a visualization result representing the spatial distribution of pollution risk under different urbanization intensity levels to identify high-risk river sections of pollution.

8. The urban composite water pollution simulation and assessment method of claim 1, wherein: The input variables and output results of the coupling model in step S4 are quantified by using a random forest model combined with SHAP explainability analysis to quantify the contribution of climate factors and urbanization factors to the pollution result, including the following sub-steps: S45, select climate factor and urbanization factor variables from the input variables in the coupling model, and select pollution result indicators from the output results to construct a driving analysis data set, the climate factor and urbanization factor variables include average temperature, annual rainfall, maximum rainfall, longest drought period, rainfall frequency, drought frequency, associated population and impervious area, and the pollution result indicators include overflow frequency, runoff pollution load and CSO pollution load; S46, based on the random forest model, a prediction model is constructed, the training data set is input into the random forest model for training, a mapping relationship from the input variables to the pollution result indicators is established, and the model performance is evaluated by using the coefficient of determination to obtain the trained prediction model; S47, a predetermined number of representative samples are selected from the training data set by K-means algorithm as a benchmark for calculating SHAP values, and the SHAP values of each input variable to the output data are calculated, the expression is: ； wherein is the baseline value, M is the number of input variables, is the SHAP value for the i th input variable. S48, according to the ranking of SHAP values, identify the climate factors and urbanization factors with the highest contribution to pollution results, and analyze the nonlinear relationship between key variables and pollution results through dependence graph to quantify the overall contribution weight of climate factors and urbanization factors to pollution results.

9. The urban composite water pollution simulation and assessment method of claim 8, wherein: The step S5 includes the following sub-steps: S51, based on the actual area and proportion of building roof, road and green land in each spatial patch, match and deploy corresponding low-impact development (LID) measures for each land use type, wherein green roof (GR) is deployed for building roof, permeable pavement (PP) is deployed for road, and bioretention (BR) is deployed for green land, the area of each LID measure accounts for 15% of the total area of the corresponding land use type, and the service life of LID measure is set to 25 years; S52, based on the coupling model, simulate the combined pollution process before the deployment of LID measures, and calculate the direct reduction of runoff pollution and the indirect reduction of CSO pollution by LID. The calculation expression of the direct reduction amount of runoff pollution of the LID is: ； wherein is the removal load of pollutants by runoff, i is the removal coefficient of pollutants by LID practices during the runoff process on the land cover type j i is the area of the land cover type j is the concentration of pollutants in the upstream surface runoff on the land cover type t at the time point j i is the runoff volume upstream of the land cover type at the time point t j is the interception coefficient of pollutants by LID practices during the runoff capture process on the land cover type j i is the interception coefficient of pollutants by LID practices during the runoff capture process on the land cover type​​​​​​​​ The calculation expression of the indirect reduction amount of CSO pollution of the LID is: ； In the formula, pollutants i CSO removes load; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The concentration of pollutants; and These are time points t2 before and after the implementation of LID, from the CSO emission point. k The amount of sewage; S53, based on the direct reduction amount and the indirect reduction amount, calculating performance indicators of the LID measure under hydrological simulation of a long sequence, including annual average runoff reduction rate, annual average runoff pollution reduction rate and annual average CSO pollution reduction rate; and dividing rainstorm events in the simulation period into multiple severity grades according to total rainfall and intensity, respectively calculating CSO frequency reduction rate and pollution load reduction rate under each grade event, to evaluate the adaptability of the LID under extreme climate.

10. A system for urban complex water pollution simulation and evaluation, implemented by the method for urban complex water pollution simulation and evaluation according to any one of claims 1-9, characterized in that the system Comprise: A spatial foundation construction module, configured to divide sub-basins and spatial patches based on surface elevation data on the basis of a stormwater management model (SWMM) as a basic platform, and establish spatial correlation relationships among the spatial patches, the sub-basins, a sewer network and inspection wells; A model coupling construction module, configured to respectively construct a hydrodynamic module, a runoff pollution module and a combined sewer overflow (CSO) pollution module based on the spatial correlation relationships, and establish coupling among the modules to obtain a coupled model, wherein an output of the hydrodynamic module drives load calculation of the runoff pollution module and the CSO pollution module, and the pollution loads of the two modules are aggregated to obtain a total output of composite pollution; A scenario simulation module, configured to obtain climate prediction data and urbanization prediction data of future scenarios, respectively perform downscaling and standardization processing on the data, generate input data matched with a scale of the coupled model, and input the input data to the coupled model to simulate pollution results under different future scenarios; A pollution risk analysis and quantification module, configured to perform attribute grading and spatial correlation visualization based on urbanization attributes of the spatial patches and the pollution results, to identify high-risk areas of pollution; And based on input variables and output results of the coupled model, adopt a random forest model combined with SHAP explainability analysis to quantify contribution degrees of climate factors and urbanization factors to the pollution results; An evaluation module, configured to perform patch deployment of low-impact development (LID) measures based on the spatial patches, and respectively evaluate pollution reduction effects of the LID measures under normal and extreme events by using the coupled model.

Images