Method for identifying key source area of nitrogen and phosphorus non-point source pollution based on weighted feature importance

Abstract

The present application belongs to the technical field of nitrogen and phosphorus non-point source pollution key source area identification, and relates to a nitrogen and phosphorus non-point source pollution key source area identification method based on weighted feature importance. Watershed water system, underlying surface conditions, meteorological, hydrological and water environment data are collected and arranged to establish a watershed nitrogen and phosphorus non-point source pollution tracing basic data set. A RF model, a GBRT model, an XGBoost model and a CatBoost model are used to construct a basic model for simulating total nitrogen and total phosphorus flux, and a Ridge model is used to construct a meta-model for simulating total nitrogen and total phosphorus flux. The basic model is applied to carry out sub-basin feature importance analysis, and the meta-model is applied to carry out basic model weight analysis. By analyzing the weighted feature importance of the sub-basin, the nitrogen and phosphorus non-point source pollution key source area is identified according to the importance from high to low. The present application makes up for the shortage of the existing non-point source pollution key source area identification method based on mechanism model, i.e. more parameters and difficult calibration.

Description

Technical Field

[0001] This invention belongs to the field of eco-hydrology technology under geophysics, specifically involving a method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance. Background Technology

[0002] Source apportionment of nitrogen and phosphorus non-point source pollution in agriculture is a crucial foundation for non-point source pollution control. A few sub-basins often contribute the majority of the pollution load to the entire basin, and sub-basins with a high proportion of agricultural land also exhibit relatively high non-point source pollutant runoff intensity, making them key areas that decisively influence the water quality of receiving water bodies. The randomness, complexity, and lag of nitrogen and phosphorus non-point source pollution's migration and transformation make its pollution load estimation and control challenging. Model simulation is an important means of quantifying pollution load and identifying key source areas, mainly divided into receptor models and diffusion models. Receptor models take polluted areas as the research object, identify pollution sources that have environmental impacts on the study area and contribute to the pollution receptors by measuring the physicochemical properties of pollution receptor samples, and determine the contribution rate of each pollution source to the environmental impact using mathematical methods, mainly including chemical mass balance, multivariate statistical analysis, and stable isotope analysis. Diffusion models take pollution sources as the research object and calculate the environmental impact of pollution sources on the research area based on the emission volume of each pollution source, the research area, meteorological factors, and the physical and chemical properties of pollution. They mainly include the HSPF (Hydrological Simulation Program Fortran) model, the SWAT (Soil and Water Assessment Tool) model, and the GWLF (Generalized Watershed Loading Function) model.

[0003] By applying watershed hydrological and water environment models and combining them with high spatiotemporal resolution satellite remote sensing technology, a series of methods for assessing agricultural nitrogen and phosphorus non-point source pollution loads and identifying key source areas have been proposed. Examples include: a method and device for predicting rainwater runoff non-point source pollution loads based on watersheds (CN202411618904.0); a method, device, electronic equipment, and read / storage medium for predicting non-point source pollution loads (CN202311181692.X); a method and system for identifying agricultural non-point source pollution risks applicable to lake basins (CN202410539369.3); a method and system for assessing regional phosphorus pollution loads (CN202311841844.4); and a method, device, equipment, medium, and product for identifying the spatial pattern of crop non-point source pollution risks (CN202411471422.7). This includes methods for estimating the load of agricultural non-point source pollution in large watersheds using an improved output-inflow coefficient model (CN202411433965.X), a method for assessing agricultural non-point source pollution in confluence zones based on the coupling of CFD and SWAT models (CN202410670863.3), a method for estimating the load of agricultural non-point source pollution in confluence zones based on a mechanistic model (CN202310334947.5), a method for calculating the inflow coefficient of agricultural non-point source pollution in watersheds based on standard data (CN202110296375.7), and a method and system for identifying key source areas of agricultural non-point source pollution (CN202110275391.8).

[0004] While the mechanisms of HSPF, SWAT, and GWLF models are relatively well-defined, their application suffers from issues such as a large number of parameters and difficulties in calibration. With the rise of artificial intelligence, machine learning algorithms have been gradually applied to construct models of nitrogen and phosphorus non-point source pollution. For example, a large-scale watershed total phosphorus simulation method based on deep learning and attention mechanisms (ZL202411505973.0) applies deep learning algorithms to the training of LSTM models and introduces an attention mechanism to improve the model's focus on important time steps and features. This method enables the simultaneous simulation of daily total phosphorus data from hundreds of watersheds, learning the characteristics and dynamic behaviors of hundreds of watersheds, and has the advantages of a wide spatial range and large time span in water quality simulation. However, currently, there is a lack of methods to identify key source areas of nitrogen and phosphorus non-point source pollution using machine learning algorithms to address the problems of numerous parameters and difficult calibration in mechanistic models. Therefore, it is necessary to utilize machine learning algorithms suitable for regression analysis, such as Random Forest (RF), Gradient Boosting Regression Trees (GBRT), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost), to establish watershed total nitrogen and total phosphorus flux models. By comprehensively applying feature importance analysis of the basic and meta-models, the impact of each sub-watershed on the total nitrogen and total phosphorus fluxes of the watershed can be identified, thereby identifying key source areas of nitrogen and phosphorus non-point source pollution. In summary, a method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on machine learning-based weighted feature importance analysis is needed. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance.

[0006] To achieve the objectives of this invention, the following technical solutions are adopted.

[0007] The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance includes the following steps:

[0008] S1. Collect and organize data on watershed systems, underlying surface conditions, meteorology, hydrology, and water environment to establish a basic dataset for tracing nitrogen and phosphorus non-point source pollution in the watershed;

[0009] S2. Construct basic models of total nitrogen flux and total phosphorus flux using RF model, GBRT model, XGBoost model and CatBoost model respectively, taking forest vegetation cover area, grassland vegetation cover area, cultivated land crop cover area, total precipitation data, temperature data and potential evapotranspiration data of each sub-basin in the basic dataset as inputs, and taking total nitrogen flux and total phosphorus flux as outputs respectively. At the same time, construct meta-models of total nitrogen flux and total phosphorus flux using Ridge model, taking the outputs of the basic models of total nitrogen flux and total phosphorus flux based on RF model, GBRT model, XGBoost model and CatBoost model respectively as inputs, and taking total nitrogen flux and total phosphorus flux as outputs respectively.

[0010] S3. Using the RF model, GBRT model, XGBoost model, and CatBoost model respectively, conduct feature importance analysis on the basic models of total nitrogen flux and total phosphorus flux to obtain the feature importance of each input variable on the total nitrogen flux and total phosphorus flux of the watershed. At the same time, use the total nitrogen flux meta-model and the total phosphorus flux meta-model respectively to conduct weight analysis on the basic models of total nitrogen flux and total phosphorus flux based on the RF model, GBRT model, XGBoost model, and CatBoost model to obtain the weights of the basic models of total nitrogen flux and total phosphorus flux of each model.

[0011] S4. Based on the feature importance analysis in step S3, analyze the weighted feature importance of the impact of each sub-basin on the total nitrogen flux and total phosphorus flux of the basin, and identify the key source areas of nitrogen and phosphorus non-point source pollution by ranking them from high to low according to the weighted feature importance.

[0012] As a preferred embodiment of the present invention, the basic dataset for tracing nitrogen and phosphorus non-point source pollution in the watershed includes: watershed system and sub-watershed distribution data; forest area, grassland area, cultivated land area and vegetation coverage data of each sub-watershed; total precipitation data of each sub-watershed; temperature data of each sub-watershed; potential total evapotranspiration data of each sub-watershed; and flow rate, total nitrogen concentration and total phosphorus concentration data at the watershed outlet section.

[0013] As a preferred embodiment of the present invention, the forest vegetation coverage area is obtained by multiplying the forest area by the vegetation coverage within the forest distribution range; the grassland vegetation coverage area is obtained by multiplying the grassland area by the vegetation coverage within the grassland distribution range; and the cultivated land crop coverage area is obtained by multiplying the cultivated land area by the vegetation coverage within the cultivated land distribution range.

[0014] As a preferred embodiment of the present invention, the total nitrogen flux basic model is constructed from the RF model. and the basic model of total phosphorus flux They are represented as follows:

[0015] ;

[0016] In the formula: Representing the RF model, This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins;

[0017] ;

[0018] In the formula: Representing the RF model, This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins;

[0019] As a preferred embodiment of the present invention, a basic model of total nitrogen flux constructed from the GBRT model is used. and the basic model of total phosphorus flux They are represented as follows:

[0020] ;

[0021] In the formula: Represents the GBRT model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins;

[0022] ;

[0023] In the formula: Represents the GBRT model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

[0024] As a preferred embodiment of the present invention, the total nitrogen flux basic model is constructed using the XGBoost model. and the basic model of total phosphorus flux They are represented as follows:

[0025] ;

[0026] In the formula: This represents the XGBoost model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins;

[0027] ;

[0028] In the formula: This represents the XGBoost model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

[0029] As a preferred embodiment of the present invention, the total nitrogen flux basic model is constructed using the CatBoost model. and the basic model of total phosphorus flux They are represented as follows:

[0030] ;

[0031] In the formula: This represents the CatBoost model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: Number of sub-basins

[0032] ;

[0033] In the formula: This represents the CatBoost model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

[0034] As a preferred embodiment of the present invention, the total nitrogen flux element model and the total phosphorus flux element model constructed by the Ridge model are respectively expressed as:

[0035] ;

[0036] In the formula: , , , As input, As output For the Ridge model;

[0037] ;

[0038] In the formula: , , , As input, As output This is the Ridge model.

[0039] As a preferred embodiment of the present invention, a feature importance analysis is conducted on the total nitrogen flux basic model and the total phosphorus flux basic model constructed based on the RF model to obtain the feature importance of the influence of each input variable on the total nitrogen flux and total phosphorus flux of the watershed. Represented as:

[0040] ;

[0041] In the formula: For the sub-basin of the RF-based total nitrogen flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the RF-based total nitrogen flux model The significance of the characteristics of grassland vegetation cover area For the sub-basin of the RF-based total nitrogen flux model The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the RF-based total nitrogen flux model The importance of the characteristics of total precipitation For the sub-basin of the RF-based total nitrogen flux model The importance of temperature characteristics For the sub-basin of the RF-based total nitrogen flux model The significance of the characteristics of total potential evapotranspiration;

[0042] Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the RF-based total nitrogen flux model, the sub-watersheds are statistically analyzed. The characteristic importance of the impact on total nitrogen flux in the watershed is denoted as: :

[0043] ;

[0044] Based on the RF-based total phosphorus flux model, eigenvalue importance analysis was conducted to obtain the eigenvalue importance of the influence of each input variable on the total phosphorus flux of the watershed, denoted as . :

[0045] ;

[0046] In the formula: For the sub-basin of the RF-based total phosphorus flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on RF The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on RF The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on RF The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on RF The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on RF The significance of the characteristics of total potential evapotranspiration.

[0047] Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the RF-based total phosphorus flux model, the sub-watersheds are statistically analyzed. The characteristic importance of the impact on total phosphorus flux in the watershed is denoted as :

[0048] .

[0049] As a preferred embodiment of the present invention, a feature importance analysis is conducted on the total nitrogen flux basic model and the total phosphorus flux basic model constructed based on the GBRT model to obtain the feature of the influence of each input variable on the total nitrogen flux and total phosphorus flux of the sub-basin.

[0050] importance :

[0051] ;

[0052] In the formula: For the sub-basin of the total nitrogen flux basic model based on GBRT The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on GBRT The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on GBRT The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on GBRT The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on GBRT The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on GBRT The significance of the characteristics of total potential evapotranspiration;

[0053] Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the GBRT-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed :

[0054] ;

[0055] Based on the GBRT total phosphorus flux fundamental model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the influence of each input variable on the total phosphorus flux of the watershed. :

[0056] ;

[0057] In the formula: For the sub-basin of the total phosphorus flux basic model based on GBRT The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on GBRT The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on GBRT The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on GBRT The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on GBRT The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on GBRT The significance of the characteristics of total potential evapotranspiration.

[0058] Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the GBRT-based total phosphorus flux model, the sub-watersheds were statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed :

[0059] .

[0060] As a preferred embodiment of the present invention, based on the XGBoost total nitrogen flux fundamental model, feature importance analysis is performed to obtain the feature importance of the influence of each input variable on the total nitrogen flux of the watershed. :

[0061] ;

[0062] In the formula: For the sub-basin of the total nitrogen flux basic model based on XGBoost The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on XGBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on XGBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on XGBoost The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on XGBoost The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on XGBoost The significance of the characteristics of total potential evapotranspiration;

[0063] Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the XGBoost-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed :

[0064] ;

[0065] Based on the XGBoost total phosphorus flux model, eigenvalue importance analysis was conducted to determine the eigenvalue importance of each input variable on the total phosphorus flux of the watershed. :

[0066] ;

[0067] In the formula: For the sub-basin of the total phosphorus flux basic model based on XGBoost The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on XGBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on XGBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on XGBoost The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on XGBoost The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on XGBoost The significance of the characteristics of total potential evapotranspiration.

[0068] Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the XGBoost-based total phosphorus flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed :

[0069] .

[0070] As a preferred embodiment of the present invention, based on the CatBoost total nitrogen flux fundamental model, feature importance analysis is performed to obtain the feature importance of the influence of each input variable on the total nitrogen flux of the watershed. :

[0071] ;

[0072] In the formula: For the sub-basin of the CatBoost-based total nitrogen flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on CatBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on CatBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on CatBoost The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on CatBoost The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on CatBoost The significance of the characteristics of total potential evapotranspiration;

[0073] Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the CatBoost-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed :

[0074] ;

[0075] Based on the CatBoost total phosphorus flux model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed. :

[0076] ;

[0077] In the formula: For the sub-basin of the CatBoost-based total phosphorus flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on CatBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on CatBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on CatBoost The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on CatBoost The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on CatBoost The significance of the characteristics of total potential evapotranspiration;

[0078] Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the CatBoost-based total phosphorus flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed :

[0079] .

[0080] As a preferred embodiment of the present invention, the specific analysis process of the weight analysis is as follows:

[0081] In the total nitrogen flux meta-model based on Ridge , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient, and the weights of the total nitrogen flux basic models based on RF, GBRT, XGBoost, and CatBoost were obtained as follows: , , , :

[0082] ;

[0083] ;

[0084] ;

[0085] ;

[0086] In Ridge's total phosphorus flux element model , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient to obtain the weights of the total phosphorus flux basic models based on RF, GBRT, XGBoost, and CatBoost. , , , :

[0087] ;

[0088] ;

[0089] ;

[0090] .

[0091] As a preferred embodiment of the present invention, the specific process of identification is as follows:

[0092] ;

[0093] In the formula: For sub-basins The weighted characteristic importance of the impact on total nitrogen flux in the watershed will Sort from largest to smallest, the sub-basins with the highest ranking are the key source areas for total nitrogen non-point source pollution, for example, taking their maximum value ( The corresponding sub-basin is a key source area for total nitrogen non-point source pollution;

[0094] ;

[0095] In the formula: For sub-basins The weighted characteristic importance of the impact on total phosphorus flux in the watershed. Sort from largest to smallest, the sub-basins with the highest ranking are the key source areas for total phosphorus non-point source pollution, for example, taking their maximum value ( The corresponding sub-basin is the key source area for total phosphorus non-point source pollution.

[0096] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention is a method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on machine learning weighted feature importance analysis. It overcomes the shortcomings of existing non-point source pollution key source area identification methods based on mechanistic models, which have many parameters and are difficult to calibrate. It improves the system of methods for identifying and controlling key source areas of nitrogen and phosphorus non-point source pollution, enhances the utilization efficiency of data such as watershed water system, underlying surface conditions, meteorology, hydrology, and water environment, provides scientific and technological support for watershed non-point source pollution control, and improves the watershed water quality safety guarantee and comprehensive management capabilities. Attached Figure Description

[0097] Figure 1 This is a flowchart of the method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on machine learning weighted feature importance analysis according to the present invention;

[0098] Figure 2 This is a schematic diagram of a typical small watershed sub-watershed distribution in an example of the present invention;

[0099] Figure 3 This describes the characteristic importance of the influence of each sub-basin on total nitrogen flux in the RF-based total nitrogen flux basic model of this invention.

[0100] Figure 4 This is the characteristic importance of the influence of each sub-basin on total phosphorus flux in the RF-based total phosphorus flux basic model of the present invention;

[0101] Figure 5 This describes the characteristic importance of the influence of each sub-basin on total nitrogen flux in the GBRT-based total nitrogen flux basic model of this invention.

[0102] Figure 6 This is the characteristic importance of the influence of each sub-basin on total phosphorus flux in the GBRT-based total phosphorus flux basic model of the present invention;

[0103] Figure 7 This describes the characteristic importance of the influence of each sub-basin on total nitrogen flux in the XGBoost-based total nitrogen flux basic model of this invention.

[0104] Figure 8 This is the characteristic importance of the influence of each sub-basin on total phosphorus flux in the XGBoost-based total phosphorus flux basic model of this invention;

[0105] Figure 9 This is the characteristic importance of the influence of each sub-basin on total nitrogen flux in the CatBoost-based total nitrogen flux basic model of this invention;

[0106] Figure 10 This is the characteristic importance of the influence of each sub-basin on total phosphorus flux in the CatBoost-based total phosphorus flux basic model of this invention;

[0107] Figure 11 This refers to the weighted characteristic importance of the impact of each sub-basin on the total nitrogen flux of the basin in the examples of this invention;

[0108] Figure 12 This represents the weighted characteristic importance of the impact of each sub-basin on the total phosphorus flux of the basin in the examples of this invention. Detailed Implementation

[0109] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0110] The technical solution of the present invention will be further described below with reference to the accompanying drawings and examples.

[0111] As an embodiment of the present invention, such as Figure 1As shown, the key source area identification method for nitrogen and phosphorus non-point source pollution based on weighted feature importance is applied in a typical small watershed in the Yangtze River Basin. This watershed is located in southwestern Henan Province, at the junction of Hubei, Shaanxi, and Henan provinces, with geographical coordinates between 110°17′~111°50′E and 32°55′~34°01′N. The river originates near Nannihu Village, Lengshui Town, Luanchuan County, flowing from northwest to southeast through Lushi County, Xixia County, Neixiang County, and Xichuan County, finally flowing into the Danjiangkou Reservoir at Shanghe Village, Xichuan County. It is a first-order tributary of the Danjiang River, a tributary of the Han River, a tributary of the Yangtze River. The watershed is situated on the southern slope of the Funiu Mountains, with a terrain that slopes from north to south. The northern part mainly consists of mountains, hills, and ridges, while the southwest part features open, fan-shaped intermontane valleys. The terrain is highly undulating with steep slopes. The basin has a well-developed river system, with major tributaries including the Ding River, Shewei River, Yanzhen River, Guzhuang River, Wangou River, and Jiuligou River.

[0112] Step 1: Collect and organize data on typical small watershed systems, underlying surface conditions, meteorology, hydrology, and water environment to establish a basic dataset for tracing nitrogen and phosphorus non-point source pollution in typical small watersheds. Specifically:

[0113] 1. Collect and organize data on the distribution of typical small watershed systems and sub-watersheds, and denote the sub-watersheds as follows: ,like Figure 2 As shown.

[0114] 2. Collection and organization Data on forest area, grassland area, and cultivated land area for each sub-basin from 2013 to 2023, and monthly vegetation cover data for each sub-basin from 2013 to 2023. The forest area of ​​each sub-basin is multiplied by the vegetation cover within the forest distribution area to obtain the monthly forest vegetation cover area for each sub-basin from 2013 to 2023, in km². 2 , recorded as Multiplying the grassland area of ​​each sub-basin by the vegetation cover within the grassland distribution range yields the monthly grassland vegetation cover area for each sub-basin from 2013 to 2023, in km². 2 , recorded as Multiplying the cultivated land area of ​​each sub-basin by the vegetation cover within the cultivated land distribution area yields the monthly cultivated land crop cover area of ​​each sub-basin from 2013 to 2023, in km². 2 , recorded as .

[0115] 3. Collection and organization Monthly precipitation data for each sub-basin from 2013 to 2023, in 10,000 m³. 3 , recorded as Collection and organization Monthly temperature data for each sub-basin from 2013 to 2023, in °C, denoted as . Collection and organization Potential evapotranspiration data for each sub-basin from 2013 to 2023, in 10,000 m³. 3 , recorded as .

[0116] 4. Collect and organize monthly data on flow rate, total nitrogen concentration, and total phosphorus concentration at the watershed outlet section from 2013 to 2023. Calculate the total nitrogen flux using flow rate and total nitrogen concentration, in tons (t), denoted as [missing data]. Total phosphorus flux was calculated using flow rate and total phosphorus concentration, in tons (t), and denoted as . .

[0117] 5. Data such as forest vegetation coverage area, grassland vegetation coverage area, cultivated land crop coverage area, total precipitation, temperature, and potential evapotranspiration are standardized and preprocessed using the z-score method.

[0118] 6. For data such as forest vegetation coverage area, grassland vegetation coverage area, cultivated land crop coverage area, total precipitation, temperature, potential total evapotranspiration, total nitrogen flux, and total phosphorus flux, randomly divide them into training set, validation set, and test set according to the proportions of 70%, 15%, and 15%, respectively.

[0119] Step 2: Based on the source data of nitrogen and phosphorus non-point source pollution in the watershed from Step 1, construct a machine learning model suitable for regression analysis. The RF model, GBRT model, XGBoost model, and CatBoost model are used to construct the base model, and the Ridge model is used to construct the meta-model. Specifically:

[0120] 1. With Forest vegetation cover area, grassland vegetation cover area, cultivated land crop cover area, total precipitation, temperature, and potential evapotranspiration in each sub-basin were used as inputs, with total nitrogen flux and total phosphorus flux as outputs, respectively. RF model, GBRT model, XGBoost model, and CatBoost model were used to construct basic models for total nitrogen flux and total phosphorus flux, respectively. Specifically:

[0121] (1) with , , , , , As input, As output, the RF model is used ( ), construct a basic model for total nitrogen flux based on radio frequency (RF):

[0122] ;

[0123] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the RF-based basic model for total nitrogen flux.

[0124] Dataset RMSE / t R NSE KGE training set 100.58 0.86 0.73 0.69 Validation set 28.43 0.93 0.81 0.75 test set 399.57 0.61 0.31 0.36

[0125] by , , , , , As input, As output, the RF model is used ( ), construct a basic model for total phosphorus flux based on radio frequency (RF):

[0126] ;

[0127] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the RF-based total phosphorus flux fundamental model.

[0128] Dataset RMSE / t R NSE KGE training set 2.65 0.93 0.84 0.72 Validation set 2.23 0.95 0.90 0.81 test set 1.85 0.51 0.18 0.45

[0129] A long-term simulation of monthly total nitrogen flux from 2013 to 2023 was conducted using a fundamental RF-based model for total nitrogen flux, generating... Using an RF-based total phosphorus flux fundamental model, a long-term simulation of monthly total phosphorus flux from 2013 to 2023 was conducted to generate... .

[0130] (2) with , , , , , As input, As output, the GBRT model is used ( Construct a basic model for total nitrogen flux based on GBRT:

[0131] ;

[0132] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the GBRT-based total nitrogen flux fundamental model.

[0133] Dataset RMSE / t R NSE KGE training set 94.83 0.95 0.90 0.86 Validation set 25.85 0.98 0.96 0.95 test set 117.45 0.19 -2.31 0.08

[0134] by , , , , , As input, As output, the GBRT model is used ( Construct a basic model for total phosphorus flux based on GBRT:

[0135] ;

[0136] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the GBRT-based total phosphorus flux fundamental model.

[0137] Dataset RMSE / t R NSE KGE training set 1.05 0.99 0.97 0.97 Validation set 0.54 0.99 0.99 0.95 test set 4.08 0.94 0.47 0.03

[0138] A long-term simulation of monthly total nitrogen flux from 2013 to 2023 was conducted using the GBRT-based total nitrogen flux fundamental model, generating... A long-term simulation of monthly total phosphorus flux from 2013 to 2023 was conducted using the GBRT-based total phosphorus flux model, generating... .

[0139] (3) with , , , , , As input, As output, the XGBoost model is used ( ), construct a basic model for total nitrogen flux based on XGBoost:

[0140] ;

[0141] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the XGBoost-based total nitrogen flux fundamental model.

[0142] Dataset RMSE / t R NSE KGE training set 112.33 0.99 0.85 0.67 Validation set 61.38 0.99 0.91 0.69 test set 57.77 0.52 -0.02 0.32

[0143] by , , , , , As input, As output, the XGBoost model is used ( ), construct a basic model for total phosphorus flux based on XGBoost:

[0144] ;

[0145] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the XGBoost-based total phosphorus flux fundamental model.

[0146] Dataset RMSE / t R NSE KGE training set 1.50 0.99 0.95 0.87 Validation set 1.70 0.97 0.93 0.84 test set 2.88 0.82 -0.04 0.48

[0147] A long-term simulation of monthly total nitrogen flux from 2013 to 2023 was conducted using the XGBoost-based total nitrogen flux fundamental model, generating... A long-term simulation of monthly total phosphorus flux from 2013 to 2023 was conducted using the XGBoost-based total phosphorus flux model, generating... .

[0148] (4) with , , , , , As input, As output, the CatBoost model is used ( Construct a basic model for total nitrogen flux based on CatBoost:

[0149] ;

[0150] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the basic CatBoost-based model for total nitrogen flux.

[0151] Dataset RMSE / t R NSE KGE training set 207.53 0.73 0.50 0.53 Validation set 84.07 0.82 0.66 0.74 test set 175.76 0.44 -3.25 0.08

[0152] by , , , , , As input, As output, the CatBoost model is used ( Construct a basic model for total phosphorus flux based on CatBoost:

[0153] ;

[0154] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the CatBoost-based total phosphorus flux fundamental model.

[0155] Dataset RMSE / t R NSE KGE training set 1.83 0.94 0.87 0.84 Validation set 8.88 0.62 0.35 0.47 test set 2.25 0.53 -0.11 0.36

[0156] A long-term simulation of monthly total nitrogen flux from 2013 to 2023 was conducted using a CatBoost-based total nitrogen flux fundamental model, generating... A long-term simulation of monthly total phosphorus flux from 2013 to 2023 was conducted using a CatBoost-based total phosphorus flux model, generating... .

[0157] 2. Elementary models for total nitrogen flux and total phosphorus flux were constructed using the Ridge model. Specifically:

[0158] by , , , As input, As output, the Ridge model is used ( ), and construct a total nitrogen flux element model based on Ridge.

[0159] ;

[0160] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the Ridge-based total nitrogen flux element model.

[0161] Dataset RMSE / t R NSE KGE training set 47.49 0.97 0.94 0.96 Validation set 45.68 0.85 0.72 0.77 test set 106.46 0.99 0.95 0.80

[0162] by , , , As input, As output, the Ridge model is used ( ), and construct a total phosphorus flux element model based on Ridge.

[0163] ;

[0164] The root mean square error (RMSE), correlation coefficient (R), Nash efficiency coefficient (NSE), and Kling-Gupta efficiency coefficient (KGE) were used to evaluate the simulation accuracy of the Ridge-based total phosphorus flux element model.

[0165] Dataset RMSE / t R NSE KGE training set 1.26 0.98 0.96 0.94 Validation set 1.95 0.97 0.93 0.90 test set 0.55 0.98 0.94 0.82

[0166] Step 3: Based on the basic model and meta-model constructed in Step 2, apply the basic model to conduct sub-basin feature importance analysis, and apply the meta-model to conduct basic model weight analysis. Specifically:

[0167] 1. Apply basic models such as RF model, GBRT model, XGBoost model, and CatBoost model to conduct sub-basin feature importance analysis. Specifically:

[0168] (1) Using the RF-based total nitrogen flux basic model, feature importance analysis was conducted to obtain the feature importance of the influence of each input variable on the total nitrogen flux of the watershed. Based on this, the sub-watersheds were statistically analyzed. The characteristic importance of its impact on total nitrogen flux in the watershed. Figure 3 This describes the characteristic importance of the influence of each sub-basin on total nitrogen flux in the RF-based total nitrogen flux basic model of this invention.

[0169] Using a fundamental RF-based model of total phosphorus flux, characteristic importance analysis was conducted to determine the characteristic importance of each input variable's impact on the total phosphorus flux of the watershed. Based on this, sub-watershed statistics were performed. The characteristic importance of its impact on total phosphorus flux in the watershed. Figure 4 This describes the characteristic importance of the influence of each sub-basin on the total phosphorus flux in the RF-based total phosphorus flux basic model of this invention.

[0170] (2) Using the GBRT-based total nitrogen flux model, feature importance analysis was conducted to obtain the feature importance of each input variable on the total nitrogen flux of the watershed. Based on this, the sub-watersheds were statistically analyzed. The characteristic importance of its impact on total nitrogen flux in the watershed. Figure 5 This describes the characteristic importance of the influence of each sub-basin on the total nitrogen flux in the GBRT-based total nitrogen flux basic model of this invention.

[0171] Using the GBRT-based total phosphorus flux model, characteristic importance analysis was conducted to determine the characteristic importance of each input variable's impact on the total phosphorus flux of the watershed. Based on this, sub-watershed statistics were performed. The characteristic importance of its impact on total phosphorus flux in the watershed. Figure 6 This describes the characteristic importance of the influence of each sub-basin on the total phosphorus flux in the GBRT-based total phosphorus flux basic model of this invention.

[0172] (3) Using the XGBoost-based total nitrogen flux model, feature importance analysis was conducted to obtain the feature importance of each input variable on the total nitrogen flux of the watershed. Based on this, the sub-watersheds were statistically analyzed. The characteristic importance of its impact on total nitrogen flux in the watershed. Figure 7 This describes the characteristic importance of the influence of each sub-basin on the total nitrogen flux in the XGBoost-based total nitrogen flux basic model of this invention.

[0173] Using an XGBoost-based total phosphorus flux model, feature importance analysis was conducted to determine the feature importance of each input variable's impact on the total phosphorus flux of the watershed. Based on this, sub-watershed statistics were performed. The characteristic importance of its impact on total phosphorus flux in the watershed. Figure 8 This describes the characteristic importance of the influence of each sub-basin on total phosphorus flux in the XGBoost-based total phosphorus flux basic model of this invention.

[0174] (4) Using the CatBoost-based total nitrogen flux model, feature importance analysis was conducted to obtain the feature importance of each input variable on the total nitrogen flux of the watershed. Based on this, the sub-watersheds were statistically analyzed. The characteristic importance of its impact on total nitrogen flux in the watershed. Figure 9 This describes the characteristic importance of the influence of each sub-basin on total nitrogen flux in the CatBoost-based total nitrogen flux basic model of this invention.

[0175] Using a CatBoost-based total phosphorus flux model, feature importance analysis was conducted to determine the feature importance of each input variable's impact on the total phosphorus flux of the watershed. Based on this, sub-watershed statistics were performed. The characteristic importance of its impact on total phosphorus flux in the watershed. Figure 10 This describes the characteristic importance of the influence of each sub-basin on total phosphorus flux in the CatBoost-based total phosphorus flux basic model of this invention.

[0176] 2. Apply the Ridge-based total nitrogen flux element model and the Ridge-based total phosphorus flux element model to conduct weight analysis of basic models such as the RF model, GBRT model, XGBoost model, and CatBoost model. Specifically:

[0177] In the total nitrogen flux meta-model based on Ridge , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient to obtain the weights of the total nitrogen flux basic models based on RF, GBRT, XGBoost, and CatBoost. , , , .

[0178] In Ridge's total phosphorus flux element model , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient to obtain the weights of the total phosphorus flux basic models based on RF, GBRT, XGBoost, and CatBoost. , , , .

[0179] Step 4: Based on the feature importance analysis in Step 3, analyze... The weighted importance of the impact of each sub-basin on the total nitrogen and total phosphorus fluxes of the basin was used to identify key source areas of nitrogen and phosphorus non-point source pollution, ranked from highest to lowest weighted importance. Specifically:

[0180] ;

[0181] Figure 11 This represents the weighted characteristic importance of the impact of each sub-basin on the total nitrogen flux of the watershed in this invention example. Sort from largest to smallest, the sub-basins with the highest ranking are the key source areas for total nitrogen non-point source pollution, for example, taking their maximum value ( Sub-basin 2, corresponding to the total nitrogen non-point source pollution, is a key source area.

[0182] ;

[0183] Figure 12 This represents the weighted characteristic importance of the impact of each sub-basin on the total phosphorus flux of the basin in this invention example. Sort from largest to smallest, the sub-basins with the highest ranking are the key source areas for total phosphorus non-point source pollution, for example, taking their maximum value ( Sub-basin 9, corresponding to the total phosphorus non-point source pollution, is a key source area.

[0184] The application of this invention's method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on machine learning-based weighted feature importance analysis is as follows: It can be applied to constructing machine learning-based total nitrogen flux models and machine learning-based total phosphorus flux models, analyzing the weighted feature importance of the impact of each sub-basin on the total nitrogen and total phosphorus fluxes of the basin, and identifying key source areas of total nitrogen and total phosphorus non-point source pollution in the basin. This invention overcomes the shortcomings of existing mechanistic model-based methods for identifying key source areas of non-point source pollution, which suffer from numerous parameters and difficulties in calibration. It improves the system of methods for identifying and controlling key source areas of nitrogen and phosphorus non-point source pollution, enhances the utilization efficiency of data on watershed systems, underlying surface conditions, meteorology, hydrology, and water environment, provides scientific and technological support for watershed non-point source pollution control, and improves the watershed's water quality safety and comprehensive management capabilities.

[0185] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance, characterized in that: Includes the following steps: S1. Collect and organize data on watershed systems, underlying surface conditions, meteorology, hydrology, and water environment to establish a basic dataset for tracing nitrogen and phosphorus non-point source pollution in the watershed; S2. Construct basic models of total nitrogen flux and total phosphorus flux using RF model, GBRT model, XGBoost model and CatBoost model respectively, taking forest vegetation cover area, grassland vegetation cover area, cultivated land crop cover area, total precipitation data, temperature data and potential evapotranspiration data of each sub-basin in the basic dataset as inputs, and taking total nitrogen flux and total phosphorus flux as outputs respectively. At the same time, construct meta-models of total nitrogen flux and total phosphorus flux using Ridge model, taking the outputs of the basic models of total nitrogen flux and total phosphorus flux based on RF model, GBRT model, XGBoost model and CatBoost model respectively as inputs, and taking total nitrogen flux and total phosphorus flux as outputs respectively. S3. Using the RF model, GBRT model, XGBoost model, and CatBoost model respectively, conduct feature importance analysis on the basic models of total nitrogen flux and total phosphorus flux to obtain the feature importance of each input variable on the total nitrogen flux and total phosphorus flux of the watershed. At the same time, use the total nitrogen flux meta-model and the total phosphorus flux meta-model respectively to conduct weight analysis on the basic models of total nitrogen flux and total phosphorus flux based on the RF model, GBRT model, XGBoost model, and CatBoost model to obtain the weights of the basic models of total nitrogen flux and total phosphorus flux of each model. S4. Based on the feature importance analysis in step S3, analyze the weighted feature importance of the impact of each sub-basin on the total nitrogen flux and total phosphorus flux of the basin, and identify the key source areas of nitrogen and phosphorus non-point source pollution by ranking them from high to low according to the weighted feature importance.

2. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 1, characterized in that: The basic dataset for tracing nitrogen and phosphorus non-point source pollution in the basin includes: water system and sub-basin distribution data; forest area, grassland area, cultivated land area, and vegetation coverage data for each sub-basin; total precipitation data for each sub-basin; temperature data for each sub-basin; potential total evapotranspiration data for each sub-basin; and flow rate, total nitrogen concentration, and total phosphorus concentration data at the basin outlet section.

3. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 2, characterized in that: The forest vegetation coverage area is obtained by multiplying the forest area by the vegetation coverage within the forest distribution range; the grassland vegetation coverage area is obtained by multiplying the grassland area by the vegetation coverage within the grassland distribution range; the cultivated land crop coverage area is obtained by multiplying the cultivated land area by the vegetation coverage within the cultivated land distribution range.

4. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 3, characterized in that: The basic model of total nitrogen flux built from the RF model and the basic model of total phosphorus flux They are represented as follows: ； In the formula: Representing the RF model, This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins; ； In the formula: Representing the RF model, This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

5. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 4, characterized in that: The basic model of total nitrogen flux built from the GBRT model and the basic model of total phosphorus flux They are represented as follows: ； In the formula: Represents the GBRT model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins; ； In the formula: Represents the GBRT model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

6. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 5, characterized in that: The basic model of total nitrogen flux built by the XGBoost model and the basic model of total phosphorus flux They are represented as follows: ； In the formula: This represents the XGBoost model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins; ； In the formula: This represents the XGBoost model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

7. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 6, characterized in that: The basic model of total nitrogen flux built by the CatBoost model and total phosphorus Flux Basic Model They are represented as follows: ； In the formula: This represents the CatBoost model. This represents the total nitrogen flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins; ； In the formula: This represents the CatBoost model. This represents the total phosphorus flux as output. This represents the forest vegetation cover area of ​​each sub-basin, which is used as input. This represents the grassland vegetation cover area of ​​each sub-basin, which is used as input. This represents the crop cover area of ​​each sub-basin, which is used as input. This represents the total precipitation data for each sub-basin, which is used as input. This represents the temperature data for each sub-basin, which is used as input. This represents the total potential evapotranspiration data for each sub-basin, which is used as input. Denotes a sub-basin, where: The number of sub-basins.

8. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 7, characterized in that: The total nitrogen flux element model and the total phosphorus flux element model constructed from the Ridge model are expressed as follows: ； In the formula: , , , As input, As output For the Ridge model; ； In the formula: , , , As input, As output This is the Ridge model.

9. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 8, characterized in that: Based on the RF model, fundamental models for total nitrogen flux and total phosphorus flux were constructed. Characteristic importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total nitrogen and total phosphorus fluxes of the watershed. Represented as: ； In the formula: For the sub-basin of the RF-based total nitrogen flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the RF-based total nitrogen flux model The significance of the characteristics of grassland vegetation cover area For the sub-basin of the RF-based total nitrogen flux model The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the RF-based total nitrogen flux model The importance of the characteristics of total precipitation For the sub-basin of the RF-based total nitrogen flux model The importance of temperature characteristics For the sub-basin of the RF-based total nitrogen flux model The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the RF-based total nitrogen flux model, the sub-watersheds are statistically analyzed. The characteristic importance of the impact on total nitrogen flux in the watershed is denoted as: : ； Based on the RF-based total phosphorus flux model, eigenvalue importance analysis was conducted to obtain the eigenvalue importance of the influence of each input variable on the total phosphorus flux of the watershed, denoted as . : ； In the formula: For the sub-basin of the RF-based total phosphorus flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on RF The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on RF The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on RF The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on RF The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on RF The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the RF-based total phosphorus flux model, the sub-watersheds are statistically analyzed. The characteristic importance of the impact on total phosphorus flux in the watershed is denoted as : 。 10. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 9, characterized in that: Based on the GBRT model, fundamental models for total nitrogen flux and total phosphorus flux were constructed. Characteristic importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total nitrogen and total phosphorus fluxes of the sub-basin. : ； In the formula: For the sub-basin of the total nitrogen flux basic model based on GBRT The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on GBRT The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on GBRT The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on GBRT The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on GBRT The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on GBRT The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the GBRT-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed : ； Based on the GBRT total phosphorus flux fundamental model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the influence of each input variable on the total phosphorus flux of the watershed. : ； In the formula: For the sub-basin of the total phosphorus flux basic model based on GBRT The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on GBRT The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on GBRT The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on GBRT The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on GBRT The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on GBRT The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the GBRT-based total phosphorus flux model, the sub-watersheds were statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed : 。 11. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 10, characterized in that: Based on the XGBoost total nitrogen flux model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed. : ； In the formula: For the sub-basin of the total nitrogen flux basic model based on XGBoost The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on XGBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on XGBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on XGBoost The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on XGBoost The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on XGBoost The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the XGBoost-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed : ； Based on the XGBoost total phosphorus flux model, eigenvalue importance analysis was conducted to determine the eigenvalue importance of each input variable on the total phosphorus flux of the watershed. : ； In the formula: For the sub-basin of the total phosphorus flux basic model based on XGBoost The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on XGBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on XGBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on XGBoost The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on XGBoost The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on XGBoost The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the XGBoost-based total phosphorus flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed : 。 12. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 11, characterized in that: Based on the CatBoost total nitrogen flux model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed. : ； In the formula: For the sub-basin of the CatBoost-based total nitrogen flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total nitrogen flux model based on CatBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total nitrogen flux model based on CatBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total nitrogen flux model based on CatBoost The importance of the characteristics of total precipitation For the sub-basin of the total nitrogen flux model based on CatBoost The importance of temperature characteristics For the sub-basin of the total nitrogen flux model based on CatBoost The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total nitrogen flux of the watershed in the CatBoost-based total nitrogen flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total nitrogen flux in the watershed : ； Based on the CatBoost total phosphorus flux model, eigenvalue importance analysis was conducted to obtain the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed. : ； In the formula: For the sub-basin of the CatBoost-based total phosphorus flux fundamental model The characteristics and importance of forest vegetation cover area For the sub-basin of the total phosphorus flux model based on CatBoost The significance of the characteristics of grassland vegetation cover area For the sub-basin of the total phosphorus flux model based on CatBoost The significance of characteristics of crop cover area of ​​arable land For the sub-basin of the total phosphorus flux model based on CatBoost The importance of the characteristics of total precipitation For the sub-basin of the total phosphorus flux model based on CatBoost The importance of temperature characteristics For the sub-basin of the total phosphorus flux model based on CatBoost The significance of the characteristics of total potential evapotranspiration; Based on the characteristic importance of the impact of each input variable on the total phosphorus flux of the watershed in the CatBoost-based total phosphorus flux model, the sub-watersheds are statistically analyzed. Characteristic importance of the impact on total phosphorus flux in the watershed : 。 13. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 12, characterized in that: The specific analysis process of the weight analysis is as follows: In the total nitrogen flux meta-model based on Ridge , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient, and the weights of the total nitrogen flux basic models based on RF, GBRT, XGBoost, and CatBoost were obtained as follows: , , , : ； ； ； ； In Ridge's total phosphorus flux element model , , , The coefficients are respectively , , , Normalization analysis was performed on each coefficient to obtain the weights of the total phosphorus flux basic models based on RF, GBRT, XGBoost, and CatBoost. , , , : ； ； ； 。 14. The method for identifying key source areas of nitrogen and phosphorus non-point source pollution based on weighted feature importance according to claim 13, characterized in that: The specific process of the identification is as follows: ； In the formula: For sub-basins The weighted characteristic importance of the impact on total nitrogen flux in the watershed will Sorted from largest to smallest, the sub-basins with the highest ranking are the key source areas of total nitrogen non-point source pollution; ； In the formula: For sub-basins The weighted characteristic importance of the impact on total phosphorus flux in the watershed will Sorted from largest to smallest, the sub-basins with the highest ranking are the key source areas for total phosphorus non-point source pollution.

Images