A method and system for preventing and controlling watershed non-point source pollution based on key landscape indicators

Abstract

The present invention relates to a method and system for preventing and controlling watershed non-point source pollution based on key landscape indicators, including: collecting multiple non-point source pollution indicators at each sampling point in the target watershed; obtaining multiple landscape indicators at each sampling point in the target watershed; after inputting the same non-point source pollution indicator and the corresponding landscape indicator of all sampling points into a random forest model, screening out the key landscape indicators of the non-point source pollution indicator and applying a non-parametric deviation reduction algorithm and a bootstrap method to obtain the mutation threshold data of the key landscape indicator; optimizing the landscape pattern according to the response relationship between the key landscape indicator and the non-point source pollution indicator and the mutation threshold of the key landscape indicator, and realizing the prevention and control of non-point source pollution. By screening the key landscape indicators that affect non-point source pollution in the watershed and determining the mutation threshold of the key landscape indicator, the landscape pattern of the target watershed is more accurately optimized, and the non-point source pollution in the watershed is accurately and efficiently prevented and controlled.

Description

Technical Field

[0001] The present invention relates to the field of watershed non-point source pollution prevention and control, and in particular to a watershed non-point source pollution prevention and control method and system based on key landscape indicators. Background Art

[0002] Data from the Second National Pollution Source Census Bulletin show that non-point source pollution has become the main source of pollution in my country's water bodies and an important factor restricting economic development. Unreasonable landscape configuration accelerates the loss of nutrients such as nitrogen and phosphorus, which is the main cause of non-point source pollution. As human activities have greatly changed the natural landscape configuration in the basin, the problem of non-point source pollution has become increasingly prominent. Regulating the spatial configuration of different types of patches from the perspective of landscape planning will be the key to improving the effect of non-point source pollution control and maintaining the long-term effectiveness of governance results. Reasonable planning of landscape patterns has great potential for preventing and controlling non-point source pollution, protecting water quality and ensuring economic development.

[0003] However, researchers mainly select landscape indicators based on their own understanding of the landscape and refer to relevant literature to quantify the landscape pattern and qualitatively analyze its relationship with non-point source pollution. There are differences in the angles and scales of researchers' understanding of landscape patterns, and there has been a long-term debate on the selection of key landscape indicators. In addition, due to the nonlinear response of landscape indicators to non-point source pollution, there are also differences in the regulation thresholds of different landscape indicators in the basin. The screening of landscape indicators and the setting of thresholds in the prior art have low accuracy and great limitations, and cannot establish an effective and reliable standard system for the prevention and control of non-point source pollution in the basin. In this regard, the present invention proposes a method and system for the prevention and control of non-point source pollution in the basin based on key landscape indicators. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for preventing and controlling watershed non-point source pollution based on key landscape indicators, which can accurately screen out key landscape indicators and quantify the mutation thresholds that cause water quality deterioration, and provide a specific and quantifiable design basis for future landscape sustainability planning, thereby effectively preventing and controlling watershed non-point source pollution.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for preventing and controlling watershed non-point source pollution based on key landscape indicators, the method comprising:

[0007] Collect multiple non-point source pollution indicators at each sampling point in the target watershed;

[0008] Acquire multiple landscape indicators at each sampling point in the target watershed, and record them as a landscape indicator set; each non-point source pollution indicator of all the sampling points corresponds to the landscape indicator set;

[0009] For the non-point source pollution index whose index value exceeds the preset value, the same non-point source pollution index of all the sampling points and the corresponding landscape index set are respectively input into the random forest model and the key landscape index corresponding to each non-point source pollution index is screened out in combination with the significance test method;

[0010] Applying a nonparametric deviation reduction algorithm and a bootstrap method to each of the nonpoint source pollution indicators and the corresponding key landscape indicators to obtain a mutation threshold value of each of the key landscape indicators;

[0011] For each of the non-point source pollution indicators, based on the relationship between the key landscape indicators and the non-point source pollution indicators, a landscape planning scheme is designed with the mutation threshold of the key landscape indicators as a standard.

[0012] A watershed non-point source pollution prevention and control system based on key landscape indicators, the system comprising:

[0013] The non-point source pollution index collection module is used to collect various non-point source pollution indicators at each sampling point in the target watershed;

[0014] A landscape index acquisition module is used to acquire a plurality of landscape indexes at each sampling point in the target watershed, which are recorded as a landscape index set; each non-point source pollution index of all the sampling points corresponds to the landscape index set;

[0015] A key landscape index screening module is used to input the same non-point source pollution index of all sampling points and the corresponding landscape index set into the random forest model for non-point source pollution indexes whose index values ​​exceed preset values, and screen out the key landscape index corresponding to each non-point source pollution index in combination with a significance test method;

[0016] A key landscape indicator threshold calculation module, used for applying a non-parametric deviation reduction algorithm and a bootstrap method to each of the non-point source pollution indicators and the corresponding key landscape indicators to obtain a mutation threshold of the key landscape indicator;

[0017] The landscape pattern optimization module adjusts the landscape pattern in the target watershed according to the mutation threshold data of each key landscape indicator corresponding to each non-point source pollution indicator and the response relationship between each non-point source pollution indicator and each corresponding key landscape indicator, so as to achieve the prevention and control of non-point source pollution.

[0018] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0019] The present invention relates to a method and system for preventing and controlling watershed non-point source pollution based on key landscape indicators, including: collecting multiple non-point source pollution indicators at each sampling point in a target watershed; obtaining multiple landscape indicators at each sampling point in the target watershed, recorded as a landscape indicator set; for each non-point source pollution indicator, inputting the same non-point source pollution indicator of all sampling points and the corresponding landscape indicator set into a random forest model and screening out the key landscape indicators corresponding to each non-point source pollution indicator in combination with a significance test method; applying a non-parametric deviation reduction algorithm and a bootstrap method to each non-point source pollution indicator and each corresponding key landscape indicator to obtain the mutation threshold data of the key landscape indicator corresponding to each non-point source pollution indicator; adjusting the landscape pattern in the target watershed according to the response relationship between the non-point source pollution indicator and the corresponding key landscape indicator, and the mutation threshold data of the corresponding key landscape indicator, so as to achieve prevention and control of non-point source pollution in the watershed. The random forest algorithm is used to screen the key landscape indicators that affect non-point source pollution in the watershed, and the non-parametric deviation reduction algorithm and the bootstrap method are used to determine the mutation threshold of the key landscape indicator, so as to optimize the landscape pattern of the target watershed according to the mutation threshold of the key landscape indicator, and accurately and efficiently prevent and control non-point source pollution in the watershed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A flow chart of a method for preventing and controlling watershed non-point source pollution based on key landscape indicators provided in Example 1 of the present invention;

[0022] Figure 2 A block diagram of a watershed non-point source pollution prevention and control system based on key landscape indicators provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] At present, there are more than 200 landscape indices that describe landscape pattern characteristics from the perspectives of fragmentation, shape, connectivity, diversity, etc. However, from the perspective of actual planning, a few indicators can comprehensively evaluate the characteristics of watershed landscape pattern. Therefore, a widely applicable indicator screening method is developed to select key landscape indicators that have regulatory functions for non-point source pollutants in the target watershed (riparian zone scale and / or sub-basin scale), and quantify the threshold of sudden change in water quality caused by them. The process of quantifying the threshold of sudden change in landscape indicators is extended to a general research paradigm, providing a scientific basis for watershed landscape planning, and becoming a key technology for preventing and controlling non-point source pollution in watersheds.

[0025] The purpose of the present invention is to provide a method and system for preventing and controlling watershed non-point source pollution based on key landscape indicators, which can accurately screen out key landscape indicators that significantly affect watershed non-point source pollution and quantify the mutation threshold that causes water quality deterioration, so as to provide a specific and quantifiable design scheme for future landscape sustainability planning, thereby effectively preventing and controlling watershed non-point source pollution.

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1

[0028] like Figure 1 As shown, this embodiment 1 provides a method for preventing and controlling watershed non-point source pollution based on key landscape indicators, the method comprising:

[0029] S1: Collect multiple non-point source pollution indicators at each sampling point in the target watershed;

[0030] Several sampling points are set up in the target watershed, and surface water of the watershed is collected at different times. The indicators of non-point source pollution are screened with reference to the Surface Water Environmental Quality Standard (GB3838-2002). In order to ensure the accuracy of subsequent calculations, the cumulative sampling is required to be greater than 50.

[0031] S2: obtaining a plurality of landscape indicators at each sampling point in the target watershed, which are recorded as a landscape indicator set; the non-point source pollution indicators of all sampling points correspond to the landscape indicator set;

[0032] Considering the scale effect of the impact of non-point source pollution on the landscape pattern in the target watershed, in order to optimize the landscape pattern of the target watershed, the target watershed can be divided into sub-watersheds and riparian zones. Set the sampling point as the outlet to divide the sub-watershed in ArcGIS software, and select an appropriate buffer distance to divide the riparian zone according to the size of the watershed. A 100m buffer zone can be selected as the riparian zone, and the value can be adjusted according to actual needs.

[0033] The scale division results of the target watershed were input into Fragstats4.2 to calculate 109 commonly used landscape indicators such as shape, connectivity, edge density, etc. at the class level, and the riparian zone landscape indicator set and sub-watershed landscape indicator set were obtained, which respectively characterize the landscape patterns at the riparian zone scale and the sub-watershed scale.

[0034] S3: For non-point source pollution indicators whose index values ​​exceed the preset values ​​(i.e., non-point source pollution indicators with pollution), the same non-point source pollution indicators of all sampling points and the corresponding landscape indicator set are input into the random forest model and the key landscape indicators corresponding to the non-point source pollution indicators are screened out in combination with the significance test method; the preset values ​​can be determined according to the national standards for non-point source pollution indicators.

[0035] The non-point source index selected in step S1 (this embodiment uses total nitrogen TN as an example for illustration, which is not limiting) and the corresponding landscape index calculated in step S2 are input into the random forest model to evaluate the importance of different landscape indicators to the change of each non-point source pollution index (total nitrogen TN) in the watershed. The mean square error (%IncMSE) of the average increase in model parameters is used as a measurement indicator to quantify the influence of landscape indicators of different scales on the change of watershed TN, and a significance test is performed.

[0036] If the target watershed is divided into sub-watersheds and riparian zones, each non-point source pollution index and the corresponding landscape index in the sub-watershed or riparian zone are input into the random forest model respectively to obtain each non-point source pollution index and key landscape index in the sub-watershed, and each non-point source pollution index and key landscape index in the riparian zone.

[0037] Step S3 specifically includes:

[0038] Step S31: For the non-point source indicators with pollution, the indicators of all sampling points and the corresponding landscape indicator sets are combined to obtain the original sample data set of the non-point source pollution indicators; one non-point source pollution indicator and the corresponding landscape indicator set of each sampling point are a sample; the landscape indicator is used as the independent variable, and the TN indicator is used as the dependent variable.

[0039] Step S32: for each original sample data set of the non-point source pollution index, extract the same number of samples based on the method of resampling with replacement to form multiple new data sets, and use the new data sets to train the decision tree in the random forest model. Among them, one new data set corresponds to one decision tree.

[0040] Step S33: For each decision tree, the out-of-bag data of the decision tree is used to calculate the mean square error of each decision tree, which is recorded as the first mean square error MSE1.

[0041] Out-of-bag data refers to data that is not involved in the establishment of a decision tree when training a decision tree through resampling with replacement. Assume that D is a data set containing M samples. We perform M resampling with replacement on it, thus obtaining a new data set D' of size M. We can be sure that the new data set must contain repeated sampling of a sample in the original data set, so it can be estimated. In each round of sampling, the probability of sample x being drawn is 1 / M, and the probability of not being drawn is 1-1 / M. Therefore, after M rounds of sampling, the probability calculation formula for the sample still not being drawn is as follows:

[0042]

[0043] Therefore, approximately 36.8% of the samples in the original data set are not drawn, and these samples can be used to verify the decision tree trained with the new data set D'.

[0044] MSE is calculated by the following formula:

[0045]

[0046] In the formula, E is the number of samples of out-of-bag data, f(x e ) is the predicted value of a non-point source pollution index (TN) of the e-th sample in the out-of-bag data, y e Represents the actual value of the surface source pollution index (TN) corresponding to the e-th sample in the original sample data set.

[0047] Step S34: shuffling the landscape indicators in the out-of-bag data of each decision tree to obtain the out-of-bag data after shuffling, and calculating the mean square error of each out-of-bag data after shuffling according to the decision tree predicted value and the actual value of the non-point source pollution indicator to obtain a second mean square error MSE2;

[0048] When the order of the characteristic values ​​of a column of landscape indicators is disrupted, it is necessary to ensure that other characteristic variables of the out-of-bag data remain unchanged, and then use the decision tree to predict the samples after the disrupted order. The mean square error between the predicted value and the true value is MSE2.

[0049] Step S34 specifically includes:

[0050] Step S341: for each out-of-bag data, a landscape index is randomly selected, recorded as the selected landscape index, and the order of the currently selected landscape indexes at all the sampling points is shuffled to obtain the out-of-bag data after shuffling;

[0051] Step S342: for each out-of-bag data after the shuffle, the mean square error is calculated according to the decision tree prediction value and the actual value of the non-point source pollution index of the out-of-bag data after the shuffle, to obtain the second mean square error MSE2 corresponding to the currently selected landscape index;

[0052] Step S343: Obtain the landscape indicators that have not been selected and return to the step of "arbitrarily select one of the landscape indicators, record it as the selected landscape indicator, shuffle the order of the current selected landscape indicators at all the sampling points, and obtain the shuffled out-of-bag data", until all the landscape indicators in the out-of-bag data are traversed to obtain the MSE2 corresponding to each landscape indicator.

[0053] For ease of understanding, assume that the non-point source pollution index is TN; then the landscape index set corresponding to TN includes landscape indicators X1, X2, X3, ..., X m ; Therefore, the original data set can be regarded as a matrix with each landscape indicator as a column vector and the sampling points as row vectors. When the order is disordered, the order of the TN values ​​of all sampling points of a landscape indicator is disrupted. For example, a landscape indicator has values ​​at 5 sampling points, and the data is 1, 2, 3, 4, 5. After random arrangement, this column of data may be 5, 2, 3, 1, 4. After the disorder, keep the other column data unchanged, that is, create the original variable as a "bad" variable that does not match the actual situation, and then calculate the MSE2 of the landscape indicator. Each decision tree can obtain the MSE2 of all landscape indicators.

[0054] Step S35: determining a first mean score of each of the landscape indicators according to the first mean square error and the second mean square error;

[0055] Step S35 specifically includes:

[0056] Compare the first mean square error with the second mean square error of each of the landscape indicators to determine a first score for each of the landscape indicators;

[0057] The first scores of the same landscape indicator of all the out-of-bag data are averaged to obtain a first average score of each landscape indicator.

[0058] Since the MSE2 indicator of each landscape is created using completely random "bad" variables that are inconsistent with reality, we expect MSE2>MSE1 (the higher the MSE, the worse the model results). IncMSE=MSE2–MSE1, so the higher the IncMSE, the greater the error caused by the current corresponding disordered landscape indicator to the prediction result, and the more important the un-disordered landscape indicator is;

[0059] Traverse the entire random forest model and get the landscape index X m Impact on all decision trees (each decision tree includes incMSE m1 , incMSE m2 ), then calculate the landscape index X mThe overall impact on the accuracy of the random forest model is the average increase in mean squared error (% incMSE), which is the sum of the incMSE of all decision trees. m Then take the average and get the average increased mean square error %incMSE m , %incMSE Xm As a landscape indicator X m Importance score.

[0060] Step S36: randomly permuting the first mean square error and the second mean square error corresponding to each landscape indicator in all the decision trees, and determining the second mean score of each landscape indicator according to the permuted mean square error;

[0061] Step S36 specifically includes:

[0062] Step S361: Randomly permuting the first mean square error MSE1 and the second mean square error MSE2 corresponding to each of the landscape indicators in each of the decision trees;

[0063] Detailed description of random permutation: The order of MSE1 and MSE2 of each decision tree for each landscape indicator in the random forest is shuffled. For example, the mean square error of the original and shuffled decision tree of the i-th decision tree is MSE i1 , MSE x2 , the mean square error of the original and shuffled decision trees of the xth tree is MSE x1 , MSE i2 .

[0064] Step S362: determining a second score for each of the landscape indicators according to the first mean square error and the second mean square error after substitution corresponding to each of the landscape indicators;

[0065] After the mean square error of the i-th decision tree and the x-th decision tree is permuted, IncMSE i =MSE x2 -MSE i1 ,IncMSEx=MSE i2 -MSE x1 .

[0066] Step S363: taking the average of the second scores of the same landscape indicator corresponding to all the decision trees to obtain a second average score of each landscape indicator.

[0067] Step S37: performing a significance test on the first mean score and the second mean score of each of the landscape indicators, and screening out the key landscape indicators corresponding to each of the non-point source pollution indicators.

[0068] Evaluate whether the difference between the replaced %incMSE (second mean score) and the original %incMSE (first mean score) reaches a significant level (p<0.05). If the significance test is passed, it is considered that this indicator is a key landscape indicator affecting TN.

[0069] S4: applying a non-parametric bias reduction algorithm (nCPA) and a bootstrap method to each of the non-point source pollution indicators and the corresponding key landscape indicators to obtain a mutation threshold of each of the key landscape indicators;

[0070] Step S4 specifically includes:

[0071] Step S41: sorting the non-point source pollution indicators in the original sample data set of each non-point source pollution indicator according to the numerical value to obtain each sorted non-point source pollution indicator;

[0072] Step S42: for each of the sorted non-point source pollution indicators, calculating the total deviation value D of the non-point source pollution indicator; Where D is the entire set of numbers Y1, Y2, Y3, ...Y n The deviation, K is the number of samples in the original sample data set, Y k is the value of a non-point source pollution index in the kth sample (TN variable value), μ is the K Y k The average value of .

[0073] Step S43: for each of the sorted non-point source pollution indicators, a plurality of indicator value segmentation points are randomly selected, and the sorted non-point source pollution indicators are divided into a first group of indicators and a second group of indicators according to each of the indicator value segmentation points;

[0074] By selecting the split point, the sample is divided into two different random samples (the first group of indicators and the second group of indicators) without significant trend changes. For example, the TN indicator is divided into two groups (Y1, Y2, ..., Y g and Y g+1 …Y n , 1≤g≤n), so as to compare and analyze the respective characteristics of the two random samples and the differences between them.

[0075] Step S44: for each of the index value segmentation points, respectively calculating the whole group deviation values ​​of the non-point source pollution index for the first group of indexes and the second group of indexes, recorded as the first group deviation value and the second group deviation value;

[0076] Calculate the sum of the first group of deviation values ​​and the second group of deviation values, and record it as group deviation sum;

[0077] Calculate the difference between the total deviation value and the group deviation sum, and record it as the total group difference;

[0078] Step S45: Select the index value segmentation point corresponding to the maximum value in the total group difference as the mutation point of the non-point source pollution index, and determine the value of each key landscape index corresponding to each mutation point of the non-point source pollution index as the mutation threshold of the key landscape index.

[0079] For each split point g, the two groups of data divided have two deviations, and the split point with the largest difference between the sum and the total deviation is the mutation point;

[0080] Δg=D-(D ≤g +D ＞g )

[0081] Where D is the entire number set Y1, Y2, Y3, ..., Y n The deviation of , D≤g and D>g are the deviations of the two columns of samples, g=1, 2, 3, ..., n, Δg is the difference between the deviation after segmentation and the total deviation. At this time, the landscape index corresponding to the value at g of the divided TN is the mutation threshold of the landscape index, that is, when the change of the landscape index exceeds the mutation threshold, the corresponding non-point source pollution index will mutate.

[0082] In order to ensure that the mutation threshold of the key landscape indicator is selected more accurately, multiple samples can be newly created based on the bootstrap method for calculation, so as to determine the variation range of the mutation threshold of the same key landscape indicator, so that the landscape pattern can be subsequently optimized based on the variation range. The mutation threshold of each key landscape indicator is calculated in step S4, specifically including:

[0083] Extracting multiple sub-sample sets with the same number of samples from the original sample data set of each non-point source pollution indicator;

[0084] For each of the sub-sample sets, according to the non-point source pollution index and each of the corresponding key landscape indicators combined with a non-parametric deviation reduction algorithm, the mutation threshold of each of the key landscape indicators corresponding to each of the non-point source pollution indicators is obtained;

[0085] For each of the key landscape indicators corresponding to each of the non-point source pollution indicators, the mutation thresholds of the same key landscape indicator of all the sub-sample sets are sorted to obtain the mutation threshold range of the key landscape indicator corresponding to each of the non-point source pollution indicators. Specifically, the obtained mutation thresholds are arranged by size, and according to the normality of the data, the mutation point values ​​with a confidence interval of [2.5%, 97.5%] are retained to obtain the mutation threshold range of the key landscape indicator.

[0086] S5: For each of the non-point source pollution indicators, based on the relationship between the key landscape indicators and the non-point source pollution indicators (positive correlation or negative correlation), a landscape planning scheme is designed with the mutation threshold of the key landscape indicators as a standard.

[0087] Taking TN as the research object, it is found that the key landscape indicators affecting TN at the riparian zone scale and the sub-basin scale are different, and the corresponding landscape management measures also change accordingly. The landscape planning scheme for controlling total nitrogen is as follows;

[0088] (1) Maintain the maximum patch index (LPI) of riparian grassland>4.19%, the grassland edge density index (ED)>27.99m / ha, and the farmland shape index (LSI)<3.2.

[0089] (2) Maintain the construction land edge density index (ED) <1.69m / ha, the construction land shape index (LSI) <2.46, and the farmland similarity proximity percentage index (PLADJ) <89.0%.

[0090] By investigating and counting the main non-point source pollution types and landscape indexes in the basin, a random forest model was established to identify the key landscape indicators that affect non-point source pollution. The potential for non-point source pollution prevention and control was developed from qualitative landscape pattern analysis, and scientific planning standards were established for basin landscape planning in a quantitative manner by calculating the mutation threshold of key landscape indicators affecting non-point source pollutants. From the perspective of landscape ecology, this method achieves the goal of controlling pollution with development. This method not only achieves the goal of controlling non-point source pollution, protecting water quality and ensuring economic development, but also provides a theoretical basis for the subsequent improvement of environmental governance measures from the perspective of landscape planning, and thus provides the possibility of achieving intensive use of cultivated land, economic maximization, environmental optimization, and reducing regional non-point source pollution during urban expansion.

[0091] By considering the types of major non-point source pollutants in the basin, key landscape indicators are screened at different scales (sub-basin scale and riparian zone scale) to quantify the landscape pattern. By simulating and optimizing the impact of landscape spatial configuration on non-point source pollution, a landscape pattern optimization plan is established to achieve coordinated regulation between economic benefits, agricultural production, urban planning and non-point source pollution. This method transforms qualitative regulation into quantitative planning schemes, combines ecological monitoring with intelligent algorithms to form a clear, simple and unified research paradigm, which is conducive to macro-control of non-point source pollution in the basin and solving the problem of eutrophication of water bodies.

[0092] Example 2

[0093] like Figure 2 As shown, this embodiment provides a watershed non-point source pollution prevention and control system based on key landscape indicators, and the system includes:

[0094] The non-point source pollution index collection module T1 is used to collect various non-point source pollution indicators at each sampling point in the target watershed;

[0095] The landscape index acquisition module T2 is used to acquire a plurality of landscape indexes at each sampling point in the target watershed, which are recorded as a landscape index set; each non-point source pollution index of all the sampling points corresponds to the landscape index set;

[0096] The key landscape index screening module T3 is used to input the same non-point source pollution index of all the sampling points and the corresponding landscape index set into the random forest model for the non-point source pollution index whose index value exceeds the preset value, and screen out the key landscape index corresponding to each non-point source pollution index by combining the significance test method;

[0097] The key landscape index threshold calculation module T4 is used to apply a non-parametric deviation reduction algorithm and a bootstrap method to each of the non-point source pollution indicators and the corresponding key landscape indicators to obtain a mutation threshold of each of the key landscape indicators;

[0098] The landscape pattern optimization module T5 is used to design a landscape planning scheme for each of the non-point source pollution indicators based on the relationship between the key landscape indicators and the non-point source pollution indicators and taking the mutation threshold of the key landscape indicators as the standard.

[0099] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0100] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for preventing and controlling watershed non-point source pollution based on key landscape indicators. It is characterized in that The method comprises: Collect multiple non-point source pollution indicators at each sampling point in the target watershed; Acquire multiple landscape indicators at each sampling point in the target watershed, and record them as a landscape indicator set; each non-point source pollution indicator of all the sampling points corresponds to the landscape indicator set; For the non-point source pollution index whose index value exceeds the preset value, the same non-point source pollution index of all the sampling points and the corresponding landscape index set are input into the random forest model and combined with the significance test method to screen out the key landscape index corresponding to each non-point source pollution index; Applying a nonparametric deviation reduction algorithm and a bootstrap method to each of the nonpoint source pollution indicators and the corresponding key landscape indicators to obtain mutation threshold data for each of the key landscape indicators; For each of the non-point source pollution indicators, based on the relationship between the key landscape indicators and the non-point source pollution indicators, a landscape planning scheme is designed with the mutation threshold of the key landscape indicators as the standard; For the non-point source pollution index whose index value exceeds the preset value, the same non-point source pollution index of all the sampling points and the corresponding landscape index set are respectively input into the random forest model and the key landscape index corresponding to each non-point source pollution index is screened out in combination with the significance test method, specifically including: For the non-point source index with pollution, the index of all sampling points and the corresponding landscape index set are combined to obtain the original sample data set of the non-point source pollution index; one non-point source pollution index and the corresponding landscape index set of each sampling point are one sample; For each of the original sample data sets of the non-point source pollution indicators, extracting the same number of samples based on a resampling method with replacement to form multiple new data sets, and using the new data sets to train the decision tree in the random forest model; For each of the decision trees, the mean square error of each decision tree is calculated using the out-of-bag data of the decision tree, which is recorded as the first mean square error; The landscape indicators in the out-of-bag data of each decision tree are respectively arranged in random order to obtain the out-of-bag data after random order, and for each out-of-bag data after random order, the mean square error is calculated according to the decision tree predicted value and the actual value of the non-point source pollution indicator to obtain a second mean square error; Determine a first mean score for each of the landscape indicators according to the first mean square error and the second mean square error; Randomly permuting the first mean square error and the second mean square error corresponding to each landscape indicator in all the decision trees, and determining a second mean score for each landscape indicator according to the permuted mean square error; A significance test is performed on the first mean score and the second mean score of each of the landscape indicators to screen out the key landscape indicators corresponding to each of the non-point source pollution indicators.

2. The method according to claim 1, It is characterized in that The landscape indicators in the out-of-bag data of each decision tree are respectively arranged in random order to obtain the out-of-bag data after random order, and for each out-of-bag data after random order, the mean square error is calculated according to the decision tree predicted value and the actual value of the non-point source pollution indicator to obtain the second mean square error, specifically including: For each of the out-of-bag data, a landscape index is randomly selected, recorded as the selected landscape index, and the order of the current selected landscape indexes at all the sampling points is shuffled to obtain the out-of-bag data after shuffling; For each out-of-bag data after the shuffle, the mean square error is calculated according to the decision tree predicted value and the actual value of the non-point source pollution index of the out-of-bag data after the shuffle, to obtain the second mean square error corresponding to the currently selected landscape index; Obtain the unselected landscape indicators and return to step "arbitrarily select one of the landscape indicators, record it as the selected landscape indicator, shuffle the order of the current selected landscape indicators at all the sampling points to obtain the shuffled out-of-bag data", until all the landscape indicators in the out-of-bag data are traversed to obtain the second mean square error corresponding to each landscape indicator.

3. The method according to claim 2, It is characterized in that Determining a first mean score of each of the landscape indicators according to the first mean square error and the second mean square error specifically includes: Compare the first mean square error with the second mean square error of each of the landscape indicators to determine a first score for each of the landscape indicators; The first scores of the same landscape indicator of all the out-of-bag data are averaged to obtain a first average score of each landscape indicator.

4. The method according to claim 1, It is characterized in that The randomly permuting the first mean square error and the second mean square error corresponding to each landscape indicator in all the decision trees, and determining the second mean score of each landscape indicator according to the permuted mean square error, specifically includes: Randomly permuting the first mean square error and the second mean square error corresponding to each of the landscape indicators in each of the decision trees; Determine a second score for each of the landscape indicators according to the first mean square error and the second mean square error after substitution corresponding to each of the landscape indicators; The second scores of the same landscape indicator corresponding to all the decision trees are averaged to obtain a second average score for each landscape indicator.

5. The method according to claim 1, It is characterized in that The non-parametric deviation reduction algorithm and the bootstrap method are applied to each of the non-point source pollution indicators and the corresponding key landscape indicators to obtain the mutation threshold of each of the key landscape indicators, specifically including: Sorting the non-point source pollution indicators in the original sample data set of each non-point source pollution indicator according to the numerical value to obtain each sorted non-point source pollution indicator; For each of the ranked non-point source pollution indicators, calculating the total deviation value of the non-point source pollution indicator; For each of the ranked non-point source pollution indicators, a plurality of indicator value segmentation points are randomly selected, and the ranked non-point source pollution indicators are divided into a first group of indicators and a second group of indicators according to each of the indicator value segmentation points; For each of the index value segmentation points, the whole group deviation values ​​of the non-point source pollution index are calculated for the first group of indexes and the second group of indexes, respectively, and recorded as the first group deviation value and the second group deviation value; Calculate the sum of the first group of deviation values ​​and the second group of deviation values, and record it as group deviation sum; Calculate the difference between the total deviation value and the group deviation sum, and record it as the total group difference; The index value segmentation point corresponding to the maximum value in the total group difference is selected as the mutation point of the non-point source pollution index, and the value of the key landscape index corresponding to the mutation point of each non-point source pollution index is determined as the mutation threshold of each key landscape index.

6. The method according to claim 5, It is characterized in that After determining that the value of each of the key landscape indicators corresponding to the mutation point of each of the non-point source pollution indicators is the mutation threshold of each of the key landscape indicators, the method further includes: Extracting multiple sub-sample sets with the same number of samples from the original sample data set of each non-point source pollution indicator; For each of the sub-sample sets, according to the non-point source pollution index and each of the corresponding key landscape indicators, a non-parametric deviation reduction algorithm and a bootstrap method are combined to obtain the mutation threshold of each of the key landscape indicators corresponding to each of the non-point source pollution indicators; For each of the key landscape indicators corresponding to each of the non-point source pollution indicators, the mutation thresholds of the same key landscape indicator of all the sub-sample sets are sorted to obtain the mutation threshold range of each of the key landscape indicators corresponding to each of the non-point source pollution indicators.

7. A watershed non-point source pollution prevention and control system based on key landscape indicators based on the watershed non-point source pollution prevention and control method based on key landscape indicators according to any one of claims 1 to 6, It is characterized in that The system comprises: The non-point source pollution index collection module is used to collect various non-point source pollution indicators at each sampling point in the target watershed; A landscape index acquisition module is used to acquire a plurality of landscape indexes at each sampling point in the target watershed, which are recorded as a landscape index set; each non-point source pollution index of all the sampling points corresponds to the landscape index set; A key landscape index screening module is used to input the same non-point source pollution index of all sampling points and the corresponding landscape index set into the random forest model for non-point source pollution indexes whose index values ​​exceed preset values, and screen out the key landscape index corresponding to each non-point source pollution index in combination with a significance test method; A key landscape indicator threshold calculation module, used for applying a non-parametric deviation reduction algorithm and a bootstrap method to each of the non-point source pollution indicators and the corresponding key landscape indicators to obtain a mutation threshold of the key landscape indicator; The landscape pattern optimization module adjusts the landscape pattern in the target watershed according to the mutation threshold data of each key landscape indicator corresponding to each non-point source pollution indicator and the response relationship between each non-point source pollution indicator and each corresponding key landscape indicator, so as to achieve the prevention and control of non-point source pollution.

Images