An ecological monitoring network point position evaluation method and system based on gradient coverage analysis and machine learning enhancement

Abstract

The application discloses a kind of based on gradient coverage analysis and machine learning enhanced ecological monitoring network point position evaluation method and system.Method includes: step 1: obtaining the total gradient range of each environmental variable in ecological research area;Step 2: import historical point position data set, and calculate the covered interval and proportion of each historical point position to each environmental variable, and the calculation result is calibrated using coverage boundary refining module;Step 3: based on the corrected calculation result, identify the gradient interval not covered on each environmental variable;Step 4: set new point position in the gradient interval not covered, based on high-dimensional joint coverage evaluation module, evaluate the filling capacity of new point position to the gradient interval not covered, obtain evaluation result;Step 5: based on the evaluation result, set machine learning prediction module to dynamically adjust the filling demand of the gradient interval not covered;Step 6: according to the filling demand, output new point position priority ranking and recommendation scheme.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring network deployment technology, and more specifically to a method and system for evaluating ecological monitoring network sites based on gradient coverage analysis and machine learning enhancement. Background Technology

[0002] In the actual deployment of ecological monitoring networks, the current mainstream approach is as follows: spatially uniform distribution: random sampling is carried out in the study area according to grids or stratification to ensure that the points are roughly evenly distributed; habitat type representativeness judgment: based on field experience, it is ensured that the selected points can cover the main habitat types in the study area (such as tidal flats, salt marshes, reed beds, etc.).

[0003] Furthermore, some studies employ environmental stratification or spatial optimization algorithms based on protection priority zones (such as Marxan), but their core remains focused on discrete partitioning or spatial distance as the primary optimization objective, failing to systematically address the coverage quality problem of continuous environmental gradients. The aforementioned methods have the following shortcomings: (1) The coverage of historical data points to continuous environmental gradients was not systematically quantified. Current practices focus on whether the monitoring points are distributed across different habitat types, but do not quantify the range of values ​​for each continuous environmental variable (such as soil salinity, soil bulk density, annual precipitation, elevation, etc.) across the entire study area, or the percentage of the range already covered by historical monitoring points.

[0004] (2) The boundary of the point coverage is too crude.

[0005] When a historical data point has a value of 3‰ to 4‰ for a certain environmental variable, the traditional approach usually assumes that this data point represents a fixed radius interval centered on its value (e.g., 3‰ ± 0.5‰) or directly uses the data point value as a substitute. However, in reality, due to environmental heterogeneity, a data point may only represent a part of this interval (e.g., 3.2‰ to 3.8‰), and the remaining part still needs to be covered by other data points. Existing methods lack precise quantitative means for this "coverage boundary".

[0006] (3) Each environmental variable was considered in isolation, and a high-dimensional joint coverage assessment was not conducted.

[0007] Current practices often involve calculating the coverage of each environmental variable separately (e.g., 40% salinity coverage, 50% soil bulk density coverage), and then simply weighting or presenting the data side-by-side. However, in reality, when a location simultaneously covers the blank areas of multiple environmental variables, its overall value is far greater than that of a location covering only the blank area of ​​a single variable. Existing methods cannot evaluate the joint coverage value of locations in high-dimensional environmental gradient spaces.

[0008] (4) Lack of a predictive mechanism for the decay of the representativeness of historical data points over time.

[0009] Once established, historical data points are generally considered to permanently represent the environmental gradient range in which they are located. However, in reality, environmental conditions (such as water level, vegetation cover, and intensity of human activity) may change over time, causing the original representativeness of the data points to decline. Existing methods lack the ability to predict this decline and cannot provide early warnings of which historical data points need to be supplemented or replaced.

[0010] Therefore, providing a monitoring point optimization method with historical point gradient coverage filling as the core and machine learning as an enhancement module is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0011] In view of this, the present invention provides a method and system for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement, which integrates gradient cover quantitative analysis and three types of machine learning enhancement modules to effectively overcome the shortcomings of traditional site selection methods.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: An ecological monitoring network site evaluation method based on gradient cover analysis and machine learning enhancement includes: Step 1: Obtain the total gradient range of each environmental variable in the ecological study area; Step 2: Import the historical data set, calculate the coverage range and percentage of each historical data point for each environmental variable, and use the coverage boundary refinement module to perform boundary calibration on the calculation results; Step 3: Identify the gradient intervals that were not covered on each historical environmental variable based on the corrected calculation results; Step 4: Set up new points in the uncovered gradient interval, evaluate the ability of the new points to fill the uncovered gradient interval based on the high-dimensional joint coverage evaluation module, and obtain the evaluation results; Step 5: Based on the evaluation results, the machine learning prediction module dynamically adjusts the filling requirements for uncovered gradient intervals; Step 6: Output the priority ranking and recommended solutions for the new locations based on the filling requirements.

[0013] Preferably, step 2 refines the coverage boundary using a coverage boundary refinement module, specifically including: Using a single historical point as the center, multiple auxiliary sample points within a preset radius are extracted; the kernel density calculation method is used to calculate the local density function of environmental variables, and a density threshold is set; the preset radius can be adaptively adjusted according to the spatial autocorrelation scale of the corresponding environmental variables.

[0014] Preferably, in step 4, the high-dimensional joint coverage assessment module constructs a multi-dimensional environmental gradient space, calculates high-dimensional volume coverage, and obtains the assessment capability for filling newly added points based on the high-dimensional volume coverage results. Specifically, this includes: Construct an m-dimensional environmental gradient space, where each newly added point corresponds to a point in the space, representing an m-dimensional hyperrectangle corresponding to the interval; The covered volume of historical points in high-dimensional space The calculation formula is as follows: ; in, V history This represents the union volume of all historical monitoring points in the high-dimensional environmental gradient space; n This indicates the total number of existing historical monitoring points; m This represents the total number of dimensions of the environmental gradient; I ij Indicates the first i The historical point at the j Representative intervals in each environmental dimension; Indicates the first i A hyperrectangular region formed by historical points in high-dimensional space; ∪ represents the union operator of sets; The formula for calculating the total volume of the ecological research area in high-dimensional space is as follows: ; in, V total : Represents the theoretical total volume of the entire ecological research area in high-dimensional space; L j Indicates the first j The total gradient length of each environmental dimension; The incremental formula after adding the new candidate points to the high-dimensional coverage volume is as follows: ; in, To fill the gaps in capabilities for high-dimensional collaboration; Q t Indicates the number currently to be evaluated t One new candidate site; V history : ∪ indicates that the first ∪ will be the ∪th ... t After a candidate point is added, the volume of the union of all historical points and the current point.

[0015] Preferably, the machine learning prediction module in step 5 dynamically adjusts the filling requirement for uncovered gradient intervals by training a decay prediction model using time-series environmental data, specifically including: Collect time series data of environmental variables from historical locations over many years, and calculate the magnitude of change over time for each historical location and environmental variable; Using the initial environmental characteristics, topographic features, and climate zone characteristics of historical locations as input features, and the magnitude of change over time as the target variable, a regression model is trained. After training, a training decay prediction model is obtained. For any historical point, the training decay prediction model can predict the change range of each environmental variable in the next few years and calculate the degree of shrinkage of the representative interval to obtain the size of the newly added gradient blank interval.

[0016] Preferably, the calculation formula for the priority sorting of newly added points in step 6 is as follows: ; in, To improve univariate imputation capabilities, To fill the gaps in capabilities for high-dimensional collaboration, To compensate for attenuation, λ1, λ2, and λ3 are weighting coefficients that satisfy λ1 + λ2 + λ3 = 1. The final output is calculated according to P. t A list of recommended new locations, sorted in descending order.

[0017] Preferably, an ecological monitoring network site evaluation system based on gradient cover analysis and machine learning enhancement includes: a user interface layer, a business logic layer, and a data storage layer, with each layer interacting with each other through a standardized data interface; The user interface layer receives user input parameters, ecological research area and site data, and visualizes the site priority ranking and high-dimensional gradient space coverage results. The business logic layer includes the gradient total range calculation module, historical point coverage analysis module, uncovered interval identification module, new point filling capability assessment module, representative decay prediction module, and comprehensive priority ranking recommendation module, which are called in sequence. The data storage layer is used to store the environmental gradient base database, historical point database, auxiliary sample database, and model parameter library.

[0018] Preferably, the historical point coverage analysis module has a built-in first machine learning enhancement unit for performing coverage boundary refinement based on kernel density estimation; The newly added point filling capability assessment module has a built-in second machine learning enhancement unit, which is used to perform joint coverage assessment of high-dimensional environmental gradient space; The representative decay prediction module has a built-in third machine learning enhancement unit, which is used to perform representative time-series decay prediction of historical points.

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method and system for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement, with the following beneficial effects: 1. Quantitative Gradient Coverage: The present invention quantifies the coverage of historical data points on multidimensional continuous environmental gradients, upgrading the deployment of monitoring networks from qualitative experience to quantitative analysis, filling the technical gap in the field of quantitative assessment methods for gradient coverage. 2. Precise coverage boundary: By using the kernel density estimation method, the representative interval of each point is adaptively determined, avoiding the coarse processing of fixed radius or point value substitution in traditional methods. The width of the representative interval is dynamically adjusted according to the heterogeneity of the local environment, and the accuracy is significantly better than the fixed radius method. 3. High-dimensional joint evaluation: Evaluating the joint filling value of points in a high-dimensional environmental gradient space is superior to the approach of calculating the coverage of each variable separately and then simply weighting them. It can identify points that do not perform well in a single variable dimension but have high multi-dimensional joint coverage value. 4. Dynamic adaptability: Through the attenuation prediction model, the monitoring network has the ability to dynamically adjust, which can provide early warning of the decline in the representativeness of historical points and arrange for replenishment, reducing the risk of decline in monitoring data quality due to environmental changes; 5. Transparent Decision-Making: The three components of the comprehensive priority index (univariate imputation capability, high-dimensional joint imputation capability, and attenuation compensation requirement) all have clear physical meanings. Decision-makers can understand why each recommended point was selected, avoiding the "black box" decision-making defects of traditional machine learning methods. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 The overall flowchart provided for this invention.

[0022] Figure 2 This is a schematic diagram of the coverage boundary refinement provided by the present invention.

[0023] Figure 3 This is a schematic diagram of the high-dimensional environmental gradient space coverage volume provided by the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention discloses a method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement, comprising: Step 1: Obtain the total gradient range of each environmental variable in the ecological study area; Step 2: Import the historical data set, calculate the coverage range and percentage of each historical data point for each environmental variable, and use the coverage boundary refinement module to perform boundary calibration on the calculation results; Step 3: Identify the gradient intervals that were not covered on each historical environmental variable based on the corrected calculation results; Step 4: Set up new points in the uncovered gradient interval, evaluate the ability of the new points to fill the uncovered gradient interval based on the high-dimensional joint coverage evaluation module, and obtain the evaluation results; Step 5: Based on the evaluation results, the machine learning prediction module dynamically adjusts the filling requirements for uncovered gradient intervals; Step 6: Output the priority ranking and recommended solutions for the new locations based on the filling requirements.

[0026] In a specific embodiment of the present invention, such as Figure 1 As shown, it includes: Determine the total gradient range of each environmental variable in the study area.

[0027] Calculate the coverage range and percentage of historical data points for each environmental variable, and use the machine learning enhancement module ① (coverage boundary refinement module) to refine the coverage boundary and perform boundary calibration on the results.

[0028] Based on the above calculation results, the gradient intervals that were not historically covered for each environmental variable are identified.

[0029] The ability of newly added points to fill in the uncovered areas is evaluated. In this process, machine learning enhancement module ② (high-dimensional joint coverage evaluation module) is introduced to improve the evaluation accuracy.

[0030] Considering the representative decay of historical data points, the representative decay prediction of the machine learning enhancement module ③ (machine learning prediction module) is used to dynamically adjust and fill the demand.

[0031] Output the priority ranking and recommended solution for newly added points.

[0032] The above steps, executed in sequence, constitute a complete incremental optimization decision-making process for the monitoring network. Once the new monitoring point is deployed, it is automatically transferred to the historical monitoring point database to support the next round of optimization iterations.

[0033] Furthermore, let there be m continuous environmental variables within the study area, denoted as... V 1, V 2,..., V m For the j-th environment variable V j Its value range within the study area is:

[0034] The total gradient length is:

[0035] For example, for soil salinity, the minimum value in the study area is 1‰ and the maximum value is 10‰, then L salinity = 9‰.

[0036] The determination of the total gradient range depends on the spatial continuous distribution data of various environmental variables within the study area, which is usually stored in the environmental gradient base database in the form of raster layers or vector surface data.

[0037] Specifically, step 2 involves refining the coverage boundary using the coverage boundary refinement module, which includes: Using a single historical point as the center, multiple auxiliary sample points within a preset radius are extracted; the kernel density calculation method is used to calculate the local density function of environmental variables, and a density threshold is set; the preset radius can be adaptively adjusted according to the spatial autocorrelation scale of the corresponding environmental variables.

[0038] In a specific embodiment of the present invention, the coverage range and percentage of historical data points for each environmental variable are calculated.

[0039] There are n historical points. P 1, P 2,..., P n For the i-th historical point P i Its value for the j-th environment variable is v. ij .

[0040] Traditional methods directly v ij Treating a point as a single value means that the point represents only that single value. However, a more reasonable approach is to consider that each point represents a range within its local environment.

[0041] Suppose that after machine learning enhancement, point P i The representative interval for the j-th environment variable is:

[0042] in, delta ij - and delta ij + The point is located at V. j The left and right directions represent the radius, and their values ​​are adaptively determined by the kernel density estimation method, rather than being preset fixed values.

[0043] The covered intervals of all historical data points on the j-th environment variable are the union of the intervals represented by each data point:

[0044] The coverage percentage of historical data points on the j-th environmental variable is:

[0045] Among them, |C j | represents the total length of the covered interval.

[0046] Furthermore, the process of determining the range of the interval includes: How large is the representative range of a historical data point on its corresponding environmental variables? Traditional methods, such as simply treating the data point as a single point or arbitrarily defining a fixed radius, are not precise enough.

[0047] The machine learning method used in this invention is an adaptive interval estimation method based on local environmental heterogeneity.

[0048] For historical point P i Multiple auxiliary sample points were collected within a certain spatial range (e.g., a radius of 500 meters) around the central point. These auxiliary sample points were used to collect environmental variable V. j The values ​​taken on the vector form a distribution. Using the kernel density estimation (KDE) method, the density function f of this distribution is estimated. j (x).

[0049] Point P i In V j The representative interval on is defined as:

[0050] Where θ is the density threshold, which can be set according to the research needs (e.g., 10% of the density peak value).

[0051] The advantages of this method are: ① The width of the representative interval is automatically determined by the heterogeneity of the local environment: the interval is wide in homogeneous environments and narrow in heterogeneous environments. ② There is no need to manually set a fixed radius, adapting to the environmental characteristics of different regions.

[0052] Furthermore, the nuclear density estimation formula is calculated as follows: Let historical point P be... i Within a radius of r (e.g., r = 500 m), N auxiliary sample points were collected in the environment variable V. j The values ​​taken on the sample form the sample set:

[0053] The probability density function given by kernel density estimation is: .

[0054] The meanings of the symbols in the formula are shown in Table 1: Table 1

[0055] Explanation of physical meaning: j The larger (x) is, the better the environment variable V. j The denser the auxiliary sample points appearing near value x, the higher the historical point P. i The more "typical" the microenvironment is near x, the more "typical" it is; conversely, a lower density indicates that the value deviates from point P. i The representative range.

[0056] Furthermore, the determination of the representative interval includes: Historical point P i The representative interval I on the j-th environment variable ij Defined as the set of values ​​whose density is higher than a threshold:

[0057] in: θ j The density threshold can be taken as... Typically, κ = 0.1 (i.e., the interval enclosed by the contour line with a density peak of 10%). like j If (x) exhibits a unimodal distribution, then I ij For containing v ij (Point P) i In V j A continuous interval [a] of observations) ij ,b ij ].

[0058] For example: Historical points P 1 For example, auxiliary sample values ​​for soil salinity (in ‰): Auxiliary samples: x={2.8,3.1,3.3,3.5,3.6,3.9,4.2} Calculated using KDE Salinity (x) reaches its maximum value of 0.38 at x=3.5, and the density drops to <0.038 at x=2.5 and x=4.5. If κ = 0.1, then θ_salinity = 0.038 satisfy The range of x for salinity (x) ≥ 0.038 is [2.5, 4.5]. Therefore, I1, salinity = [2.5, 4.5], meaning that this historical point is considered to represent the salinity range of 2.5‰ to 4.5‰. Compared to directly taking a single point value or a fixed radius (such as ±0.5‰), the KDE method makes the width of the representative interval adaptively change with the heterogeneity of the local environment—wider time intervals in homogeneous environments and narrower time intervals in heterogeneous environments.

[0059] In another specific embodiment of the present invention, identifying gradient intervals not covered by historical data includes: the interval on the j-th environmental variable that is not covered by any historical data point is:

[0060] G j This refers to the gradient gaps that need to be filled first when new points are added. If G j If the set is empty, it means that the gradient of the j-th environmental variable has been completely covered by historical points.

[0061] Specifically, in step 4, the high-dimensional joint coverage assessment module constructs a multi-dimensional environmental gradient space, calculates high-dimensional volume coverage, and obtains the assessment capability for filling newly added points based on the high-dimensional volume coverage results. This includes: Construct an m-dimensional environmental gradient space, where each newly added point corresponds to a point in the space, representing an m-dimensional hyperrectangle corresponding to the interval; The covered volume of historical points in high-dimensional space The calculation formula is as follows: ; The formula for calculating the total volume of the ecological research area in high-dimensional space is as follows: ; The incremental formula after adding the new candidate points to the high-dimensional coverage volume is as follows: ; in, To fill the gaps in high-dimensional joint capabilities.

[0062] In a specific embodiment of the present invention, there are k candidate new locations. Q 1, Q 2,..., Q k For candidate locations Q t Its value for the j-th environment variable is u tj The represented interval is J tj =[ u tj - epsilon tj - , u tj + epsilon tj + ].

[0063] The length of the uncovered interval that the candidate point can fill on the j-th environmental variable is:

[0064] The univariate imputation capability of this candidate site is:

[0065] Among them, w j The weight of the j-th environmental variable can be set according to the research objective (e.g., equal weighting or weighting based on ecological significance). When |G j When |=0, this item is defined as 0, meaning that when the variable has been completely covered, there is no longer a need to fill it.

[0066] Furthermore, a new data point may fill gaps in multiple environmental variables simultaneously. How can we comprehensively evaluate its multidimensional filling value?

[0067] like Figure 3 As shown, the machine learning method used in this invention is based on a multidimensional environmental gradient space to calculate high-dimensional volume coverage.

[0068] We construct an m-dimensional environmental gradient space using m environmental variables as coordinate axes. Each point in this space corresponds to a point, and its represented interval corresponds to an m-dimensional hyperrectangle.

[0069] The covered volume of historical points in high-dimensional space is: ; in, V historyThis represents the union volume (i.e., the covered volume) of all historical monitoring points in the high-dimensional environmental gradient space; n represents the total number of existing historical monitoring points; m represents the total number of dimensions of the environmental gradient (e.g., salinity, temperature, and depth, a total of 3 dimensions). I ij It represents the representative interval of the i-th historical point in the j-th environmental dimension (i.e., the range covered by the point in this dimension). The ∪ symbol represents the hyperrectangular region formed by the i-th historical point in high-dimensional space; ∪ represents the union operator, used to merge the hyperrectangular regions and eliminate overlaps. The total volume of the study area in high-dimensional space is: ; in, V total : Represents the theoretical total volume of the entire ecological research area in high-dimensional space; L j This represents the total gradient length of the j-th environmental dimension (i.e., the difference between the maximum and minimum values ​​of this dimension within the study area). Added candidate point Q t The high-dimensional filling capability is defined as: the increment of the high-dimensional coverage volume after adding the point:

[0070] in, This refers to the ability to fill gaps in higher dimensions (i.e., net volume increment). Q t This represents the t-th newly added candidate point to be evaluated. V history : ∪ represents the volume of the union of all historical points and the current point after adding the t-th candidate point.

[0071] This indicator intuitively reflects the extent to which the candidate point can fill the gaps in the high-dimensional environmental gradient space of historical points.

[0072] Because high-dimensional volume computation suffers from the "curse of dimensionality"—that is, as the dimension m increases, the number of sampling points required per unit volume grows exponentially, and direct high-dimensional numerical integration will lead to exhaustion of computational resources—this study adopts Monte Carlo sampling approximation or principal component analysis for dimensionality reduction in practical calculations to achieve a balance between computational accuracy and computational efficiency.

[0073] (1) Construction of high-dimensional environmental gradient space

[0074] There are m continuous environment variables. V 1 , V 2 ,…,V m Each variable V j The total area within the study region is:

[0075] Using the ranges of these m variables as coordinate axes, we span an m-dimensional space called the environment gradient space. The total volume of this space is the product of the lengths of each axis:

[0076] For example: with only two variables—salinity (range 1‰~10‰, length 9) and soil bulk density (range 1.0~1.8g / cm³, length 0.8)—the total area of ​​the two-dimensional space is 9×0.8=7.2 (unit: ‰·g / cm³).

[0077] (2) The representative region of a single point in high-dimensional space

[0078] Let P be the i-th historical point. i In the j-th environment variable V j The representative interval on is (obtained by the kernel density estimation method in §4.3.1):

[0079] Where a ij and b ij The point is located at V. j The upper and lower limits are represented by the symbol above.

[0080] So, point P i The region occupied in the m-dimensional environment gradient space is an m-dimensional hyperrectangle, defined as the Cartesian product of the intervals represented by each dimension:

[0081] Where □ represents a hyperrectangle, the volume of which is the product of the lengths of each dimension:

[0082] Example: Location P 1 The representative range for salinity is [2.5, 4.5], and the representative range for bulk density is [1.2, 1.4]. Therefore, it occupies a rectangular area of ​​2.0‰ × 0.2 g / cm³ = 0.40 (‰·g / cm³) in two-dimensional space.

[0083] (3) The combined coverage area of ​​the historical location set

[0084] The regions occupied by all n historical points in the environmental gradient space are the union of the hyperrectangles of each point:

[0085] The volume of this union is denoted as V. history =Vol(Ω history ).

[0086] Note: When hyperrectangles at different locations overlap, the overlapping portion is only counted once in the volume calculation. Therefore, V history It is generally less than the sum of the volumes of all points.

[0087] (4) Calculation method of high-dimensional union volume

[0088] Since there is no simple closed-form expression for calculating the volume of the union of hyperrectangles (especially when m≥3 and there is a large amount of overlap), this invention uses Monte Carlo integration for approximate calculation: Step ①: Within the total range of the environmental gradient space Within the sample, randomly and uniformly sample Nsample points (Nsample ≥ 10 is recommended). 5 The coordinates of each sampling point are:

[0089] Step 2: Determine whether each sampling point falls within the hyperrectangle of at least one historical point:

[0090] Step 3: The estimated value of the combined coverage volume of historical data points is:

[0091] That is: the proportion of covered sampling points multiplied by the total volume.

[0092] (5) High-dimensional filling capability of newly added points

[0093] For candidate new locations Q t Its representative interval for the j-th environmental variable is (also obtained by kernel density estimation):

[0094] Its hyperrectangle in m-dimensional space is:

[0095] Will Q t After being added, the combined coverage area of ​​the historical and new locations is:

[0096] Its volume is also calculated using Monte Carlo integration:

[0097] in:

[0098] Candidate points Q t The high-dimensional filling capability is defined as the volume increment before and after the addition:

[0099] This indicator intuitively reflects Q t How much "environmental gradient spatial coverage area" (in two dimensions) or "volume" (in three dimensions and above) can be added to the existing monitoring network?

[0100] Assuming the study area has only two environmental variables: salinity (1‰~10‰) and soil bulk density (1.0~1.8 g / cm³), and two historical sites and one candidate site, as shown in Table 2.

[0101] Table 2

[0102] Total volume: V total =(10-1)×(1.8-1.0)=9×0.8=7.2

[0103] The rectangles of P1 and P2 do not overlap in the salinity dimension (4.5 < 6.0), therefore they are not overlapping in two dimensions. The historical joint cover volume V history =0.40 + 0.40 = 0.80

[0104] The overlap between Q1 and P1 is as follows: salinity overlaps [4.0, 4.5], bulk density overlaps [1.2, 1.4], and the 2D overlap area is 0.5 × 0.2 = 0.10.

[0105] The overlap between Q1 and P2 is as follows: salinity overlaps [6.0, 6.5], bulk density overlaps [1.3, 1.5], and the two-dimensional overlap area is 0.5 × 0.2 = 0.10.

[0106] The area of ​​the union of sets = 0.40 + 0.40 + 1.25 - 0.10 - 0.10 = 1.85

[0107] High-dimensional filling capability S1 ( ² ) =1.85 - 0.80 = 1.05

[0108] This means that although Q1 itself has an area of ​​1.25, due to the overlap of 0.20 with historical locations, the net newly added coverage area is 1.05.

[0109] Specifically, the machine learning prediction module in step 5 dynamically adjusts the filling requirement for uncovered gradient intervals by training a decay prediction model using time-series environmental data, specifically including: Collect time-series data of environmental variables from historical sites over many years. The time-series environmental data comes from on-site observation records and / or satellite remote sensing inversion data of historical sites. Calculate the magnitude of change over time for each historical site and environmental variable. Using the initial environmental characteristics, topographic features, and climate zone characteristics of historical locations as input features, and the magnitude of change over time as the target variable, a regression model is trained. After training, a training decay prediction model is obtained. For any historical point, the training decay prediction model can predict the change range of each environmental variable in the next few years and calculate the degree of shrinkage of the representative interval to obtain the size of the newly added gradient blank interval.

[0110] In a specific embodiment of the present invention, the historical point P i The representativeness of the j-th environmental variable is not constant. Let the time factor t represent the number of years since the location was established; then its representative interval may shrink over time.

[0111] Where, α ij and β ij The attenuation coefficient can be learned from historical environmental change data through machine learning models.

[0112] Attenuated historical coverage interval C j (t) will be smaller than the initial coverage interval C j (0), thus creating a new gradient blank interval, which requires new points to fill.

[0113] Furthermore, does the environmental representativeness of historical data points decay over time? How fast does it decay? When is it necessary to supplement or replace them?

[0114] The machine learning method used in this embodiment utilizes time-series environmental data to train a decay prediction model.

[0115] Collect time-series data of environmental variables (such as annual NDVI, annual groundwater level, annual soil salinity, etc.) for historical sites over many years. For each historical site P i and environment variable V j Calculate its variation over time:

[0116] Where v ij (0) represents the environmental variable value of this location in the initial year of deployment, v ij(t) represents the observation value in year t after deployment.

[0117] The initial environmental characteristics of the location, the topographic features of its location, and the characteristics of its climate zone are used as input features, with Δv as the input feature. ij (t) is used as the target variable to train a regression model (such as XGBoost or Random Forest).

[0118] After training, for any historical point, the model can predict the magnitude of changes in various environmental variables over the next few years, thereby calculating the degree of contraction of its representative interval and the size of the resulting new gradient blank interval.

[0119] The output of this module can be used for: Early warning of which historical data points need to be supplemented or replaced The priority of newly added points is dynamically adjusted to prioritize filling the blank areas caused by attenuation.

[0120] Specifically, the calculation formula for the priority ranking of newly added points in step 6 is as follows: ; in, To improve univariate imputation capabilities, To fill the gaps in capabilities for high-dimensional collaboration, To compensate for attenuation, λ1, λ2, and λ3 are weighting coefficients that satisfy λ1 + λ2 + λ3 = 1. The final output is calculated according to P. t A list of recommended new locations, sorted in descending order.

[0121] Specifically, an ecological monitoring network site evaluation system based on gradient coverage analysis and machine learning enhancement includes: a user interface layer, a business logic layer, and a data storage layer, with data interaction between each layer through standardized data interfaces; The user interface layer receives user input parameters, ecological research area and site data, and visualizes the site priority ranking and high-dimensional gradient space coverage results. The business logic layer includes the gradient total range calculation module, historical point coverage analysis module, uncovered interval identification module, new point filling capability assessment module, representative decay prediction module, and comprehensive priority ranking recommendation module, which are called in sequence. The data storage layer is used to store the environmental gradient base database, historical point database, auxiliary sample database, and model parameter library.

[0122] Specifically, the historical point coverage analysis module has a built-in first machine learning enhancement unit, which is used to perform coverage boundary refinement based on kernel density estimation; The newly added point filling capability assessment module has a built-in second machine learning enhancement unit, which is used to perform joint coverage assessment of high-dimensional environmental gradient space; The representative decay prediction module has a built-in third machine learning enhancement unit, which is used to perform representative time-series decay prediction of historical points.

[0123] The specific implementation method is as follows: Collect the following data within the study area: For raster layers of various environmental variables (such as soil salinity, soil bulk density, annual precipitation, elevation, human footprint index, NDVI seasonal values, etc.), a spatial resolution of no less than 30m×30m is recommended to ensure the density of auxiliary sample collection. Location and year of establishment of historical monitoring points; Location of candidate new points; Environmental data of auxiliary sample points around historical locations (used for coverage boundary refinement). Time series data of environmental variables from historical locations over many years (for decay prediction).

[0124] like Figure 2 As shown, the implementation of coverage boundary refinement includes: Historical point P i For example: 1. Using P i Centered on the grid, extract the environmental variable values ​​of all raster pixels within a 500-meter radius; 2. Regarding the environmental variable V j Kernel density estimation is performed using a Gaussian kernel function; 3. Determine the density threshold θ (e.g., take 10% of the density peak value); 4. The interval where the density is greater than θ is taken as P. i In V j The representative interval I on ij .

[0125] To accommodate the spatial heterogeneity scale differences of different environmental variables, the radius r can be adjusted according to the spatial autocorrelation range of the variable: for variables with drastic spatial variation (such as soil salinity), r can be taken as 200~300 m; for variables with moderate spatial variation (such as annual precipitation), r can be taken as 500~1000 m.

[0126] The implementation of the high-dimensional joint coverage assessment includes: 1. Standardize all environmental variables to the [0,1] range to eliminate the influence of different units on distance / volume calculations; Preferably, before constructing the high-dimensional environmental gradient space, each environmental variable is first standardized to the [0,1] interval to eliminate the influence of dimensional differences on volume calculation. Standardization does not change the relative positions of each point in the environmental gradient space or the subsequent priority ranking results, but it ensures that variables with different dimensions are treated equally in the joint coverage assessment. Those skilled in the art may also choose not to perform standardization; in this case, the weights of each variable automatically include their dimensional information, and the two methods are mathematically equivalent.

[0127] 2. Perform principal component analysis on the standardized environmental variables, and take the first 3 principal components (cumulative variance contribution rate ≥ 85%) as the coordinates of the environmental gradient space after dimensionality reduction; the purpose of PCA dimensionality reduction is to project the original high-dimensional space (m is usually 5~20 dimensions) to a low-dimensional space (3 dimensions), so as to make volume calculation feasible while retaining the main environmental gradient information, and the 3-dimensional space is convenient for visualization and user understanding; 3. In 3D space, the Monte Carlo integration method is used (number of sampling points N). sample ≥10 5 Calculate the volume V covered by historical data points. history (i.e., the volume of the union of the hyperrectangles at each point); 4. Calculate the volume V after adding each candidate point. history ∪Q t 5. Calculate high-dimensional filling capability

[0128] The implementation of representative decay prediction includes: 1. Collect annual (or biennial) environmental variable observations from historical data points, with a recommended time span of at least 5 years to capture trends in environmental change; 2. Calculate the annual variation of each location for each environmental variable; 3. Using the geographic coordinates, topographic features, and climate zones of the points as input features, and the annual variation as the target variable, train an XGBoost regression model (the model hyperparameters can be determined through grid search or Bayesian optimization). 4. For each historical location, input its characteristics to predict the magnitude of environmental change over the next 5 years; 5. Based on the predicted change magnitude, calculate the shrinkage of the representative interval, and then calculate the additional gradient blank interval that needs to be filled by new points.

[0129] The implementation of comprehensive priority ranking includes: 1. For each candidate point, calculate S ( ¹ ) t , S (² ) t , D t ; 2. Set weighting coefficients lambda 1 = 0.4 lambda 2 = 0.4 lambda 3 = 0.2; the weighting coefficient can be adjusted according to different management objectives—for example, if the focus is on the overall coverage quality of the multi-dimensional environmental space, it can be increased. lambda 2; Focusing on gradient imputation for specific variables can improve... lambda 1; If the environment in the study area changes drastically, it can improve lambda 3; 3. Calculate the overall priority index P t ; 4. Press P t Sort the data in descending order and output the top N recommended locations.

[0130] Input salinity, bulk density, and NDVI data from 30 historical locations. Based on the calculations of this method, it is recommended to add a new location at the X and Y coordinates. This location increases the high-dimensional coverage volume from 0.8 to 1.05.

[0131] Taking a region in the Yellow River Delta as an example, input observational data on salinity, bulk density, and NDVI from 30 historical monitoring points, along with location and environmental data from 50 candidate new monitoring points. The calculations performed using this method are as follows: The joint cover volume V of historical points in the three-dimensional environmental gradient space history =0.80 (standardized units); Based on comprehensive priority ranking, the candidate site ranked first Q Located at 118°57′E, 37°45′N, its univariate completion capability S 1 ( ¹ ) =0.32, high-dimensional filling capability S 1 ( ² ) =1.05, attenuation compensation requirement D 1 = 0.18; Comprehensive Priority Index P 1 = 0.4 × 0.32 + 0.4 × 1.05 + 0.2 × 0.18 = 0.584; Will Q After the addition of 1, the high-dimensional coverage volume increased from 0.80 to 1.85 (normalized units), an improvement rate of 131%.

[0132] This embodiment verifies that the present invention can effectively identify candidate sites with high multidimensional filling value and significantly improve the coverage quality of the monitoring network over the environmental gradient space.

[0133] The present invention also provides comparative examples, as follows: Comparative Example 1: Traditional Spatial Uniform Point Distribution Method 1. Method Content Monitoring points are deployed within the study area using a regular grid (e.g., 5 km × 5 km) or stratified random sampling methods to ensure uniform distribution. This method does not distinguish the spatial distribution characteristics of environmental variables, but only uses geographical distance or area coverage as the optimization objective. After determining the monitoring points, a habitat type map of the study area is overlaid, and a subjective judgment is made as to whether the selected points cover the main habitat types (e.g., wetlands, woodlands, farmland, etc.). If a certain habitat type is not covered, 1-2 additional points are deployed within it.

[0134] The core logic of this method is that the more uniform the distribution of points, the stronger the representativeness of the study area.

[0135] 2. Defects

[0136] The continuity of environmental gradients is not considered: uniform sampling assumes that environmental variables are spatially random or uniformly distributed, but in reality, environmental variables such as soil salinity, elevation, and precipitation exhibit significant spatial autocorrelation and heterogeneity. Uniformly distributed sampling points may largely fall within the median range of environmental variable values, while extreme value regions at the extremes of the gradient (such as high salinity areas and low elevation areas) are often undersampled. Taking the Yellow River Delta as an example, uniform sampling coverage of the salinity gradient's extremes (e.g., 0-2‰ and 8-10‰) is only 12% and 8%, respectively, far lower than the 89% coverage of the middle range (e.g., 4-6‰). The habitat type method is too crude: habitat types are usually discrete classifications (taking coastal wetlands as an example, such as "tidal flats" and "salt marshes"), but the range of environmental variables within the same habitat type can be extremely wide. For example, the soil salinity in "tidal flats" may vary from 1‰ to 15‰, and setting up 1-2 points in this type cannot represent the diversity of its internal environmental gradients; Lack of coverage increment assessment of historical sites: When adding new sites, this method only considers the "blank area of ​​the study area" and does not consider the "blank interval of the environmental gradient", which leads to the new sites often being highly redundant with historical sites in terms of environmental characteristics and having low marginal contribution. Insufficient objectivity: The number and location of habitat type supplementation sites depend on expert experience and judgment. Different experts may give different solutions, and the decision-making process lacks repeatability and quantifiability.

[0137] Comparative Example 2: Univariate Stratified Sampling Method

[0138] 1. Method Content

[0139] Select one or two key environmental variables (usually elevation or soil type), divide their values ​​within the study area into several equally spaced intervals (e.g., divide elevation into low, medium, and high layers), and then deploy an equal number of monitoring points within each interval to ensure that each environmental variable interval has coverage. This method typically aims to achieve an X% coverage rate for each environmental variable, calculates the proportion of intervals where historical monitoring points for each variable have been covered, and uses the arithmetic mean or weighted average of the coverage rates of each variable as a comprehensive evaluation index for the monitoring network coverage quality.

[0140] 2. Defects

[0141] The method evaluates each variable in isolation, ignoring the combined coverage value: it calculates "40% salinity coverage" and "50% bulk density coverage" separately, then takes the average (40%+50%) / 2=45% as the comprehensive evaluation. However, when a site fills gaps in both salinity and bulk density, its value is far higher than a site that only fills a gap in a single variable. For example, the combined contribution of a candidate site in the salinity gap (contribution +10%) and the bulk density gap (contribution +10%) is not 20%, but should be 22% (the value of the intersection of the two is not counted twice). The univariate method cannot identify this combined value and will underestimate high-value sites. Overlapping gaps between variables are counted repeatedly: When the uncovered gaps of multiple environmental variables highly overlap geographically, the univariate method will repeatedly calculate the fill value of this area, resulting in the deployment of multiple points in the area and wasting resources. Fixed stratification intervals cannot be adaptively adjusted: equal-interval stratification ignores the actual distribution density of environmental variables within the study area. For example, elevation may be skewed within the study area (most areas are low elevation, only a small portion are high elevation). Equal-interval stratification can lead to denser samples in low-elevation areas and sparser samples in high-elevation areas, or unnecessary oversampling in sparse data intervals. Difficulty in Dimensional Expansion: When the number of environmental variables exceeds three, traditional stratified sampling faces the "combinatorial explosion" problem—if 10 variables are each divided into three layers, then 3¹ layers are needed. 0 =59049 combination units, it is practically impossible to place points in every combination, so this method fails in actual operation.

[0142] Comparative Example 3: Point Selection Method Based on Spatial Clustering or Environmental Variable Clustering

[0143] 1. Method Content

[0144] Raster data of various environmental variables within the study area were collected, treating all raster pixels as sample points, each with m environmental variable values. Clustering algorithms such as K-means, ISODATA, or Self-Organizing Map (SOM) were used to divide all raster pixels into K clusters, each representing an "environmental type." Then, 1-2 representative locations were selected from each cluster to deploy monitoring points, ensuring coverage for each "environmental type." This method has been adopted in some ecological monitoring network planning studies in recent years.

[0145] 2. Defects

[0146] Clustering results provide a discrete characterization of environmental types: clustering algorithms divide the continuous environmental gradient space into discrete "types," losing information about the continuity within and between types. For example, two pixels that are very close to each other in the environmental space may be assigned to different clusters due to the clustering threshold, while two pixels that are very far apart may be assigned to the same cluster, leading to unreasonable jumps or redundancies in the placement of points. Unable to identify blank sub-intervals within covered intervals: This method can only determine "whether there are points for a certain environment type", but cannot quantify "which specific intervals within that environment type have been covered and which are still blank". For example, if the salinity range within a cluster is 2-8‰, and historical points have already covered the 4-6‰ interval, this method will still consider the cluster "uncovered" and recommend adding new points, resulting in waste. There is no clear priority ranking mechanism for the selected points: Clustering point selection methods typically select points randomly within each cluster or based on the distance from the center, making it impossible to calculate a uniform priority index for all candidate points. When the budget is limited and only a subset of points can be selected, this method lacks a scientific basis for selection. The marginal contribution of newly added sites cannot be quantified: a candidate site may be located at the boundary of multiple clusters, making it difficult for clustering methods to assess the joint contribution of the site to multiple environmental types. The high-dimensional joint coverage assessment in this invention is designed to solve this problem. Clustering performance degrades in high-dimensional environmental spaces: As the dimension m of environmental variables increases, the distance between pixels in high-dimensional space tends to become uniform (another manifestation of the "curse of dimensionality"), the rationality and stability of clustering results decrease significantly, and the clustering schemes generated by different initial parameters vary greatly and lack robustness.

[0147] Table 3 shows a summary table comparing the effects of the comparative method. Table 3

[0148] Comparative Examples 1-3 represent the technological evolution paths in the field of ecological monitoring network deployment, from "spatial uniformity" to "univariate hierarchical" and then to "clustering and point selection." However, none of these three existing methods have solved the four core technical problems of continuous quantitative assessment of environmental gradient cover, adaptive definition of representative intervals, high-dimensional joint value quantification, and dynamic prediction in the time dimension. This invention, through a combination of steps S1-S6 and three machine learning enhancement modules, achieves significant performance breakthroughs in all four dimensions. Furthermore, there is a synergistic effect between the modules—coverage boundary refinement (module ①) provides accurate superrectangular boundaries for high-dimensional joint assessment (module ②), and attenuation prediction (module ③) dynamically adjusts the calculation benchmark for filling needs. The non-obviousness of this overall technical solution is far greater than the simple superposition of the modules, and this invention significantly outperforms the comparative methods in terms of the overall improvement of environmental cover quality.

[0149] In summary, the core innovation of this invention lies in: Framework Innovation: This is the first time a monitoring network optimization framework centered on "historical point gradient coverage filling" has been proposed. It uses "how many historical points have been covered, how many are still missing, and how many can be filled by new points" as its core logic, distinguishing it from existing methods that rely solely on machine learning scoring or clustering for point selection. This framework transforms monitoring network optimization from "selecting good points" to "filling gaps," representing a methodological innovation. Innovation in boundary refinement: An adaptive coverage boundary refinement method based on kernel density estimation is proposed, which solves the problem of coarse delineation of point representative intervals in traditional methods. Compared with the traditional fixed radius method, this method dynamically adjusts the width of the representative interval according to the heterogeneity of the local environment, avoiding the subjectivity of manually setting parameters; Innovation in high-dimensional assessment: A joint coverage assessment method for high-dimensional environmental gradient space was constructed, and two optional engineering solutions, Monte Carlo integration and PCA dimensionality reduction, were provided to address the "curse of dimensionality" problem, realizing the comprehensive quantification of the multidimensional filling value of candidate points; Dynamic Prediction Innovation: This module introduces a representative attenuation prediction model based on time-series environmental data, enabling the monitoring network to dynamically adapt. This module makes the optimization scheme of this invention applicable not only to one-time deployments but also to monitoring networks requiring long-term operation and maintenance.

[0150] The inventors intend to focus on protecting the combined scheme of the above four innovative points, namely the complete technical chain of "gradient coverage identification—gap filling evaluation—high-dimensional joint optimization—attenuation dynamic adjustment," rather than any single link within it. This is because there are synergistic effects between the links—for example, high-dimensional joint evaluation relies on the accurate representative interval provided by coverage boundary refinement, and attenuation prediction dynamically influences the priority of filling requirements—this coupling relationship between the technical links is a key feature that distinguishes this invention from existing technologies.

[0151] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0152] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement, characterized in that, include: Step 1: Obtain the total gradient range of each environmental variable in the ecological study area; Step 2: Import the historical data set, calculate the coverage range and percentage of each historical data point for each environmental variable, and use the coverage boundary refinement module to perform boundary calibration on the calculation results; Step 3: Identify the gradient intervals that were not covered on each historical environmental variable based on the corrected calculation results; Step 4: Set up new points in the uncovered gradient interval, evaluate the ability of the new points to fill the uncovered gradient interval based on the high-dimensional joint coverage evaluation module, and obtain the evaluation results; Step 5: Based on the evaluation results, the machine learning prediction module dynamically adjusts the filling requirements for uncovered gradient intervals; Step 6: Output the priority ranking and recommended solutions for the new locations based on the filling requirements.

2. The method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement as described in claim 1, characterized in that, Step 2 involves refining the coverage boundary using the coverage boundary refinement module, specifically including: Using a single historical point as the center, multiple auxiliary sample points within a preset radius are extracted; the kernel density calculation method is used to calculate the local density function of environmental variables, and a density threshold is set; the preset radius can be adaptively adjusted according to the spatial autocorrelation scale of the corresponding environmental variables.

3. The method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement as described in claim 1, characterized in that, In step 4, the high-dimensional joint coverage assessment module constructs a multi-dimensional environmental gradient space, calculates high-dimensional volume coverage, and obtains the assessment capability for filling newly added points based on the high-dimensional volume coverage results. Specifically, this includes: Construct an m-dimensional environmental gradient space, where each newly added point corresponds to a point in the space, representing an m-dimensional hyperrectangle corresponding to the interval; The covered volume of historical points in high-dimensional space The calculation formula is as follows: ； in, V history This represents the union volume of all historical monitoring points in the high-dimensional environmental gradient space; n This indicates the total number of existing historical monitoring points; m This represents the total number of dimensions of the environmental gradient; I ij Indicates the first i The historical point at the j Representative intervals in each environmental dimension; Indicates the first i A hyperrectangular region formed by historical points in high-dimensional space; ∪ represents the union operator of sets; The formula for calculating the total volume of the ecological research area in high-dimensional space is as follows: ； in, V total : Represents the theoretical total volume of the entire ecological research area in high-dimensional space; L j Indicates the first j The total gradient length of each environmental dimension; The incremental formula after adding the new candidate points to the high-dimensional coverage volume is as follows: in, To fill the gaps in capabilities for high-dimensional collaboration; Q t Indicates the number currently to be evaluated t One new candidate site; V history : ∪ indicates that the first ∪ will be the ∪th ... t After a candidate point is added, the volume of the union of all historical points and the current point.

4. The method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement as described in claim 1, characterized in that, The machine learning prediction module in step 5 dynamically adjusts the filling requirement for uncovered gradient intervals by training a decay prediction model using time-series environmental data. Specifically, this includes: Collect time series data of environmental variables from historical locations over many years, and calculate the magnitude of change over time for each historical location and environmental variable; Using the initial environmental characteristics, topographic features, and climate zone characteristics of historical locations as input features, and the magnitude of change over time as the target variable, a regression model is trained. After training, a training decay prediction model is obtained. For any historical point, the training decay prediction model can predict the change range of each environmental variable in the next few years and calculate the degree of shrinkage of the representative interval to obtain the size of the newly added gradient blank interval.

5. The method for evaluating ecological monitoring network sites based on gradient cover analysis and machine learning enhancement as described in claim 1, characterized in that, The formula for calculating the priority ranking of newly added points in step 6 is as follows: ； in, To improve univariate imputation capabilities, To fill the gaps in capabilities for high-dimensional collaboration, To compensate for attenuation, λ1, λ2, and λ3 are weighting coefficients that satisfy λ1 + λ2 + λ3 = 1. The final output is calculated according to P. t A list of recommended new locations, sorted in descending order.

6. A site evaluation system for an ecological monitoring network based on gradient cover analysis and machine learning enhancement, characterized in that, include: The user interface layer, business logic layer, and data storage layer interact with each other through standardized data interfaces. The user interface layer receives user input parameters, ecological research area and site data, and visualizes the site priority ranking and high-dimensional gradient space coverage results. The business logic layer includes the gradient total range calculation module, historical point coverage analysis module, uncovered interval identification module, new point filling capability assessment module, representative decay prediction module, and comprehensive priority ranking recommendation module, which are called in sequence. The data storage layer is used to store the environmental gradient base database, historical point database, auxiliary sample database, and model parameter library.

7. The ecological monitoring network site evaluation system based on gradient cover analysis and machine learning enhancement as described in claim 6, characterized in that, The historical point coverage analysis module has a built-in first machine learning enhancement unit, which is used to perform coverage boundary refinement based on kernel density estimation; The newly added point filling capability assessment module has a built-in second machine learning enhancement unit, which is used to perform joint coverage assessment of high-dimensional environmental gradient space; The representative decay prediction module has a built-in third machine learning enhancement unit, which is used to perform representative time-series decay prediction of historical points.

Images