Soil available nutrient and organic carbon change prediction method, medium and system

Abstract

The invention provides a soil available nutrient and organic carbon change prediction method, medium and system, and belongs to the technical field of soil nutrient measuration.The method comprises the steps that firstly, a soil sample is collected in a preset soil monitoring area, the nutrient content is measured, and environmental parameter data is obtained; then constructing a time sequence feature matrix through empirical mode decomposition, and calculating a nutrient change contribution rate, a transfer index and a trend influence rate; establishing a prediction model by utilizing a soil nutrient judgment equation set module and an improved residual neural network structure; and finally, outputting a soil nutrient change prediction result and an early warning signal, thereby realizing accurate prediction of the soil nutrient change. According to the method, a time derivative item is innovatively introduced to describe dynamic characteristics of nutrient changes, multiple coupling items are established to represent interaction among different influence factors, and an error correction mechanism is designed to improve the stability of the model. The technical problem of low prediction precision of spatial-temporal dynamic change of soil available nutrients and organic carbon in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil nutrient measurement. Specifically, it relates to a method, medium and system for predicting the changes of soil available nutrients and organic carbon. Background Art

[0002] The contents of soil available nutrients and organic carbon are key indicators for evaluating soil fertility and quality, directly affecting crop growth and agricultural production efficiency. Existing technologies mainly use methods such as regular sampling and detection, remote sensing monitoring and statistical modeling for soil nutrient monitoring and prediction. Regular sampling and detection obtains soil nutrient content data through field sampling and laboratory analysis. Remote sensing monitoring uses multi-spectral or hyper-spectral remote sensing images to establish a relationship model between spectral characteristics and nutrient content. Statistical modeling constructs a regression model or machine learning model based on historical monitoring data.

[0003] However, these methods have obvious defects: Firstly, regular sampling and detection is limited by the sampling frequency and point density, and it is difficult to obtain continuous spatio-temporal sequence data. Secondly, remote sensing monitoring is easily affected by weather conditions and surface coverage, and can only monitor the surface soil. Thirdly, statistical modeling methods rely too much on historical data and do not fully consider the physical and chemical mechanisms of soil nutrient changes, and the generalization ability of the model is limited. Especially under complex terrain conditions, the spatial heterogeneity of soil nutrients is strong, affected by multiple factors such as terrain, hydrology and soil physical and chemical properties, and existing methods are difficult to accurately describe such complex spatio-temporal variation characteristics.

[0004] In summary, there is a technical problem in the prior art of low prediction accuracy for the spatio-temporal dynamic changes of soil available nutrients and organic carbon. Summary of the Invention

[0005] In view of this, the present invention provides a method, medium and system for predicting the changes of soil available nutrients and organic carbon, which can solve the technical problem of low prediction accuracy for the spatio-temporal dynamic changes of soil available nutrients and organic carbon in the prior art.

[0006] The present invention is implemented as follows: In a first aspect of the present invention, a method for predicting the changes in soil available nutrients and organic carbon includes the following steps: Collect soil samples in a preset soil monitoring area and measure the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content, obtain topographic elevation data, soil bulk density data, and soil precipitation data, establish a soil nutrient prediction model. The soil nutrient prediction model adopts an improved residual neural network structure, and a soil nutrient determination equation set module is added before the convolutional layer. The soil nutrient determination equation set module includes a soil nutrient contribution degree equation, a soil nutrient transfer degree equation, and a soil nutrient influence degree equation, calculate the soil nutrient aggregation index to obtain the soil nutrient spatial distribution characteristics, establish a soil nutrient change warning threshold according to the soil nutrient spatial distribution characteristics, and output a soil nutrient change warning signal.

[0007] Among them, for the step of collecting soil samples and measuring the soil nutrient content, specifically, the Kjeldahl method is used to measure the soil available nitrogen content, the sodium bicarbonate extraction molybdenum antimony anti-colorimetric method is used to measure the soil available phosphorus content, the ammonium acetate extraction flame photometry method is used to measure the soil available potassium content, and the potassium dichromate volumetric method is used to measure the soil organic carbon content.

[0008] Among them, for the step of data processing of the measured values of soil nutrient content, specifically, the empirical mode decomposition method is used to decompose the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content, construct a soil nutrient time series feature matrix, and calculate the soil nutrient change contribution rate, soil nutrient transfer index, and soil nutrient trend influence rate.

[0009] Among them, the soil nutrient contribution degree equation is used to calculate the contribution ratio of soil nutrients to soil fertility. The input includes the soil nutrient change contribution rate and the soil nutrient time series feature matrix, and the output is the soil nutrient contribution degree feature vector.

[0010] Among them, the soil nutrient transfer degree equation is used to calculate the spatial migration intensity of soil nutrients. The input includes the soil nutrient transfer index and the soil slope factor, and the output is the soil nutrient migration feature vector.

[0011] Among them, the soil nutrient influence degree equation is used to calculate the action intensity of soil texture on soil nutrient changes. The input includes the soil nutrient trend influence rate and the soil texture parameter matrix, and the output is the soil texture influence feature vector.

[0012] Among them, the soil nutrient contribution degree feature vector, the soil nutrient migration feature vector, and the soil texture influence feature vector are input into the convolutional layer after being normalized.

[0013] Among them, the Kriging interpolation method is used to generate the soil nutrient distribution density map, the soil nutrient aggregation index is calculated according to the soil nutrient distribution density map, the support vector machine method is used to establish a soil nutrient change early warning model, and the soil nutrient change early warning threshold is determined.

[0014] The second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions run on a computer, they are used to execute the above-mentioned method for predicting the changes of soil available nutrients and organic carbon.

[0015] The third aspect of the present invention provides a system for predicting the changes of soil available nutrients and organic carbon, which includes the above-mentioned computer-readable storage medium. The system can be any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is set inside the system, and a microprocessor for executing the program instructions stored in the computer-readable storage medium is set inside the system.

[0016] Compared with the prior art, the present invention provides a method, medium, and system for predicting the changes of soil available nutrients and organic carbon. The present invention proposes a method for predicting the changes of soil available nutrients and organic carbon based on multiple couplings. By constructing a soil nutrient determination equation set module and an improved residual neural network structure, high-precision prediction of the dynamic changes of soil nutrients is realized. This method innovatively introduces a time derivative term to describe the dynamic characteristics of nutrient changes, establishes multiple coupling terms to characterize the interaction between different influencing factors, and designs an error correction mechanism to improve the stability of the model. Through empirical mode decomposition and time series analysis, the ability to extract the dynamic characteristics of soil nutrients is improved; by constructing equations for soil nutrient contribution degree, transfer degree, and influence degree, quantitative characterization of multiple influencing factors is realized; by the improved residual neural network structure, the nonlinear fitting ability and prediction accuracy of the model are enhanced. The method of the present invention can effectively solve the problem of low prediction accuracy of the spatio-temporal dynamic changes of soil available nutrients and organic carbon. The main reasons are as follows: The method is based on the physical and chemical mechanisms of soil nutrient cycling, and accurately describes the internal laws of nutrient changes through a mathematical model; the modeling idea of multiple couplings is adopted to effectively integrate multi-source information; deep learning technology is introduced to improve the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

[0019] As Figure 1As shown in the figure, it is a flowchart of a method for predicting the changes in soil available nutrients and organic carbon provided by the first aspect of the present invention. This method includes the following steps: S01. Arrange multiple soil sampling points in a preset soil monitoring area, collect soil samples, and measure the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content in the soil samples; S02. Obtain the terrain elevation data, soil bulk density data, and soil precipitation data of the preset soil monitoring area; S03. Decompose the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content to obtain a soil nutrient time series feature matrix; S04. Establish a terrain undulation index using the terrain elevation data, and calculate the soil slope factor of the preset soil monitoring area; S05. Calculate the soil porosity according to the soil bulk density data, and establish a soil texture parameter matrix; S06. Construct a soil water infiltration coefficient using the soil precipitation data, and calculate the soil water change trend; S07. Calculate the soil nutrient change contribution rate based on the soil nutrient time series feature matrix. The soil nutrient change contribution rate characterizes the influence degree of the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content on the change of soil fertility; S08. Calculate the soil nutrient transfer index according to the soil slope factor and the soil water infiltration coefficient. The soil nutrient transfer index characterizes the migration law of the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content in space; S09. Calculate the soil nutrient trend influence rate based on the soil texture parameter matrix. The soil nutrient trend influence rate characterizes the influence degree of soil texture on the change of the soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content; S10. Input the soil nutrient time series feature matrix, soil slope factor, soil texture parameter matrix, soil water change trend, soil nutrient change contribution rate, soil nutrient transfer index, and soil nutrient trend influence rate into a soil nutrient neural network model to establish a soil nutrient prediction model; S11. Use the soil nutrient prediction model to calculate the predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content; S12. Establish a soil nutrient distribution density map according to the predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content; S13. Calculate the soil nutrient aggregation index in the soil nutrient distribution density map to obtain the spatial distribution characteristics of soil nutrients; S14. Establish a warning threshold for soil nutrient changes based on the spatial distribution characteristics of soil nutrients; S15. When the predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content exceed the warning threshold for soil nutrient changes, output a warning signal for soil nutrient changes; S16. The soil nutrient neural network model adopts an improved residual neural network structure, and a soil nutrient determination equation set module is added before the convolutional layer. The soil nutrient determination equation set module includes a soil nutrient contribution degree equation, a soil nutrient transfer degree equation, and a soil nutrient influence degree equation; S17. The soil nutrient contribution degree equation is used to calculate the contribution ratio of soil nutrients to soil fertility. The inputs include the soil nutrient change contribution rate and the soil nutrient time series feature matrix, and the output is the soil nutrient contribution degree feature vector; S18. The soil nutrient transfer degree equation is used to calculate the spatial migration intensity of soil nutrients. The inputs include the soil nutrient transfer index and the soil slope factor, and the output is the soil nutrient migration feature vector; S19. The soil nutrient influence degree equation is used to calculate the action intensity of soil texture on soil nutrient changes. The inputs include the soil nutrient trend influence rate and the soil texture parameter matrix, and the output is the soil texture influence feature vector. The soil nutrient contribution degree feature vector, the soil nutrient migration feature vector, and the soil texture influence feature vector are input into the convolutional layer.

[0020] The following describes the specific implementation manners of the above steps in detail.

[0021] Specific implementation manner of step S01: First, divide different soil sampling areas according to the land use type, terrain characteristics, and soil type of the area to be detected, set a grid sampling point distribution, with a sampling point spacing of 50 meters. The sampling depth is divided into three soil layers: 0 to 20 cm, 20 to 40 cm, and 40 to 60 cm, and 500 grams of soil samples are taken from each layer. The collected soil samples need to be air-dried naturally, ground, and sieved. The soil available nitrogen content is determined by the Kjeldahl method, the soil available phosphorus content is determined by the sodium bicarbonate extraction molybdenum antimony anti-colorimetric method, the soil available potassium content is determined by the ammonium acetate extraction flame photometry method, and the soil organic carbon content is determined by the potassium dichromate volumetric method. This step obtains the soil basic data through standardized sampling and standardized measurement methods, providing data support for subsequent analysis.

[0022] Specific implementation of step S02: The global positioning system real-time kinematic differential technology is used to obtain the spatial coordinate data of the preset soil monitoring area, the digital elevation model is used to obtain the terrain elevation data, the ring knife method is used to measure the soil bulk density data, and an automatic weather station is deployed in the monitoring area to obtain the precipitation data. The acquisition accuracy of the terrain elevation data is 0.5 m, the measurement error of the soil bulk density data is controlled within 3%, and the precipitation data acquisition interval is 1 hour. This step obtains the soil physical and chemical properties and environmental factor data through multi-source data acquisition, providing basic parameters for subsequent model construction.

[0023] Specific implementation of step S03: The empirical mode decomposition method is used to decompose the available nitrogen content, available phosphorus content, available potassium content, and soil organic carbon content in the soil. Each index is decomposed into multiple intrinsic mode functions and a residual term. The instantaneous frequency of each intrinsic mode function is calculated through the Hilbert transform, and a time series feature matrix of soil nutrients is constructed. This step extracts the dynamic change characteristics of soil nutrients through the signal decomposition method and identifies the periodic change law of soil nutrient content.

[0024] Specific implementation of step S04: Based on the terrain elevation data, a digital terrain model is constructed, and the terrain undulation index is calculated. Specifically, the relative elevation difference method is used to calculate the elevation difference between the highest point and the lowest point per unit area, and the soil slope factor is calculated in combination with the slope and aspect. When the terrain undulation index is greater than 0.3, it indicates that the terrain undulates greatly and soil nutrients are prone to spatial migration. This step quantifies the influence of terrain on the spatial distribution of soil nutrients through terrain analysis.

[0025] Specific implementation of step S05: The soil porosity is calculated according to the soil bulk density data, and the particle size analysis method is used to measure the soil particle composition, including the clay, silt, and sand contents. A soil texture triangle diagram is established to determine the soil texture type, and a soil texture parameter matrix is constructed. The threshold of the total soil porosity is 50%. When the total porosity is greater than this threshold, the soil structure is good and conducive to nutrient retention. This step determines the soil texture characteristics through soil physical property analysis and evaluates the soil nutrient retention ability.

[0026] Specific implementation of step S06: Based on the soil precipitation data, the Green-Ampt equation is used to calculate the soil water infiltration coefficient, and the soil water characteristic curve is established in combination with the soil water content data to analyze the dynamic change characteristics of soil water. The threshold of the soil water infiltration coefficient is 10 mm / h. When the infiltration coefficient is greater than this threshold, soil nutrients are prone to leaching. This step evaluates the influence of soil water on nutrient migration through hydrological process analysis.

[0027] Specific implementation of step S07: Based on the soil nutrient time series feature matrix, the principal component analysis method is used to calculate the weight coefficients of each nutrient index, and the soil nutrient change contribution rate is calculated in combination with the nutrient content change rate. First, calculate the standardized scores of each nutrient index, then calculate the eigenvalues and eigenvectors, determine the number of principal components, and calculate the contribution rates of each nutrient index. The threshold of the soil nutrient change contribution rate is 0.2. When the contribution rate of a certain nutrient index is greater than this threshold, it indicates that this nutrient has a significant impact on the change of soil fertility. This step quantifies the influence degree of different nutrients on soil fertility through multivariate statistical analysis.

[0028] Specific implementation of step S08: According to the soil slope factor and soil water infiltration coefficient, the grey relational analysis method is used to calculate the soil nutrient transfer index. First, construct the soil nutrient spatial distribution sequence, calculate the grey relational degree, establish the soil nutrient spatial migration model, and analyze the migration law of nutrients on the slope. The threshold of the soil nutrient transfer index is 0.6. When the transfer index is greater than this threshold, it indicates that the soil nutrients are prone to spatial migration. This step evaluates the spatial migration characteristics of soil nutrients through the relational analysis method.

[0029] Specific implementation of step S09: Based on the soil texture parameter matrix, the fuzzy comprehensive evaluation method is used to calculate the soil nutrient trend influence rate. First, establish the fuzzy relation matrix, determine the influencing factors of soil texture on nutrient change, calculate the membership function, and obtain the soil nutrient trend influence rate. The threshold of the soil nutrient trend influence rate is 0.4. When the influence rate is greater than this threshold, it indicates that the soil texture has a significant impact on nutrient change. This step evaluates the influence degree of soil texture on nutrient change through the fuzzy mathematics method.

[0030] Specific implementation of step S10: An improved deep learning method is used to construct a soil nutrient neural network model. The soil nutrient time series feature matrix, soil slope factor, soil texture parameter matrix, soil water change trend, soil nutrient change contribution rate, soil nutrient transfer index, and soil nutrient trend influence rate are used as input parameters, and feature extraction and pattern recognition are realized through a multi-layer perceptron structure. This step establishes a soil nutrient prediction model through the deep learning method to realize the intelligent prediction of soil nutrient change.

[0031] Specific implementation of step S11: Using the trained soil nutrient prediction model, input the soil environmental parameters at the current moment, and calculate the predicted values of the available nitrogen content, available phosphorus content, available potassium content, and soil organic carbon content in the soil at the next moment. The root mean square error of the prediction results is controlled within 10%, ensuring the prediction accuracy. This step obtains the future change trend of soil nutrient content through model prediction.

[0032] Specific implementation of step S12: Based on the Kriging interpolation method, a soil nutrient distribution density map is constructed using the predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content. The variogram analysis method is used to determine the optimal interpolation parameters and generate a continuous nutrient spatial distribution image. This step realizes the visual expression of the soil nutrient prediction results through the spatial interpolation method.

[0033] Specific implementation of step S13: The Moran's index in the soil nutrient distribution density map is calculated using the spatial autocorrelation analysis method to determine the soil nutrient aggregation index and analyze the spatial aggregation characteristics and distribution pattern of soil nutrients. The threshold of the soil nutrient aggregation index is 0.5. When the aggregation index is greater than this threshold, it indicates that the soil nutrients have significant spatial aggregation. This step quantifies the spatial distribution characteristics of soil nutrients through the spatial statistical method.

[0034] Specific implementation of step S14: Based on the spatial distribution characteristics of soil nutrients, a support vector machine method is used to establish a soil nutrient change early warning model and determine the soil nutrient change early warning threshold. The radial basis kernel function is selected as the kernel function, and the model parameters are optimized through the cross-validation method to establish a multi-level early warning threshold system. This step establishes a soil nutrient early warning mechanism through machine learning methods to realize the dynamic monitoring of soil nutrient changes.

[0035] Specific implementation of step S15: The predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content are monitored in real time. When the predicted value exceeds the soil nutrient change early warning threshold, the system automatically outputs a soil nutrient change early warning signal, including information such as the early warning level, early warning area, and early warning type. This step realizes the timely early warning of soil nutrient changes through the release of early warning information.

[0036] In the improved residual neural network structure, the deep extraction of soil nutrient change characteristics is realized through the soil nutrient determination equation group module. The soil nutrient contribution degree equation uses the weighted average method to calculate the contribution degree of nutrients to soil fertility. The soil nutrient transfer degree equation uses the exponential decay model to describe the spatial migration characteristics of nutrients. The soil nutrient influence degree equation uses the multiple regression method to analyze the influence degree of soil texture on nutrient changes. After the normalization processing of each feature vector, it is input into the convolutional layer, and high-dimensional features are extracted through multiple convolutional operations to realize the accurate prediction of soil nutrient changes. This step optimizes the model structure through deep learning methods to improve the prediction accuracy.

[0037] The following details the calculation processes or mathematical models involved in the present invention.

[0038] 1. Construction equation of the soil nutrient time series feature matrix: ; In the formula, to represent the time - series characteristic values of the available nitrogen content in the soil; to represent the time - series characteristic values of the available phosphorus content in the soil; to represent the time - series characteristic values of the available potassium content in the soil; to represent the time - series characteristic values of the soil organic carbon content.

[0039] 2. Calculation equation for the contribution rate of soil nutrient change: ; In the formula, is the contribution rate of change of the th nutrient; is the time - weight coefficient, with a value range of 0.3 to 0.5; is the content change value of the th nutrient at the th time point; is the content value of the th nutrient at the th time point; is the time interval; is the space - weight coefficient, with a value range of 0.2 to 0.4; is the spatial variation coefficient of the th nutrient; is the maximum value of the spatial variation coefficients of all nutrients; is the environmental factor weight coefficient, with a value range of 0.2 to 0.4; is the environmental response coefficient; is the random error term, with a value range of 0 to 0.1.

[0040] 3. Calculation equation for the soil nutrient transfer index: ; In the formula, is the transfer index of the th nutrient; is the distance - decay coefficient, with a value range of 0.4 to 0.6; is the nutrient migration distance; is the reference distance, with a value of 100 m; is the slope - influence coefficient, with a value range of 0.3 to 0.5; is the soil water infiltration coefficient; is the slope angle; is the slope length; is the moisture influence coefficient, with a value range of 0.2 to 0.4; is the soil water content; is the error term, with a value range of 0 to 0.1.

[0041] 4. Soil nutrient trend influence rate calculation equation: ; In the formula, is the trend influence rate of the th nutrient; is the soil porosity weight coefficient, with a value range of 0.3 to 0.5; is the soil porosity; is the maximum porosity; is the bulk density weight coefficient, with a value range of 0.2 to 0.4; is the soil bulk density; is the maximum bulk density; is the texture weight coefficient, with a value range of 0.2 to 0.4; is the texture influence factor; is the maximum texture influence factor; is the error term, with a value range of 0 to 0.1.

[0042] 5. Soil nutrient contribution degree equation: ; In the formula, is the contribution degree of the th nutrient; is the nutrient content weight coefficient, with a value range of 0.4 to 0.6; is the nutrient change contribution rate; is the nutrient content characteristic value; is the change rate weight coefficient, with a value range of 0.2 to 0.4; is the time derivative of the nutrient contribution rate; is the time derivative of the nutrient content; is the fluctuation weight coefficient, with a value range of 0.2 to 0.4; is the nutrient content fluctuation coefficient; is the error term, with a value range of 0 to 0.1.

[0043] 6. Soil nutrient transfer degree equation: ; In the formula, is the transfer degree of the th nutrient; is the slope weight coefficient, with a value range of 0.3 to 0.5; is the nutrient transfer index; is the slope angle; is the infiltration weight coefficient, with a value range of 0.3 to 0.5; is the saturated hydraulic conductivity; is the soil water potential; is the reference water potential, with a value of 100 cm water column; is the moisture gradient weight coefficient, with a value range of 0.2 to 0.4; is the vertical gradient of soil water content; is the error term, with a value range of 0 to 0.1.

[0044] 7. Soil nutrient influence degree equation: ; In the formula, is the influence degree of the th nutrient; is the trend weight coefficient, with a value range of 0.4 to 0.6; is the nutrient trend influence rate; is the soil physical and chemical property parameter; is the spatial variation weight coefficient, with a value range of 0.2 to 0.4; is the spatial derivative of the influence rate; is the spatial derivative of the physical and chemical properties; is the correlation weight coefficient, with a value range of 0.2 to 0.4; is the physical and chemical property correlation coefficient; is the error term, with a value range of 0 to 0.1.

[0045] The design principles of the above equations are as follows: 1. The time series feature matrix adopts a two-dimensional matrix form, reflecting the characteristics of different nutrient indicators changing over time; 2. The change contribution rate equation considers the comprehensive influence of time change, spatial distribution, and environmental factors; 3. The transfer index equation is based on the distance decay theory, combined with the slope and moisture migration characteristics; 4. The trend influence rate equation integrates multiple influencing factors of soil physical properties; 5. The contribution degree equation introduces a time derivative term to reflect the dynamic characteristics of nutrient changes; 6. The transfer degree equation is based on the water flow movement theory, considering the influence of gravitational potential and matrix potential; 7. The influence degree equation considers spatial variability and introduces correlation analysis.

[0046] The derivation process of each equation is described in detail below.

[0047] 1. Derivation of the time series feature matrix of soil nutrients: First, perform empirical mode decomposition on the original soil nutrient data to obtain multiple intrinsic mode functions; then calculate the instantaneous frequency of each intrinsic mode function through Hilbert transform; next, use principal component analysis to extract eigenvalues, and select the 4 eigenvalues with the largest contribution rate to construct a feature matrix. This matrix can reflect the time evolution characteristics of different nutrient contents, effectively reduce the data dimension, and improve the calculation efficiency. The eigenvalues in the matrix are calculated from the experimentally measured data, with a measurement interval of 7 days and continuous measurement for 12 months.

[0048] 2. Derivation of the soil nutrient change contribution rate equation: The first step is to construct the nutrient content change rate term , which reflects the relative change rate of nutrient content; the second step is to introduce the spatial variation coefficient , which characterizes the spatial distribution characteristics of nutrients; the third step is to add the environmental response term , which reflects the influence of environmental factors; the fourth step is to determine the weight coefficients through multiple linear regression , , . This equation comprehensively considers time changes, spatial distribution, and environmental impacts, and can accurately evaluate the contribution degree of different nutrients to soil fertility.

[0049] 3. Derivation of the soil nutrient transfer index equation: First, construct an exponential decay term based on the distance decay principle ; then introduce the overland flow term , which describes the overland nutrient transport characteristics; finally, add the water influence term . The parameters , , are calibrated through field experiments. The experiments set 3 slope gradients (5 degrees, 15 degrees, 25 degrees) and 3 water content levels (15%, 25%, 35%), and the observation period is 6 months. This equation can quantitatively describe the transport law of nutrients on the slope.

[0050] 4. Derivation of the soil nutrient trend influence rate equation: Based on the influence mechanism of soil physical properties on nutrient changes, construct a comprehensive evaluation equation considering porosity, bulk density, and texture. The weight coefficients of each item are determined through the fuzzy Delphi method, consulting 15 soil science experts, and determining the final weight range through 3 rounds of feedback. This equation realizes the quantitative evaluation of the influence of soil physical properties.

[0051] 5. Derivation of the soil nutrient contribution degree equation: Introduce the coupling term of nutrient content and contribution rate , construct the time derivative term to describe the dynamic change characteristics, and add the fluctuation coefficient Characterize stability. The weight coefficients are obtained through BP neural network training, with 1000 groups of training samples and 200 groups of verification samples. This equation can evaluate the actual contribution effect of nutrients on soil fertility.

[0052] 6. Derivation of the soil nutrient transfer equation: Based on Darcy's law and Richards' equation, a comprehensive equation considering slope, infiltration, and water migration is constructed. Parameter calibration is carried out through indoor soil column leaching experiments, setting 5 soil types and 3 leaching intensities, with a monitoring period of 30 days. This equation realizes an accurate description of the nutrient migration process.

[0053] 7. Derivation of the soil nutrient influence equation: Construct a coupling term of the trend influence rate and soil physical and chemical properties, introduce a spatial derivative term to describe spatial variation characteristics, and add a correlation coefficient term to characterize the interaction. The parameters are optimized by co-kriging method and verified with the measured data of 500 sampling points. This equation can accurately evaluate the influence degree of soil physical and chemical properties on nutrient changes.

[0054] Methods for obtaining parameters in each equation: 1. Temporal weight coefficient : Determined through time series analysis, based on 12 months of continuous monitoring data; 2. Spatial weight coefficient : Adopt the variogram analysis method, based on the spatial distribution data of 500 sampling points; 3. Environmental factor weight coefficient : Through principal component analysis, based on environmental factor data such as temperature, precipitation, and light; 4. Distance decay coefficient : Adopt the spatial autocorrelation analysis method, based on sampling data at different scales; 5. Slope influence coefficient : Determined through field runoff plot experiments, setting different slope gradients; 6. Moisture influence coefficient : Adopt the soil water characteristic curve method, based on the measurement data under different water content conditions; 7. Soil physical and chemical property parameters are measured by conventional experimental methods, including measuring water content by oven drying method, measuring bulk density by core cutter method, and measuring mechanical composition by pipette method, etc.

[0055] The second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions run on a computer, they are used to execute the above-mentioned method for predicting the changes of soil available nutrients and organic carbon.

[0056] The third aspect of the present invention provides a system for predicting the changes in available soil nutrients and organic carbon, which includes the above-mentioned computer-readable storage medium. The system can be any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is set inside the system, and a microprocessor for executing the program instructions stored in the computer-readable storage medium is set inside the system.

[0057] Specifically, the principle of the present invention is as follows: The technical principle of the present invention is based on the basic laws of soil nutrient cycling and the modeling idea of system coupling. The spatio-temporal changes of soil nutrients are a complex dynamic process, which is affected by various factors such as soil physical and chemical properties, environmental conditions, and human management. There are non-linear interaction relationships among these factors, forming a dynamic feedback system. The present invention constructs a soil nutrient determination equation set to transform these complex influence mechanisms into computable mathematical models.

[0058] From the perspective of the change mechanism of soil nutrients, the temporal change of nutrient content reflects the dynamic balance of its accumulation and consumption, and the spatial change reflects the process of nutrient migration and redistribution. The contribution degree equation designed by the present invention describes the change rate of nutrient content by introducing a time derivative term; the transfer degree equation is based on the theory of mass transport and considers the influence of terrain and hydrological factors on the spatial distribution of nutrients; the influence degree equation analyzes the regulation effect of the substrate on nutrient changes from the perspective of soil physical and chemical properties. These three equations form a complete equation set system, which can comprehensively describe the dynamic change law of soil nutrients.

[0059] The improved residual neural network structure realizes the deep extraction of soil nutrient change characteristics through multi-layer non-linear transformation. The introduction of the residual structure solves the problem of difficult training of deep networks and improves the convergence performance of the model. The addition of the determination equation set module provides the neural network with feature inputs with physical meanings, enhancing the interpretability and prediction ability of the model. This method combining mechanism and data-driven not only ensures the physical rationality of the model but also improves the prediction accuracy.

[0060] Scheme summary: The method for predicting the changes in available soil nutrients and organic carbon proposed by the present invention first collects soil samples in a preset soil monitoring area and measures the nutrient content to obtain environmental parameter data; then constructs a time series feature matrix through empirical mode decomposition, and calculates the nutrient change contribution rate, transfer index, and trend influence rate; then uses the soil nutrient determination equation set module and the improved residual neural network structure to establish a prediction model; finally, outputs the prediction results of soil nutrient changes and early warning signals, realizing the accurate prediction of soil nutrient changes.

[0061] The following provides a specific embodiment 1 of the present invention, and the specific implementation manners of each step in this embodiment 1 are described in detail as follows.

[0062] The specific implementation of step S01 is as follows: Different soil sampling areas are divided according to the land use type, terrain features, and soil type of the area to be detected. A grid-like sampling point distribution is set, with a sampling point spacing of 50 meters. The sampling depth is divided into three soil layers: 0 to 20 cm, 20 to 40 cm, and 40 to 60 cm. 500 grams of soil samples are taken from each layer. A stainless-steel soil sampling drill is used as the sampler. To avoid cross-contamination, the sampler needs to be cleaned after each sampling. The collected soil samples need to be naturally air-dried, ground, and sieved. The air-drying temperature is controlled at 25 degrees Celsius. An agate mortar is used for grinding, and a standard sieve with a pore size of 2 mm is used for sieving. The Kjeldahl method is used to determine the available nitrogen content in the soil. Specifically, 10 grams of soil samples are weighed, 50 ml of potassium chloride solution is added for extraction, shaken for 30 minutes, filtered, and the filtrate is taken for nitrogen determination. The calculation formula is: ; In the formula, is the available nitrogen content in the soil, with the unit of milligram per kilogram; is the sample titration volume, with the unit of milliliter; is the blank titration volume, with the unit of milliliter; is the concentration of the hydrochloric acid standard solution, with the unit of mole per liter; is the mass of the soil sample, with the unit of gram. The sodium bicarbonate extraction molybdenum antimony anti-colorimetric method is used to determine the available phosphorus content in the soil. Specifically, 2.5 grams of soil samples are weighed, 50 ml of sodium bicarbonate solution is added for extraction, shaken for 30 minutes, filtered, the filtrate is taken for color development, and the absorbance is measured at a wavelength of 710 nm. The calculation formula is: ; In the formula, is the available phosphorus content in the soil, with the unit of milligram per kilogram; is the phosphorus concentration obtained from the standard curve, with the unit of microgram per milliliter; is the volume of the extraction solution, with the unit of milliliter; is the mass of the soil sample, with the unit of gram; is the dilution factor. The ammonium acetate extraction flame photometry method is used to determine the available potassium content in the soil. Specifically, 5 grams of soil samples are weighed, 50 ml of ammonium acetate solution is added for extraction, shaken for 30 minutes, filtered, and the filtrate is taken for determination. The calculation formula is: ; In the formula, is the available potassium content in the soil, with the unit of milligram per kilogram; is the potassium concentration obtained from the standard curve, with the unit of microgram per milliliter; is the volume of the extraction solution, with the unit of milliliter; is the dilution factor; Let ; In the formula, is the soil organic carbon content, in grams per kilogram; is the blank titration volume, in milliliters; is the sample titration volume, in milliliters; is the concentration of the ferrous sulfate standard solution, in moles per liter; is the soil sample mass, in grams. This step obtains the soil basic data through standardized sampling and standardized measurement methods, providing data support for subsequent analysis.

[0063] The specific implementation of step S02 is as follows: The global positioning system real-time kinematic differential technology is used to obtain the spatial coordinate data of the preset soil monitoring area. The positioning accuracy of the sampling points is better than 1 centimeter, and the longitude, latitude, and altitude of each sampling point are recorded. The digital elevation model is used to obtain the terrain elevation data. The resolution of the elevation data is 30 meters, and the bilinear interpolation method is used to resample the elevation data. The resolution after resampling is 5 meters, which is used for subsequent terrain factor calculation. The core cutter method is used to measure the soil bulk density data. Specifically, it includes using a core cutter with a volume of 100 cubic centimeters to collect undisturbed soil, drying and weighing it. The calculation formula is: ; In the formula, is the soil bulk density, in grams per cubic centimeter; is the mass of the dried soil, in grams; is the volume of the core cutter, in cubic centimeters. An automatic weather station is deployed in the monitoring area to obtain precipitation data. The weather station is equipped with a tipping bucket rain gauge with a measurement accuracy of 0.1 millimeter and a collection interval of 1 hour. The daily precipitation calculation formula is: ; In the formula, is the daily precipitation, in millimeters; is the precipitation at the th hour, in millimeters. The monthly precipitation calculation formula is: ; In the formula, is the monthly precipitation, in millimeters; is the daily precipitation on the th day, in millimeters; is the number of days in the month.

[0064] The specific implementation of step S03 is as follows: The empirical mode decomposition method is used to decompose the available nitrogen content, available phosphorus content, available potassium content, and soil organic carbon content in the soil. First, construct the original data sequence: ; Then, by finding the local extreme points of the data sequence, the upper and lower envelope lines are constructed using the cubic spline interpolation method, and the mean function is calculated: ; In the formula, is the mean function of the first decomposition; is the upper envelope line; is the lower envelope line. Calculate the first component: ; Repeat the above process until the conditions of the intrinsic mode function are met, and the first intrinsic mode function is obtained. Repeat the above process for the remaining sequence to obtain multiple intrinsic mode functions. Calculate the instantaneous frequency of each intrinsic mode function through the Hilbert transform: ; In the formula, is the instantaneous frequency of the th intrinsic mode function; is the instantaneous phase. Finally, construct the soil nutrient time series feature matrix: ; Each row of this matrix represents a nutrient index, and each column represents a time eigenvalue.

[0065] The specific implementation of step S04 is as follows: Calculate the terrain undulation index based on the digital elevation model. The calculation formula is: ; In the formula, is the terrain undulation index; is the elevation of the highest point within the unit area, with the unit of meter; is the elevation of the lowest point within the unit area, with the unit of meter; is the unit area, with a value of 1 square kilometer. When calculating the slope factor, first calculate the surface slope: ; In the formula, is the surface slope, with the unit of degree; is the elevation value, with the unit of meter; and are the horizontal coordinates. Then calculate the soil slope factor: ; In the formula, is the soil slope factor, dimensionless. When the terrain undulation index is greater than 0.3, it indicates that the terrain undulates greatly and soil nutrients are prone to spatial migration.

[0066] The specific implementation of step S05 is as follows: Calculate the soil porosity according to the soil bulk density data. The formula for the total porosity is: ; In the formula, is the total porosity, dimensionless; is the soil bulk density, with the unit of grams per cubic centimeter; is the true soil density, with the unit of grams per cubic centimeter, and the value is 2.65. The pipette method is used to determine the soil mechanical composition, and the contents of soil particles of different particle sizes are measured, including sand particles with a particle size greater than 0.05 mm, silt particles with a particle size of 0.05 to 0.002 mm, and clay particles with a particle size less than 0.002 mm. The formula is: ; In the formula, is the content of the th grade particles, with the unit of percentage; is the mass of the th grade particles, with the unit of grams; is the total mass of the soil sample, with the unit of grams. Establish a soil texture parameter matrix: ; In the formula, the three rows of the matrix represent sand particles, silt particles, and clay particles respectively, and the three columns represent three soil layers. This step determines the soil texture characteristics through soil physical property analysis and evaluates the soil nutrient retention ability. The threshold of the soil total porosity is 50%. When the total porosity is greater than this threshold, the soil structure is good, which is conducive to nutrient retention.

[0067] The specific implementation of step S06 is as follows: Based on the soil precipitation data, use the Green-Ampt equation to calculate the soil water infiltration coefficient. The formula is: ; In the formula, is the unsaturated permeability coefficient, with the unit of centimeters per second; is the saturated permeability coefficient, with the unit of centimeters per second; is the soil volumetric water content, dimensionless; is the residual water content, dimensionless; is the saturated water content, dimensionless; is the soil characteristic parameter, which is determined through experiments. Combine the soil water content data to establish a soil water characteristic curve: ; In the formula, is the soil water potential, with the unit of centimeter water column; is the air entry value, with the unit of centimeter water column; is the empirical parameter. Calculate the trend of soil moisture change: ; In the formula, is the trend of soil moisture change, with the unit of per second; is the time variable; is the spatial variable in the vertical direction. The threshold of the soil moisture permeability coefficient is 10 millimeters per hour. When the permeability coefficient is greater than this threshold, soil nutrients are prone to leaching.

[0068] The specific implementation of step S07 is: Calculate the contribution rate of soil nutrient change based on the soil nutrient time series feature matrix. The calculation formula is: ; The meanings and value-taking methods of each parameter in the formula have been given above. The parameter calibration uses the least squares method to establish the optimization objective function: ; In the formula, is the measured contribution rate of the th nutrient at the th sample; is the model calculation value; is the number of samples. Solve the optimal parameter combination through the genetic algorithm. The population size is set to 100, the number of evolutionary generations is 500, the crossover probability is 0.8, and the mutation probability is 0.1.

[0069] The specific implementation of step S08 is: Calculate the soil nutrient transfer index according to the soil slope factor and the soil moisture permeability coefficient. The calculation formula is: ; The meanings and value-taking methods of each parameter in the formula have been given above. The physical meaning of the soil nutrient transfer index is to describe the migration characteristics of nutrients on the slope surface, considering three main processes: distance attenuation, overland flow, and water movement. The parameter calibration uses a distributed hydrological model to construct an overland flow generation module: ; In the formula, is the overland flow rate, with the unit of cubic meter per second; is the thickness of the overland flow layer, with the unit of meter; and are undetermined parameters. The overland flow concentration uses the kinematic wave equation: ; In the formula, is the water depth, with the unit of meter; is the single-width flow rate, with the unit of cubic meters per second per meter; is the rainfall intensity, with the unit of meters per second.

[0070] The specific implementation of step S09 is: calculating the soil nutrient trend influence rate based on the soil texture parameter matrix, and the calculation formula is: ; The meanings and value-taking methods of the parameters in the formula have been given above. Constructing a fuzzy evaluation model, first establish the factor set and the evaluation set , and then construct the membership function: ; In the formula, is the membership function; , , are undetermined parameters. Establish the fuzzy relation matrix: ; In the formula, represents the membership degree of the th factor to the th evaluation level.

[0071] The specific implementation of step S10 is: inputting the soil nutrient time series feature matrix, soil slope factor, soil texture parameter matrix, soil moisture change trend, soil nutrient change contribution rate, soil nutrient transfer index, and soil nutrient trend influence rate into the soil nutrient neural network model. The network structure adopts an improved residual network, including an input layer, multiple residual blocks, and an output layer. The mathematical expression of the residual block is: ; In the formula, is the output of the residual block; is the residual function; is the input; is the weight parameter. Add a decision equation set module before the convolutional layer, including a contribution degree equation, a transfer degree equation, and an influence degree equation. The specific forms of these equations have been given above. The model is trained using the stochastic gradient descent method, and the loss function is the mean square error: ; In the formula, is the loss function value; is the true value; is the predicted value; is the number of samples. The initial value of the learning rate is set to 0.01, and the cosine annealing strategy is used for dynamic adjustment.

[0072] The specific implementation of step S11 is as follows: Using the trained soil nutrient prediction model, calculate the predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content. The prediction process first standardizes the input data: ; In the formula, is the standardized data; is the original data; is the minimum value of the data; is the maximum value of the data. Then calculate the predicted value through the forward propagation of the model: ; In the formula, is the predicted value; is the activation function; is the weight matrix; is the bias term. Finally, perform inverse standardization to obtain the actual predicted value: ; In the formula, is the actual predicted value; is the maximum value of the label; is the minimum value of the label. The root mean square error of the prediction result is controlled within 10%.

[0073] The specific implementation of step S12 is as follows: Use the Kriging interpolation method to construct a soil nutrient distribution density map. First, calculate the experimental variogram: ; In the formula, is the variogram value; is the number of sample point pairs with an interval distance of ; is the observation value at position . Then fit the theoretical variogram model: ; In the formula, is the nugget effect; is the sill; is the range. Establish the Kriging equations: ; ; In the formula, is the weight coefficient; is the Lagrange multiplier; is the sample point and distance between; is the sample point The distance from the point to be estimated.

[0074] The specific implementation of step S13 is as follows: The Moran's Index in the soil nutrient distribution density map is calculated using the spatial autocorrelation analysis method. The calculation formula is: ; In the formula, is the Moran's Index; is the number of sample points; is the spatial weight; and are the observed values at positions and ; is the average value. Calculate the local Moran's Index: ; In the formula, is the local Moran's Index at position . The value range of the Moran's Index is from -1 to 1. When the index is greater than 0, it indicates positive correlation, and when it is less than 0, it indicates negative correlation.

[0075] The specific implementation of step S14 is as follows: Based on the spatial distribution characteristics of soil nutrients, a support vector machine method is used to establish a soil nutrient change early warning model. Construct the optimization problem: ; The constraint conditions are: ; ; In the formula, is the weight vector; is the penalty parameter; is the slack variable; is the feature mapping function; is the class label. Use the sequential minimal optimization algorithm to solve the dual problem: ; The constraint conditions are: ; ; In the formula, is the Lagrange multiplier; is the kernel function. Select the radial basis kernel function: ; In the formula, is the kernel parameter. Determine the optimal parameter combination through the cross-validation method.

[0076] The specific implementation of step S15 is as follows: The predicted values of soil available nitrogen content, soil available phosphorus content, soil available potassium content, and soil organic carbon content are monitored using the segmented threshold method. A multi-level early warning threshold system is constructed: ; where, is the early warning level; is the predicted value; , , are the threshold demarcation points. The thresholds are determined using the clustering analysis method. First, the historical data is standardized, and then the k-means clustering algorithm is used: ; where, is the clustering objective function; is the number of clusters; is the number of samples in the th class; is the th sample in the th class; is the th class clustering center.

[0077] The specific implementation of steps S16 to S19 is as follows: The soil nutrient determination equation set module includes the soil nutrient contribution equation: ; the soil nutrient transfer equation: ; the soil nutrient influence equation: ; After the normalization process, each eigenvector is input into the convolutional layer: ; where, is the normalized eigenvector; is the original eigenvector; is the mean value; is the standard deviation. The convolutional operation uses the padding strategy to maintain the feature map size: ; where, is the output feature map; is the convolutional kernel; is the input feature map; and are the convolutional kernel sizes. The batch normalization layer is used for data standardization, and the exponential linear unit is used as the activation function: ; In the formula, is a hyperparameter with a value of 0.1. Finally, the prediction result is obtained through the fully connected layer: ; In the formula, is the weight matrix; is the feature vector; is the bias term. This module realizes the deep extraction of the characteristics of soil nutrient changes through multi-layer non-linear transformation, improving the prediction accuracy. The model is trained using the mini-batch gradient descent method, with a batch size of 64, 100 training epochs, and the learning rate is adjusted every 10 epochs. The model is verified using the five-fold cross-validation method, and the prediction accuracy of the validation set reaches more than 90%.

[0078] In this embodiment, through the organic combination of multiple mathematical models and algorithms, the accurate prediction of the changes in soil available nutrients and organic carbon is realized, which has important theoretical significance and application value. The innovation of this method lies in introducing the time derivative term to describe the dynamic change characteristics, constructing multiple coupling terms to reflect the interaction between different factors, adding an error correction term to improve the stability and reliability of the model, adopting a multi-layer nested structure to effectively express complex systems, and introducing the spatial derivative term to enhance the characterization ability of spatial variation characteristics.

[0079] To better understand and implement the present invention, the following provides Embodiment 2 of a specific application scenario of the present invention: The Soil Research Institute of a certain Agricultural Science Academy conducts soil nutrient monitoring research at a certain experimental base in Wuzhong District, Suzhou City, Jiangsu Province. The area of the experimental area is 100 hectares, and the land use types include paddy fields, dry land, and orchards, and the terrain conditions include flat land and gentle slopes. In order to achieve the accurate prediction of the changes in soil available nutrients and organic carbon, the research team uses the method of the present invention to carry out research.

[0080] First step, sampling points are arranged in a grid of 50 meters × 50 meters within the experimental area, a total of 400 sampling points are set, and the sampling depths are divided into three soil layers of 0 to 20 cm, 20 to 40 cm, and 40 to 60 cm. The sampling time is from January to December 2024, and sampling is carried out once a month. The soil samples are air-dried naturally, ground, and passed through a 2-mm sieve, and the measurement results are shown in Table 1.

[0081] Table 1 Measurement results of soil nutrient content (average value of 0-20 cm soil layer) Month Available nitrogen (mg / kg) Available phosphorus (mg / kg) Available potassium (mg / kg) Organic carbon (g / kg) January 125.6 28.4 156.8 18.6 February 118.4 26.7 148.5 18.2 March 132.5 30.1 162.4 19.1 April 145.8 32.6 175.6 19.8 May 138.4 29.8 168.3 19.4 June 128.7 27.5 159.2 18.9 Second step, the global positioning system real-time kinematic differential technology is used to obtain the spatial coordinate data of the sampling points, and the digital elevation model is used to obtain the terrain data. The elevation range of the sampling points is 2.5 to 15.8 meters, and the slope range is 0 to 8 degrees. The ring knife method is used to measure the soil bulk density, and the measurement results are shown in Table 2.

[0082] Table 2 Soil bulk density measurement results (g / cm³) Soil depth Flat land Gentle slope land 0 - 20 cm 1.32 1.28 20 - 40 cm 1.38 1.35 40 - 60 cm 1.42 1.40 The daily precipitation data recorded by the automatic weather station ranges from 0 to 85.6 mm, and the monthly precipitation ranges from 45.8 to 312.5 mm.

[0083] In the third step, the measured soil nutrient data is decomposed by empirical mode decomposition to obtain 4 intrinsic mode functions and 1 residual term. The instantaneous frequency is calculated through Hilbert transform to construct the soil nutrient time series feature matrix: .

[0084] In the fourth step, the topographic undulation index is calculated to be 0.132, indicating that the terrain of the test area is relatively flat. The calculation results of the soil slope factor range from 0.025 to 0.156.

[0085] In the fifth step, the total porosity is calculated according to the soil bulk density data. The total porosity of the flat ground soil is 50.2%, and the total porosity of the gentle slope soil is 51.7%. The measurement results of the soil mechanical composition are shown in Table 3.

[0086] Table 3 Soil mechanical composition (%) Soil depth Sand Silt Clay 0 - 20 cm 28.5 45.6 25.9 20 - 40 cm 26.8 46.2 27.0 40 - 60 cm 25.4 47.1 27.5 In the sixth step, the Green-Ampt equation is used to calculate the soil water infiltration coefficient. The saturated infiltration coefficient is 8.5 mm per hour, and the unsaturated infiltration coefficient ranges from 0.2 to 5.6 mm per hour.

[0087] In the seventh step, the contribution rate of soil nutrient change is calculated. The time weight coefficient is taken as 0.4, the space weight coefficient is taken as 0.3, and the environmental factor weight coefficient is taken as 0.3. Calculation results: the contribution rate of available nitrogen is 0.385, the contribution rate of available phosphorus is 0.256, the contribution rate of available potassium is 0.312, and the contribution rate of organic carbon is 0.298.

[0088] In the eighth step, the soil nutrient transfer index is calculated. The distance decay coefficient is taken as 0.5, the slope influence coefficient is taken as 0.4, and the water influence coefficient is taken as 0.3. Calculation results: the transfer index of available nitrogen is 0.456, the transfer index of available phosphorus is 0.389, the transfer index of available potassium is 0.412, and the transfer index of organic carbon is 0.245.

[0089] In the ninth step, the soil nutrient trend influence rate is calculated. The soil porosity weight coefficient is taken as 0.4, the bulk density weight coefficient is taken as 0.3, and the texture weight coefficient is taken as 0.3. Calculation results: the influence rate of available nitrogen is 0.365, the influence rate of available phosphorus is 0.312, the influence rate of available potassium is 0.342, and the influence rate of organic carbon is 0.328.

[0090] In the tenth step, input the above calculation results into the improved residual neural network model. The network structure includes 1 input layer, 4 residual blocks, and 1 output layer. The first 6 months of data are used for model training, and the last 6 months of data are used for validation. After 100 rounds of training, the model converges, and the training error is less than 5%.

[0091] In the eleventh step, use the trained model to predict the soil nutrient content. The prediction results are shown in Table 4.

[0092] Table 4 Prediction Results of Soil Nutrient Content Month Available nitrogen (mg / kg) Available phosphorus (mg / kg) Available potassium (mg / kg) Organic carbon (g / kg) July 135.2 28.9 164.7 19.2 August 142.6 31.2 171.5 19.6 September 138.9 29.5 167.8 19.3 October 130.4 27.8 160.6 18.8 November 122.8 26.2 152.3 18.4 December 119.5 25.6 147.9 18.1 In the twelfth step, use the Kriging interpolation method to generate the soil nutrient distribution density map. The range value is 150 meters, and the ratio of nugget value to sill value is 0.15.

[0093] In the thirteenth step, calculate the Moran index. The Moran index of available nitrogen is 0.685, available phosphorus is 0.624, available potassium is 0.658, and organic carbon is 0.712, all showing significant spatial positive correlations.

[0094] In the fourteenth and fifteenth steps, establish a soil nutrient change warning model. The warning levels are divided into four levels: safe (green), attention (yellow), warning (orange), and danger (red). When the predicted value approaches or exceeds the warning threshold, the system automatically issues a warning signal.

[0095] Traditional soil nutrient monitoring methods mainly rely on regular sampling and analysis to obtain nutrient content data through chemical determination. This method has the following problems: low sampling frequency, inability to timely reflect the dynamic changes of soil nutrients; lack of consideration of the influence of environmental factors such as terrain and hydrology; lack of prediction and warning functions. The method of the present invention has made the following improvements on the basis of the traditional method: established a complete prediction system for the dynamic changes of soil nutrients, and achieved accurate prediction of nutrient changes through the organic combination of multiple mathematical models; considered multiple influencing factors, improving the scientificity and reliability of prediction; has a warning function, capable of timely discovering soil nutrient problems and taking corresponding measures. The test results show that the prediction accuracy of the method of the present invention reaches more than 95%, which is 30% higher than the traditional method.

[0096] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Table 5 below.

[0097] Table 5 Variable Explanation Table

[0098] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A method for predicting changes in soil available nutrients and organic carbon, characterized in that: The method comprises the following steps: collecting soil samples in a preset soil monitoring area and determining the soil available nitrogen content, soil available phosphorus content, soil available potassium content and soil organic carbon content, acquiring terrain elevation data, soil bulk density data and soil precipitation data, and establishing a soil nutrient prediction model, wherein the soil nutrient prediction model adopts an improved residual neural network structure, and a soil nutrient determination equation group module is added before the convolution layer, wherein the soil nutrient determination equation group module comprises a soil nutrient contribution degree equation, a soil nutrient transfer degree equation and a soil nutrient influence degree equation, and a soil nutrient aggregation index is calculated to obtain soil nutrient spatial distribution characteristics, and a soil nutrient change warning threshold is established according to the soil nutrient spatial distribution characteristics, and a soil nutrient change warning signal is output.

2. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The steps of collecting soil samples and determining the nutrient content of the soil include: determining the available nitrogen content of the soil by the Kjeldahl method, determining the available phosphorus content of the soil by the molybdenum antimony colorimetric method with sodium bicarbonate extraction, determining the available potassium content of the soil by the ammonium acetate flame photometry, and determining the organic carbon content of the soil by the potassium dichromate volumetric method.

3. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The steps for data processing of soil nutrient content measurement values ​​are to decompose soil available nitrogen content, soil available phosphorus content, soil available potassium content and soil organic carbon content using the empirical mode decomposition method, construct a soil nutrient time series characteristic matrix, and calculate the soil nutrient change contribution rate, soil nutrient transfer index and soil nutrient trend impact rate.

4. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The soil nutrient contribution equation is used to calculate the contribution ratio of soil nutrients to soil fertility, and the input includes the soil nutrient change contribution rate and the soil nutrient time series characteristic matrix, and the output is the soil nutrient contribution characteristic vector.

5. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The soil nutrient transfer degree equation is used to calculate the spatial migration intensity of soil nutrients. The input includes the soil nutrient transfer index and the soil slope factor, and the output is a soil nutrient migration characteristic vector.

6. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The soil nutrient influence degree equation is used to calculate the effect intensity of soil texture on soil nutrient changes. The input includes the soil nutrient trend influence rate and the soil texture parameter matrix, and the output is the soil texture influence characteristic vector.

7. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The soil nutrient contribution feature vector, the soil nutrient migration feature vector and the soil texture impact feature vector are input into the convolution layer after being normalized.

8. The method for predicting changes in soil available nutrients and organic carbon according to claim 1, characterized in that: The soil nutrient distribution density map is generated by the Kriging interpolation method, the soil nutrient aggregation index is calculated according to the soil nutrient distribution density map, the soil nutrient change early warning model is established by the support vector machine method, and the soil nutrient change early warning threshold is determined.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the method for predicting changes in available nutrients and organic carbon in soil according to any one of claims 1 to 8.

10. A soil available nutrient and organic carbon change prediction system, characterized in that: The system comprises the computer-readable storage medium as claimed in claim 9, wherein the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

Images