Intelligent prediction method and device for dynamic change trend of soil organic carbon and electronic equipment

Abstract

The invention relates to the technical field of soil management, in particular to an intelligent prediction method and device for the dynamic change trend of soil organic carbon and electronic equipment. Acquiring soil monitoring data, terrain monitoring data, crop monitoring data, facility monitoring data and environment monitoring data corresponding to the target area; performing feature extraction on the soil monitoring data, the terrain monitoring data, the crop monitoring data, the facility monitoring data and the environment monitoring data, and determining a water-soil bearing capacity index corresponding to the target area; based on the water and soil bearing capacity index, the soil monitoring data, the terrain monitoring data, the crop monitoring data and the environment monitoring data, predicting a soil organic carbon dynamic change trend corresponding to the target area; and determining a target measure corresponding to the target area based on the soil organic carbon dynamic change trend. The dynamic change trend of soil organic carbon is predicted, the target area is processed based on the prediction result, and the water and soil resource utilization efficiency and the agricultural carbon emission reduction level are improved.

Description

Technical Field

[0001] This invention relates to the field of soil management technology, specifically to a method, device, and electronic equipment for intelligent prediction of dynamic changes in soil organic carbon. Background Technology

[0002] Current farmland water and soil resource management generally suffers from problems such as lagging monitoring, low integration, extensive regulation, and insufficient accuracy in carbon accounting. Current monitoring methods mostly rely on manual sampling or single sensors, and the data obtained are insufficient in terms of spatiotemporal resolution and real-time performance, failing to dynamically reflect changes in soil moisture, nutrients, and meteorological conditions.

[0003] While remote sensing technology can achieve wide-area coverage, its limitations in resolution and revisit frequency make it difficult to meet the needs of precise land-level management. Furthermore, water and soil resource management is often separated from carbon sequestration assessment and carbon reduction strategies, lacking a unified data platform and decision-making mechanism. This results in carbon accounting results not being able to directly guide agricultural management measures such as irrigation and fertilization. Existing irrigation and fertilization control methods rely heavily on experience-based judgments, lacking intelligent optimization strategies based on real-time monitoring data, which can easily lead to water and fertilizer waste, soil degradation, and even environmental pollution.

[0004] Therefore, how to predict the dynamic trend of soil organic carbon change and treat target areas based on the prediction results in order to improve the efficiency of water and soil resource utilization and the level of agricultural carbon emission reduction has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent prediction method, device and electronic device for dynamic change trend of soil organic carbon, so as to solve the problem of how to predict the dynamic change trend of soil organic carbon and process the target area based on the prediction results.

[0006] In a first aspect, the present invention provides an intelligent prediction method for the dynamic change trend of soil organic carbon. The method includes: acquiring soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to a target area; extracting features from the soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the soil and water carrying capacity index corresponding to the target area; predicting the dynamic change trend of soil organic carbon corresponding to the target area based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data; and determining the target measures corresponding to the target area based on the dynamic change trend of soil organic carbon.

[0007] In one optional implementation, feature extraction is performed on soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the soil and water carrying capacity index corresponding to the target area. This includes: determining static data and time-series dynamic data from the soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data; dividing the target area into at least one grid cell; for each grid cell, selecting target static data from the static data; constructing a static vector based on the target static data; concatenating the static vectors corresponding to each grid cell to generate a static matrix; extracting features from the static matrix to obtain static fusion features; generating a dynamic input tensor based on the duration, dimension, and number of grid cells corresponding to the time-series dynamic data; extracting features from the dynamic input tensor to obtain dynamic fusion features; fusing the static fusion features and dynamic fusion features to generate target fusion features; and inputting the target fusion features into a preset soil and water carrying capacity assessment model to output the soil and water carrying capacity index corresponding to the target area.

[0008] In one optional implementation, feature extraction is performed on the static matrix to obtain static fusion features, including: calculating the correlation degree between each sub-static feature in the static matrix and soil and water bearing capacity using the Pearson correlation coefficient; converting the correlation degree into initial feature weights; inputting the static matrix and initial feature weights into a preset random forest model; the preset random forest model binds the initial feature weights to the column vectors of the static matrix to form a sub-static feature-initial weight mapping table; for each grid cell in the static matrix, based on the sub-static feature-initial weight mapping table, modifying each sub-static feature in the static matrix, and obtaining a weighted static matrix based on the modified sub-static features; performing feature extraction on the weighted static matrix to obtain each target splitting feature; for each target splitting feature, calculating the first weighting coefficient corresponding to the target splitting feature based on the initial feature weight corresponding to the target splitting feature; and fusing each target splitting feature based on the first weighting coefficient corresponding to each target splitting feature to obtain static fusion features.

[0009] In one optional implementation, feature extraction is performed on the dynamic input tensor to obtain dynamic fusion features, including: inputting the dynamic input tensor into the input gate of a preset temporal processing model; the input gate identifies the dynamic input tensor, filters valid information through the sigmoid function, and transmits the valid information to the forget gate; the forget gate identifies the valid information, discards invalid information, and transmits the remaining valid information to the output gate; the output gate integrates the output information of the forget gate and outputs the conventional temporal features of each grid cell; the current growth period of the crop is obtained; based on the current growth period, each conventional temporal feature is corrected to obtain corrected temporal features; the resource demand threshold of the growth period corresponding to each corrected sub-dynamic feature corresponding to each corrected temporal feature corresponding to the current growth period is obtained; the matching degree between each corrected sub-dynamic feature corresponding to each corrected temporal feature and the corresponding resource demand threshold of the growth period is calculated; the matching degree of each corrected sub-dynamic feature corresponding to each corrected temporal feature is integrated to generate a growth period adaptation feature vector; and the dynamic fusion features are obtained based on the growth period adaptation feature vector.

[0010] In one optional implementation, dynamic fusion features are obtained based on the reproductive period adaptation feature vector, including: identifying environmental monitoring data and detecting whether the current environment meets a preset emergency threshold; if the current environment does not meet the preset emergency threshold, extracting the sub-dynamic feature mutation signal corresponding to the extreme time period from the dynamic input tensor; quantizing the sub-dynamic feature mutation signal into an emergency feature value; and performing an emergency correction on the reproductive period adaptation feature vector based on the emergency feature value to obtain dynamic fusion features.

[0011] In one optional implementation, based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data, the dynamic change trend of soil organic carbon in the target area is predicted, including: extracting features from crop monitoring data to obtain biomass characteristics, root exudate concentration, and current growth stage; extracting features from topographic monitoring data to obtain topographic slope gradient and aspect; decomposing the soil and water carrying capacity index to obtain water carrying capacity sub-index and nutrient carrying capacity sub-index; and extracting features from soil monitoring data to obtain basic soil organic carbon content, microbial characteristics, and... Physicochemical characteristics; feature extraction is performed on environmental monitoring data to obtain temperature, moisture, dynamic, and radiation characteristics; based on the extracted features, carbon input and output items corresponding to the target area are determined; based on the carbon input and output items, the initial soil organic carbon change corresponding to the target area is calculated; the water and soil carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, environmental monitoring data, and the initial soil organic carbon change are input into a preset intelligent prediction model for dynamic changes in soil organic carbon; the preset intelligent prediction model for dynamic changes in soil organic carbon outputs the dynamic change trend of soil organic carbon.

[0012] In one optional implementation, a pre-defined intelligent prediction model for dynamic changes in soil organic carbon outputs the dynamic change trend of soil organic carbon, including: the pre-defined intelligent prediction model for dynamic changes in soil organic carbon extracts features from soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data to obtain root-microbe interaction features, microbial activity thermodynamics, topographic slope gradient, aridity index, and aspect correction coefficient; inputs the root-microbe interaction features, microbial activity thermodynamics, topographic slope gradient, aridity index, aspect correction coefficient, and initial soil organic carbon change into a pre-defined convolutional network in the pre-defined intelligent prediction model for dynamic changes in soil organic carbon; the pre-defined convolutional network performs convolution processing on the input features and outputs a carbon fixation potential feature map and a carbon loss risk feature map corresponding to the target area; based on the carbon fixation potential feature map and the carbon loss risk feature map, the dynamic change trend of soil organic carbon is output.

[0013] In one optional implementation, the preset convolutional network includes two layers. The target region includes multiple grid cells. The preset convolutional network performs convolution processing on the input features and outputs a carbon fixation potential feature map and a carbon loss risk feature map corresponding to the target region. This includes: for each target grid cell, the first layer of the preset convolutional network extracts convolutional features based on a first attention convolutional kernel of a first preset size, focusing on root-microbe interaction features, microbial activity thermodynamics, and initial soil organic carbon changes corresponding to the target grid cell and its corresponding neighboring grid cells, to obtain the features for each grid cell. The sub-carbon fixation potential feature map corresponding to the grid is generated; a sliding window traverses the target area and stitches together the sub-carbon fixation potential feature maps to obtain a carbon fixation potential feature map; for each target grid cell, a second convolutional network in the preset convolutional network is used to extract features of the terrain slope gradient, aridity index, and aspect correction coefficient corresponding to the target grid cell and its corresponding neighboring grid cells based on a second attention convolutional kernel of a second preset size, to obtain the unidirectional feature value corresponding to the target grid cell; a sliding window traverses the target area and stitches together the unidirectional feature values ​​to obtain a carbon loss risk feature map.

[0014] Secondly, the present invention provides an intelligent prediction device for the dynamic change trend of soil organic carbon, the device comprising: The acquisition module is used to acquire soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area. The first determination module is used to extract features from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area. The prediction module is used to predict the dynamic change trend of soil organic carbon in the target area based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data. The second determination module is used to determine the target measures corresponding to the target area based on the dynamic change trend of soil organic carbon.

[0015] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent prediction method for dynamic change trend of soil organic carbon as described in the first aspect or any corresponding embodiment.

[0016] The intelligent prediction method, device, and electronic equipment for dynamic changes in soil organic carbon (SOC) provided in this application collect soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data from the target area. The multi-source data covers all elements of soil, crops, environment, and facilities, providing a comprehensive data foundation for subsequent analysis and avoiding assessment bias caused by single data sources. Key features (such as soil texture, crop water requirement, and topographic slope) are extracted from the multi-source data. The water and soil carrying capacity index is calculated by fusing static and dynamic features (such as Pearson correlation coefficient and random forest model) to quantify the regional water and soil resource support capacity. Complex data is transformed into a quantifiable index, intuitively reflecting the basic potential and dynamic adaptability of regional water and soil resources, providing constraints for subsequent carbon cycle analysis (such as the limitation of water / nutrients on carbon fixation). Next, based on the water and soil carrying capacity index and multi-source data, carbon fixation / loss features (such as root-microbe interactions and the influence of topographic slope) are extracted using a convolutional network. Combined with a carbon input (plant residues, microbial carbon fixation) and output (decomposition, erosion) model, the spatiotemporal change trend of SOC is predicted. By integrating mechanistic models and machine learning, this approach accurately captures the driving factors of SOC dynamics (such as crop growth period and topographic risk), enabling refined predictions from single-point monitoring to regional trends, thus providing a scientific basis for carbon management. Based on SOC dynamic trends (such as growth areas and loss areas), targeted measures are formulated (such as increasing organic fertilizer application in loss areas, converting farmland back to forest in steep slope areas, and optimizing irrigation during the growth period). Trend predictions are transformed into implementable management plans, achieving precise governance through "problem identification and measure matching," improving soil carbon sequestration capacity and water and soil resource utilization efficiency, and contributing to sustainable agricultural development. This application achieves full-chain optimization from data to decision-making through a closed-loop process of "data acquisition - index assessment - trend prediction - measure formulation": ensuring the scientific rigor of the assessment (supported by multi-source data) and improving the precision of management (targeted measures), ultimately helping target areas balance water and soil resource utilization and soil carbon pool stability, achieving a win-win situation for both ecological and agricultural benefits. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating another intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating another intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention. Figure 4 This is a structural block diagram of an intelligent prediction device for dynamic changes in soil organic carbon according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0020] It should be noted that the method for intelligent prediction of dynamic changes in soil organic carbon provided in this application can be executed by a device for intelligent prediction of dynamic changes in soil organic carbon. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. The electronic device can be a server or a terminal. In this application embodiment, the server can be a single server or a server cluster composed of multiple servers. The terminal in this application embodiment can be a smartphone, personal computer, tablet computer, wearable device, or other intelligent hardware device such as a smart robot. The following method embodiments will use an electronic device as an example for illustration.

[0021] According to an embodiment of the present invention, an embodiment of an intelligent prediction method for dynamic changes in soil organic carbon is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] This embodiment provides an intelligent prediction method for the dynamic change trend of soil organic carbon, which can be used in the aforementioned electronic equipment. Figure 1 This is a flowchart of an intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area.

[0023] Specifically, electronic devices can divide a target area (such as 1,000 acres of farmland) into collection units using a 10-meter × 10-meter grid, and then collect data from each collection unit.

[0024] Soil monitoring data was obtained through a combination of ground-based sensor networks (one monitoring point every 50 meters) and laboratory sampling. This included physicochemical indicators such as soil texture (sand / clay ratio), pH (4.5-8.5), bulk density (1.0-1.6 g / cm³), and field capacity (20%-40%); nutrient indicators such as organic matter content (5-50 g / kg), total nitrogen (0.5-3 g / kg), and available phosphorus (5-30 mg / kg); and biological indicators such as microbial activity (PLFA value, 5-50 nmol / g) and soil respiration rate (1-10 mg CO₂ / kg•h).

[0025] The terrain monitoring data is obtained by fusing satellite remote sensing DEM (30-meter resolution) with UAV lidar (LiDAR). It includes basic terrain features such as slope (0°-30°), aspect (coded in 8 directions), and elevation (relative elevation difference ≤50 meters); and micro-topography features such as surface relief (≤5 meters) and slope length (10-100 meters), which are used to assess the risk of soil erosion.

[0026] Crop monitoring data were obtained by combining UAV multispectral remote sensing (every 5 days) with field surveys. This included growth indicators such as plant height (50-200 cm), leaf area index (LAI, 1-6), and biomass (5-20 t / ha aboveground, 1-5 t / ha underground); phenological stages, such as current growth stage (e.g., wheat jointing stage, maize grain-filling stage), and expected maturity date (±3 days error); and stress indicators such as vegetation water index (VWI) and nitrogen nutrient index (NNI), reflecting factors limiting crop growth.

[0027] The facility monitoring data, targeting farmland irrigation and carbon management facilities, is collected in real time via IoT sensors. This includes irrigation and drainage facilities, such as irrigation flow rate (5-50 m³ / h), drainage rate (10-100 mm / d), and pipeline pressure (0.2-0.6 MPa); and carbon sink facilities, such as the operating status of organic fertilizer application equipment (fertilizer application error ±5%) and the operating area of ​​straw returning machines (coverage rate ≥80%).

[0028] Environmental monitoring data is acquired using regional meteorological stations (1 station per 1000 mu) and mobile monitoring vehicles. This includes meteorological indicators such as daily average temperature (-5-35℃), rainfall (0-200mm / month), wind speed (0-10m / s), and relative humidity (30%-90%); and atmospheric indicators such as CO2 concentration (400-600ppm) and light intensity (5000-100000lux), used to assess carbon exchange potential.

[0029] Step S102: Extract features from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area.

[0030] Specifically, electronic devices can extract core features from multi-source data and convert them to the 0-1 range by "positive normalization of high-optimal indicators and negative normalization of low-optimal indicators." Positive indicators (higher values ​​are more favorable) include field water holding capacity, organic matter content, LAI (Lower Irrigation Index), and irrigation guarantee rate; negative indicators (lower values ​​are more favorable) include slope, soil bulk density, and drought stress index (VWI < 0.5 indicates stress). Then, based on the determination of each positive and negative indicator, the water carrying capacity, nutrient carrying capacity, topographic adaptability, and plant adaptability of the target area are determined.

[0031] Then, the Analytic Hierarchy Process (AHP) is used to determine the weights corresponding to water carrying capacity, nutrient carrying capacity, topographic adaptability, and crop adaptability for the target area. Based on these weights, the soil and water carrying capacity index (WCCI) for the target area is determined. Example calculation: For a grid cell with water carrying capacity of 0.75 (weight 0.4), nutrient carrying capacity of 0.6 (weight 0.25), topographic adaptability of 0.8 (weight 0.15), and crop adaptability of 0.7 (weight 0.2), the WCCI is calculated as follows: WCCI = 0.75 × 0.4 + 0.6 × 0.25 + 0.8 × 0.15 + 0.7 × 0.2 = 0.71.

[0032] Grading criteria: WCCI > 0.7 is "high carrying capacity area", 0.5-0.7 is "medium carrying capacity area", and < 0.5 is "low carrying capacity area". Then, a grid-level WCCI distribution heat map is generated, such as "high carrying capacity area is concentrated in the center of the field (slope < 5°, sufficient irrigation), and low carrying capacity area is distributed on the edge of steep slope (slope > 15°, poor soil)".

[0033] This step will be explained in detail below.

[0034] Step S103: Based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data, predict the dynamic change trend of soil organic carbon in the target area.

[0035] Specifically, electronic devices can input water and soil carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data into a preset intelligent prediction model for dynamic changes in soil organic carbon, thereby predicting the dynamic changes in soil organic carbon in the target area.

[0036] This step will be explained in detail below.

[0037] Step S104: Based on the dynamic change trend of soil organic carbon, determine the target measures corresponding to the target area.

[0038] Specifically, electronic devices can determine target measures for target areas based on preset principles and the dynamic trends of soil organic carbon changes. These preset principles can include at least one of the following: targeting principle, economic principle, and feasibility principle. The targeting principle refers to combining WCCI grading and SOC trends, strengthening carbon sequestration in high-carrying-capacity areas, and first improving basic carrying capacity in low-carrying-capacity areas. The economic principle means that the cost of the measures should be less than or equal to the expected carbon sequestration benefits (e.g., the cost of organic fertilizer application should be less than or equal to the carbon trading benefits). The feasibility principle means implementing measures based on existing facilities (such as irrigation systems and agricultural machinery) to avoid additional investment.

[0039] For example, for high-carrying-capacity carbon sink hotspots (WCCI > 0.7, SOC annual increase > 0.5 g / kg): optimization measures include promoting the rotation of leguminous crops (increasing residual carbon input by 30%), increasing the application of biochar (200 kg per acre, increasing carbon sequestration rate by 20%), and installing intelligent irrigation systems to maintain soil moisture at 60%-80% of field capacity (promoting microbial carbon sequestration).

[0040] For the medium-bearing capacity-stable zone (WCCI 0.5-0.7, SOC annual variation ±0.2g / kg): improvement measures include straw crushing and returning to the field + deep plowing (increasing the return rate to 90%), no-till planting (reducing soil disturbance and decomposition loss); soil samples are collected every quarter, and the amount of organic fertilizer applied is dynamically adjusted.

[0041] For low-carrying-capacity carbon source areas (WCCI < 0.5, SOC annual reduction > 0.2 g / kg): emergency measures include building terraces (in areas with a slope > 15°) and planting nitrogen-fixing plants (such as alfalfa to increase soil nitrogen); long-term measures include soil maturation (applying 500 kg of peat soil per acre, which will increase organic matter to more than 15 g / kg in 2-3 years).

[0042] The intelligent prediction method for dynamic changes in soil organic carbon provided in this application collects soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data for the target area. Multi-source data covers all elements of soil, crops, environment, and facilities, providing a comprehensive data foundation for subsequent analysis and avoiding assessment bias caused by single data sources. Key features are extracted from the multi-source data to calculate the soil and water carrying capacity index, quantifying the regional soil and water resource support capacity. Complex data is transformed into a quantifiable index, intuitively reflecting the basic potential and dynamic adaptability of regional soil and water resources, providing constraints for subsequent carbon cycle analysis. Based on the soil and water carrying capacity index and multi-source data, the dynamic change trend of soil organic carbon in the target area is predicted. This achieves refined prediction from single-point monitoring to regional trends, providing a scientific basis for carbon management. Finally, based on the dynamic change trend of soil organic carbon, target measures corresponding to the target area are determined (such as increasing organic fertilizer application in erosion areas, converting farmland back to forest in steep slope areas, and optimizing irrigation during the growing season). The trend prediction is transformed into an implementable management plan, achieving precise governance through "problem identification and measure matching," improving soil carbon sequestration capacity and soil and water resource utilization efficiency, and contributing to sustainable agricultural development. The above method, through a closed-loop process of "data acquisition - index assessment - trend prediction - measure formulation", achieves full-chain optimization from data to decision-making: it not only ensures the scientific nature of the assessment (supported by multi-source data) but also improves the accuracy of management (targeted measures), ultimately helping the target area to balance the utilization of water and soil resources and the stability of the soil carbon pool, and achieve a win-win situation for ecological and agricultural benefits.

[0043] This embodiment provides an intelligent prediction method for the dynamic change trend of soil organic carbon, which can be used in the aforementioned electronic equipment. Figure 2 This is a flowchart of an intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area.

[0044] For details on this step, please refer to the above description of step S101; it will not be repeated here.

[0045] Step S202 involves extracting features from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area.

[0046] Specifically, step S202 above may include the following steps: Step S2021: Determine static data and time-series dynamic data from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data.

[0047] Specifically, electronic equipment can distinguish between static and dynamic data from multi-source monitoring data. Static data includes: soil texture (sandy / clay / loam), parent material type (e.g., loess / red soil), and soil layer thickness (cm) from soil monitoring data; slope (°), aspect (8 directions), and elevation (m) from topographic monitoring data; and irrigation and drainage facility type (fixed / mobile), pipeline distribution density (m / acre), and facility construction year from facility monitoring data. (Note: Static data is less common in crop and environmental monitoring data; dynamic data is the primary component.)

[0048] The time-series dynamic data include: soil moisture content, pH value (short-term fluctuations), and microbial activity (seasonal changes) from soil monitoring data; plant height, leaf area index, and phenological stage from crop monitoring data; and daily average temperature, rainfall, and wind speed from environmental monitoring data.

[0049] Step S2022: Divide the target area into at least one grid cell.

[0050] Specifically, the electronic device can divide the target area into 10m x 10m square grid units based on data accuracy and research needs. If the target area is 1,000 mu (approximately 667,000 square meters), then the number of grid units would be approximately 667,000 ÷ 100 ≈ 6,670. Then, the electronic device can identify each grid unit using a row-column identifier (e.g., grid (3,5) represents the 3rd row and 5th column), ensuring a unique spatial correspondence.

[0051] By calibrating with GPS coordinates, each grid cell is bound to a real plot of land: the top left grid coordinates are set as the origin (X=0, Y=0), the positive direction of the X-axis is to the right, and the positive direction of the Y-axis is downward, generating a grid-coordinate mapping table to support subsequent spatial data queries (such as "extracting grid data within the range of X=100-200 meters and Y=50-150 meters").

[0052] Step S2023: For each grid cell, target static data is selected from the static data.

[0053] Specifically, electronic devices can filter static features that directly affect soil and water bearing capacity from the static data based on their correlation with soil and water bearing capacity, eliminate irrelevant variables, and obtain the target static data. The correlation between the static data and soil and water bearing capacity can be determined based on expert knowledge and literature research.

[0054] For example, for each grid cell, the electronic device can extract four core features from static data as target static data. These target static data include: soil dimension: soil texture (coded: 1=sandy, 2=clay, 3=loam), affecting water and fertilizer retention capacity; soil thickness (graded: 1=thin <30cm, 2=medium 30-60cm, 3=thick >60cm), affecting root growth space; topography dimension: slope grade (coded: 1=gentle slope <5°, 2=medium slope 5°-15°, 3=steep slope >15°), affecting soil and water loss risk; and facility dimension: irrigation and drainage facility grade (coded: 1-5, 5=100% pipe network coverage + automatic control), affecting water regulation capacity.

[0055] Step S2024: Construct a static vector based on the target static data.

[0056] Specifically, electronic devices can convert the selected target static data into numerical feature values, forming a single-grid static vector: classification feature encoding: such as soil texture "loam" corresponding to 3, slope "medium slope" corresponding to 2; graded feature quantification: such as irrigation and drainage facility level directly using a value of 1-5 (the higher the value, the more complete the facility).

[0057] Each grid cell has a static vector that is a 1×4 dimension row vector, in the format: [soil texture, soil thickness, slope grade, irrigation and drainage facility grade].

[0058] For example, if grid (3,5) is “loam (3), thick soil layer (3), gentle slope (1), level 5 facility (5)”, then the static vector is [3,3,1,5]; if grid (10,8) is “sandy soil (1), thin soil layer (1), steep slope (3), level 2 facility (2)”, then the static vector is [1,1,3,2].

[0059] Step S2025: Concatenate the static vectors corresponding to each grid cell to generate a static matrix.

[0060] Specifically, the electronic device can vertically concatenate the static vectors of all grid cells in spatial order (row-major) to form a two-dimensional static matrix. Matrix dimensions: [number of grids, number of features], where the number of features = 4 (soil texture, soil thickness, slope grade, irrigation and drainage facility grade). For example, if the target area contains 100 grids, the static matrix is ​​a 100×4 two-dimensional array, with each row corresponding to the static feature of one grid cell.

[0061] Then, the electronic equipment checks the integrity and consistency of the static matrix: missing value handling: if a grid data is missing, it is filled with the mean of the neighboring 3×3 grids (e.g., if the slope level is missing, the median of the surrounding 9 grids is taken); outlier correction: if the irrigation and drainage facility level is "6" (out of the range of 1-5), it is corrected to the maximum value of 5.

[0062] Step S2026: Extract features from the static matrix to obtain static fusion features.

[0063] Specifically, step S2026 above may include the following steps: Step a1: The correlation between each sub-static feature in the static matrix and the soil and water bearing capacity is calculated using the Pearson correlation coefficient.

[0064] Specifically, for each sub-static feature in the static matrix (such as soil texture, slope, and irrigation and drainage facility level), the electronic device can extract data from the preset "static feature-soil and water bearing capacity level matching sample" and then use the Pearson correlation coefficient to calculate the linear correlation between each sub-static feature and the soil and water bearing capacity level.

[0065] The formula is: ; in, Let i be the i-th sub-static eigenvalue. This is the mean of the feature; To correspond to the soil and water bearing capacity level, This represents the average of the grades.

[0066] Step a2: Convert the correlation degree into initial feature weights.

[0067] Specifically, electronic devices can normalize the correlation degree, converting the correlation degree into initial feature weights.

[0068] Step a3: Input the static matrix and initial feature weights into the preset random forest model.

[0069] The default number of decision trees in the random forest model is 100 (to balance accuracy and computational efficiency). Objective function: Using soil and water carrying capacity levels as labels, train the model's predictive ability; The input data includes: Feature matrix: static matrix (e.g., [500,3], 500 grids, 3 sub-static features); Weight parameters: initial feature weights obtained in step a2.

[0070] The electronic device packages the static matrix and initial weights as input to a pre-defined random forest model, for example: Example of a static matrix row: [sandy soil (1), 10° slope, 3-level irrigation and drainage]; weight parameters: [0.26, 0.33, 0.41] (corresponding to soil texture, slope, irrigation and drainage).

[0071] Step a4: The preset random forest model binds the initial weights of features to the column vectors of the static matrix to form a sub-static feature-initial weight mapping table.

[0072] Specifically, the default random forest model binds the column vectors (sub-static features) of the static matrix to the initial weights, forming a key-value pair mapping table, a sub-static feature-initial weight mapping table. This is used for subsequent feature correction (clarifying the weight corresponding to each feature) and model interpretability.

[0073] Step a5: For each grid cell in the static matrix, based on the sub-static feature-initial weight mapping table, modify each sub-static feature in the static matrix, and obtain the weighted static matrix based on the modified sub-static features.

[0074] Specifically, the electronic device can correct the sub-static feature value for each grid cell in the static matrix according to the initial weight in the sub-static feature-initial weight mapping table.

[0075] For categorical features (such as soil texture and irrigation / drainage level), first encode them as numerical values ​​(e.g., sandy soil = 1, loam = 2), then multiply by the weight. For example, a level 3 irrigation / drainage facility × 0.41 = 1.23; for continuous features (such as slope), directly use the original value × weight, for example, a 10° slope × 0.33 = 3.3.

[0076] Next, the electronic device reassembles the eigenvalues ​​of all the modified sub-static features of the grid according to the original structure to form a weighted static matrix.

[0077] Example (for a certain grid): Original static vector: [1 (sand), 10 (slope), 3 (irrigation and drainage)]; After correction: [1×0.26=0.26, 10×0.33=3.3, 3×0.41=1.23]. Consistent with the static matrix (e.g., [500,3]), but the eigenvalues ​​now incorporate weight information.

[0078] Step a6: Extract features from the weighted static matrix to obtain the splitting features of each target.

[0079] Specifically, each decision tree in the pre-defined random forest model splits the weighted static matrix based on "Gini impurity". The splitting rule is to select the feature that minimizes impurity as the splitting point (e.g., "irrigation and drainage facility correction value > 1.0" can significantly distinguish between high and low carrying capacity).

[0080] The electronic device counts the top k features (k is usually 3-5) that are selected as split points most frequently in all decision trees. For example, the number of splits is: irrigation and drainage facilities (85 times) > slope (62 times) > soil texture (38 times) → the target split features are the top 3.

[0081] Step a7: For each target splitting feature, calculate the first weighting coefficient corresponding to the target splitting feature based on the initial feature weights corresponding to the target splitting feature.

[0082] Specifically, the electronic device can retrieve the initial weights of the target splitting features from the sub-static feature-initial weight mapping table. For example, irrigation and drainage (0.41), slope (0.33), and soil texture (0.26).

[0083] The electronic device then normalizes the initial weights again (ensuring the sum is 1), resulting in the first weighting coefficient: ; in, The initial weights of the features corresponding to the target splitting features.

[0084] Step a8: Based on the first weighting coefficient corresponding to each target splitting feature, fuse each target splitting feature to obtain static fused features.

[0085] Specifically, for each grid cell, the electronic device can sum the corrected value of its target splitting feature after multiplying it by the first weighting coefficient.

[0086] formula: ; Among them, the characteristic values ​​of the corrected target splitting features are: irrigation and drainage 1.23, slope 3.3, soil texture 0.26. The fusion value is: 1.23×0.41+3.3×0.33+0.26×0.26≈0.50+1.09+0.07=1.66.

[0087] Then, the electronic device maps the fused values ​​of all grid cells to the 0-1 range (e.g., 1.66→0.66, based on the regional maximum value of 2.5). The grid-level static fusion feature matrix (e.g., [500,1]) shows that the higher the value, the stronger the static soil and water carrying capacity of the grid.

[0088] Step S2027: Generate a dynamic input tensor based on the duration, dimension, and number of raster cells corresponding to the time series dynamic data.

[0089] Specifically, electronic devices can construct a three-dimensional dynamic input tensor based on data duration, feature dimension, and number of grids: Time dimension (T): Take the data of the most recent preset duration (e.g., 30 days) (covering changes in the key growth period of crops), and interpolate it to 30 time steps according to the sampling frequency; Feature dimension (F): Filter a preset number (e.g., 8) of core dynamic features (e.g., soil moisture content, rainfall, LAI, etc.); Spatial dimension (N): Number of grids in the target area (e.g., 500).

[0090] Dynamic input tensor format: dimensions are [N,T,F] (e.g., 500 grids × 30 days × 8 features).

[0091] Next, the electronic device can perform data preprocessing on the dynamic input tensor. Data preprocessing may include: missing value imputation: using forward imputation (e.g., if the rainfall data for a certain day is missing, use the data from the previous day); normalization: mapping all feature values ​​to the 0-1 interval (e.g., soil moisture content 20%-40% → 0-1).

[0092] Step S2028: Extract features from the dynamic input tensor to obtain dynamic fusion features.

[0093] Specifically, step S2028 above may include the following steps: Step b1: Input the dynamic input tensor into the input gate of the preset temporal processing model.

[0094] Specifically, the electronic device can input a dynamic input tensor [N,T,F] into the input gate of a preset time-series processing model to identify key features at each time step (such as a sudden increase in soil moisture content on an irrigation day).

[0095] In step b2, the input gate identifies the dynamic input tensor, filters valid information using the sigmoid function, and transmits the valid information to the forget gate.

[0096] Specifically, the input gate filters valid information using the sigmoid function (outputting 0-1): ; in, Features at time t The state was hidden in the previous moment. Here is the weight matrix for the input gate, used to quantify the importance of the preceding states and the current input. This is the bias term for the input gate, used to fine-tune the output. The output of the input gate is between 0 and 1. The closer the value is to 1, the more important the information is and it should be retained; the closer it is to 0, the more ignorable the information is. For example, the rainfall feature on a rainy day is output as 0.9 (retained), and the rainfall feature on a dry day is output as 0.1 (weakened).

[0097] In step b3, the forget gate identifies valid information, discards invalid information, and transmits the remaining valid information to the output gate.

[0098] Specifically, the forget gate discards redundant information (such as outdated soil temperature data): ; in, The output of the forget gate (between 0 and 1) indicates that the historical information should be forgotten, while the value is closer to 1. This is the weight matrix for the forgetting gate, used to determine the timeliness of historical information. This is the bias term for the forget gate. For example, retain the soil moisture content data for the most recent 7 days (weight 0.8) and forget the data from 15 days ago (weight 0.1).

[0099] Step b4: The output gate integrates the output information of the forget gate and outputs the regular timing characteristics of each grid cell.

[0100] Specifically, the output gate integrates the output information of the forget gate and outputs the regular timing characteristics of each grid cell.

[0101] ; ; in, The output of the output gate (between 0 and 1) determines which information in the cell state will be output to the hidden state. This is the weight matrix of the output gate, which controls the filtering of output information. The hidden state at the current moment (i.e., the core of the regular temporal feature vector) contains key temporal information that has been filtered and integrated. The tanh (cell state) (the "memory unit" of the LSTM) stores historical and current information after being retained by the input gate and filtered by the forget gate. ): Compresses cell state values ​​to between -1 and 1 to avoid excessively large values.

[0102] Step b5: Obtain the current growth stage of the crop.

[0103] Specifically, the electronic device can receive the current growth stage of the crop from user input. The electronic device can also identify the current growth stage of the crop through crop monitoring data.

[0104] Step b6: Based on the current reproductive period, modify each conventional time series feature to obtain the modified time series feature.

[0105] Specifically, electronic devices can extract sub-dynamic features (such as soil moisture, rainfall, temperature, and air humidity) from conventional time-series features and match them with the current growth stage-dynamic weight mapping table to determine the priority of each sub-dynamic feature in the conventional time-series features. For example, during the wheat jointing stage: soil moisture (high) > temperature (medium) > rainfall (medium) > wind speed (low); during the corn grain-filling stage: air humidity (high) > soil temperature (medium) > irrigation amount (medium).

[0106] Then, the electronic device can convert the priorities corresponding to each sub-dynamic feature into dynamic weights (summing up to 1). For example, during the wheat jointing stage, the dynamic weights are: soil moisture 0.4, temperature 0.25, rainfall 0.25, and wind speed 0.1; during the corn grain-filling stage, the dynamic weights are: soil temperature 0.35, air humidity 0.3, soil moisture 0.25, and wind speed 0.1.

[0107] Then, the electronic device corrects each sub-dynamic feature in the conventional time series feature based on dynamic weights to obtain the corrected time series feature.

[0108] Step b7: Obtain the resource demand threshold for the reproductive period corresponding to the modified sub-dynamic features of each modified time-series feature corresponding to the current reproductive period.

[0109] Specifically, the electronic device can receive the resource demand thresholds for each modified time-series feature corresponding to the current reproductive period, as input by the user.

[0110] Step b8: Calculate the matching degree between each modified sub-dynamic feature corresponding to each modified time series feature and the corresponding resource demand threshold during the reproductive period.

[0111] Specifically, the electronic device can calculate the matching degree between each modified sub-dynamic feature corresponding to each modified time-series feature and the corresponding resource demand threshold during the reproductive period.

[0112] The formula is: (The upper limit is 1.2 when the threshold is exceeded to avoid interference from extreme values).

[0113] For example, the soil moisture correction value is 0.28 ÷ threshold 0.6 ≈ 0.47 (matching degree 47%, insufficient supply).

[0114] Step b9: Integrate the matching degree of each modified sub-dynamic feature corresponding to each modified time-series feature to generate a reproductive period-adapted feature vector.

[0115] Specifically, the matching degrees corresponding to the dynamic features of each modifier are concatenated in their original order to form a vector. For example: (a grid cell during the wheat jointing stage): [0.47 (soil moisture), 0.86 (temperature), 1.2 (rainfall), 0.67 (wind speed)]; Vector meaning: Each value reflects the supporting capacity of the corresponding resource for the current growth period (the higher the value, the better the fit).

[0116] Step b10: Based on the reproductive period adaptation feature vector, obtain the dynamic fusion feature.

[0117] Specifically, step b10 above may include the following steps: Step b101: Identify the environmental monitoring data and detect whether the current environment meets the preset emergency threshold.

[0118] Specifically, electronic devices can identify extreme conditions in environmental monitoring data in real time and detect whether the current environment meets preset emergency thresholds. Examples of preset emergency thresholds: Drought: Soil moisture matching degree < 0.3 for 7 consecutive days; Rainstorm: 1-hour rainfall correction value > 0.8 (normalized).

[0119] Step b102: If the current environment does not meet the preset emergency threshold, then extract the sub-dynamic feature mutation signal corresponding to the extreme time period from the dynamic input tensor.

[0120] Specifically, if the current environment meets the preset emergency threshold, the electronic device extracts the sub-dynamic feature mutation signal corresponding to the extreme period from the dynamic input tensor. For example, in a drought scenario: the daily soil moisture decay rate (e.g., 0.05 / day); in a rainstorm scenario: the hourly increase in rainfall (e.g., 0.6 / hour).

[0121] Step b103: Quantize the sub-dynamic feature mutation signal into emergency feature value.

[0122] Specifically, electronic devices convert sudden change signals into emergency characteristic values ​​of 0-1 (the higher the value, the stronger the extreme). Drought emergency value = daily soil moisture decay rate × 20 (amplification factor), such as 0.05 × 20 = 1.0 (upper limit 1.0); Rainstorm emergency value = hourly increase in rainfall × 1.5, such as 0.6 × 1.5 = 0.9.

[0123] Step b104: Based on the emergency feature value, perform emergency correction on the reproductive period adaptation feature vector to obtain dynamic fusion features.

[0124] Specifically, the characteristics of reproductive period adaptation and emergency features are integrated: Dynamic fusion feature value = 0.8 × mean of reproductive period adaptation features + 0.2 × emergency feature value. (Weight allocation: regular adaptation accounts for 80%, emergency accounts for 20%).

[0125] Example: Mean value of reproductive period adaptation characteristics = (0.47 + 0.86 + 1.2 + 0.67) ÷ 4 = 0.8, drought emergency characteristic value = 1.0, dynamic fusion characteristic value = 0.8 × 0.8 + 0.2 × 1.0 = 0.64 + 0.2 = 0.84.

[0126] Raster-level dynamic fusion feature matrix (each raster corresponds to one fusion value, reflecting the comprehensive adaptability of dynamic resources and crop needs).

[0127] Step S2029: The static fusion features and dynamic fusion features are fused together to generate the target fusion features.

[0128] Specifically, the electronic device can detect whether the spatial dimensions of the static fusion feature and the dynamic fusion feature are consistent. If they are inconsistent, the static fusion feature and the dynamic fusion feature are unified in spatial dimension. Then, the electronic device maps the static fusion feature (range 0-1) and the dynamic fusion feature (range 0-1) to the same numerical range to eliminate the difference in dimensions.

[0129] Next, the electronic device can assign fusion weights based on the degree of influence of static and dynamic fusion characteristics on soil and water bearing capacity. For example, static fusion characteristics (such as soil texture and topography) determine the foundation bearing capacity (weight 0.4); dynamic fusion characteristics (such as growth period adaptability and emergency risk) reflect short-term fluctuations (weight 0.6).

[0130] For each grid cell, the electronic device performs a weighted summation based on the weights corresponding to the static fusion features and the dynamic fusion features respectively: Target fusion feature value = Static fusion feature value × Weight corresponding to the static fusion feature + Dynamic fusion feature value × Weight corresponding to the dynamic fusion feature.

[0131] Finally, the electronic device integrates the target fusion feature values ​​of all grids to form a target fusion feature matrix with dimensions of [number of grids, 1] (e.g., 500 grids correspond to 500 fusion values). The matrix retains spatial location information (grid coordinates and values ​​correspond one-to-one).

[0132] Step S20210: Input the target fusion features into the preset soil and water bearing capacity assessment model and output the soil and water bearing capacity index corresponding to the target area.

[0133] Among them, the preset soil and water bearing capacity assessment model can adopt the gradient boosting tree (GBDT) model, which has been trained using historical "target fusion features - soil and water bearing capacity level" samples.

[0134] The input to the preset soil and water bearing capacity assessment model is the target fusion feature matrix (the fusion value of each grid cell), and the output of the preset soil and water bearing capacity assessment model is the soil and water bearing capacity index of each grid cell (range 0-1, the higher the value, the stronger the bearing capacity).

[0135] Specifically, the electronic device can convert the target fusion feature matrix into an input format that can be recognized by the preset soil and water carrying capacity assessment model (such as a two-dimensional array [N,1], where N is the number of grids), and then load the model pre-training parameters (such as the number of decision trees, learning rate, etc.) to ensure the consistency of the assessment.

[0136] The pre-set water and soil carrying capacity assessment model infers the target fusion feature value of each grid, combines static foundation (such as soil fertility) and dynamic adaptability (such as water supply capacity during the current growing season), comprehensively judges the resource support capacity of the grid, and outputs the corresponding water and soil carrying capacity index.

[0137] For example, a raster with a target fusion feature value of 0.58 is preset to output a soil and water bearing capacity index of 0.62 (medium bearing capacity) from the soil and water bearing capacity assessment model.

[0138] Finally, the electronic device can remove outliers with indices >1 or <0 (using neighborhood mean replacement) and perform Kriging interpolation on sparse data areas to generate a continuous spatial distribution map of the soil and water bearing capacity index. Then, the continuous spatial distribution map of the soil and water bearing capacity index is displayed in three levels according to the index value: high bearing capacity: >0.7; medium bearing capacity: 0.3-0.7; low bearing capacity: <0.3.

[0139] Step S203: Based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data, predict the dynamic change trend of soil organic carbon in the target area.

[0140] Please refer to the above description of step S103 for details on this step, which will not be repeated here.

[0141] Step S204: Based on the dynamic change trend of soil organic carbon, determine the target measures corresponding to the target area.

[0142] Please refer to the above description of step S104 for details on this step, which will not be repeated here.

[0143] The intelligent prediction method for dynamic changes in soil organic carbon provided in this application distinguishes between static data and time-series dynamic data from multi-source data. It clarifies the temporal attributes of the data, laying the foundation for subsequent static feature stability modeling and dynamic feature time-series analysis, and avoiding feature distortion caused by data aliasing. The target area is divided into several raster units to capture spatial heterogeneity and improve the spatial resolution of the assessment. Target static data related to soil and water bearing capacity are selected for each raster and integrated into a static vector. Key static features are focused on, reducing redundant information, and the vector form facilitates subsequent matrix operations and model input. The static vectors of all raster units are concatenated to form a static matrix. This integrates the scattered raster features into a unified matrix, achieving a structured expression of static features at the regional scale, facilitating batch processing. The Pearson correlation coefficient is used to calculate the correlation between static features and soil and water bearing capacity, and the correlation is converted into initial weights. This quantifies feature importance, avoids interference from irrelevant features, and makes the initial weights more closely reflect actual influence patterns. The static matrix and initial weights are input into a random forest model to generate a sub-static feature-initial weight mapping table. Machine learning is used to optimize weight allocation, making feature importance more consistent with the inherent patterns of the data and improving weight reliability. The static features of each grid are corrected based on the sub-static feature-initial weight mapping table to obtain a weighted static matrix. The contribution of highly important features is strengthened while secondary features are weakened, improving the relevance of static features. Key splitting features are extracted from the weighted static matrix and fused into static fusion features according to the first weighting coefficient. Through feature dimensionality reduction and fusion, core static information is preserved, data dimensionality is reduced, and model efficiency is improved.

[0144] Then, based on the duration, dimensions, and number of raster cells of the time-series dynamic data, a dynamic input tensor is constructed. The time-series dynamic data is structured while retaining three-dimensional information of time, features, and space, adapting to the input requirements of the time-series model. Effective dynamic information is filtered through an input gate, noise is discarded through a forget gate, and information is integrated through an output gate to obtain conventional time-series features. The temporal dependencies of the dynamic data are captured, and invalid information is filtered to improve the quality of the time-series features. Combined with the current crop growth stage (e.g., jointing stage), the time-series features are modified according to the growth stage-feature priority to obtain modified time-series features. The dynamic features are adapted to the needs of the crop growth stage (e.g., strengthening soil moisture weight during the jointing stage), improving the matching degree between features and actual growth requirements. The matching degree is calculated and a growth stage-adapted feature vector is generated. The modified time-series features are compared with growth stage resource thresholds (e.g., suitable soil moisture), the matching degree is calculated, and they are integrated into an adaptation vector. The degree to which dynamic features meet crop needs is quantified, converting absolute values ​​into relative adaptability indicators, which better aligns with the carrying capacity assessment logic. Extreme environments are detected, mutation signals are extracted and quantified into emergency values, and the adaptation vector is modified to obtain dynamic fusion features. Incorporating the impact of extreme weather enhances the robustness of dynamic characteristics and avoids assessment bias caused by sudden environmental events.

[0145] Finally, static and dynamic features are integrated to generate target fusion features. Static fusion features (basic attributes) and dynamic fusion features (time-series adaptability) are weighted and fused. This integrates long-term stable attributes and short-term dynamic changes in the region to form a comprehensive feature representation, providing complete input for carrying capacity assessment. The target fusion features are input into a pre-defined model, outputting the regional soil and water carrying capacity index. Through multi-step feature optimization and fusion, a precise and quantified carrying capacity index is ultimately obtained, providing a reliable basis for subsequent carbon cycle analysis and management decisions.

[0146] This embodiment provides an intelligent prediction method for the dynamic change trend of soil organic carbon, which can be used in the aforementioned electronic equipment. Figure 3 This is a flowchart of an intelligent prediction method for dynamic changes in soil organic carbon according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area.

[0147] Please refer to the above description of step S201 for details on this step, which will not be repeated here.

[0148] Step S302: Extract features from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area; Please refer to the above description of step S202 for details on this step, which will not be repeated here.

[0149] Step S303: Based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data, predict the dynamic change trend of soil organic carbon in the target area.

[0150] Specifically, step S303 above may include the following steps: Step S3031: Extract features from crop monitoring data to obtain biomass characteristics, root exudate concentration, and current growth stage.

[0151] Specifically, electronic devices can extract three core features directly related to carbon input from crop monitoring data: biomass characteristics, root exudate concentration, and current growth stage. Biomass characteristics include: aboveground biomass (e.g., straw dry weight, kg / acre): retrieved via the vegetation index (NDVI) obtained from UAV remote sensing; and belowground biomass (e.g., root dry weight): estimated at 30% of aboveground biomass (based on crop physiological patterns). Root exudate concentration is retrieved via the hyperspectral red-edge index (mg / L), reflecting the amount of organic carbon released from the roots into the soil. The current growth stage, such as the grain-filling stage (peak biomass accumulation) and the maturity stage (preparation stage for crop residue return), influences the pace of carbon input.

[0152] Step S3032: Extract features from the terrain monitoring data to obtain the terrain slope gradient and terrain aspect.

[0153] Specifically, electronic devices can extract features from terrain monitoring data to extract key features that influence carbon migration, including terrain slope gradient and terrain aspect.

[0154] Among them, topographic slope gradient: graded by degree (e.g., 0-5°, 5-15°, >15°), the greater the slope, the higher the risk of carbon loss due to surface runoff. Topographic slope aspect: divided into sunny slopes (sufficient sunlight, high temperature → accelerated carbon decomposition), shady slopes (stable temperature and humidity → good carbon retention), and level slopes (no significant bias).

[0155] Step S3033: Decompose the soil and water bearing capacity index to obtain the water bearing capacity sub-index and the nutrient bearing capacity sub-index.

[0156] Specifically, the electronic device can decompose the soil and water carrying capacity index output in step S20210 into two constraining sub-indices: the water carrying capacity sub-index (W): ranging from 0 to 1, reflecting the soil moisture's ability to support microbial activity (e.g., W=0.7 indicates suitable water conditions). The nutrient carrying capacity sub-index (N): ranging from 0 to 1, reflecting the ability of nutrients such as nitrogen and phosphorus to ensure plant growth and microbial activity (e.g., N=0.8 indicates sufficient nutrients).

[0157] Step S3034: Feature extraction is performed on the soil monitoring data to obtain the basic soil organic carbon content, microbial characteristics, and physicochemical characteristics.

[0158] Specifically, electronic devices can extract features from soil monitoring data, extracting three types of features directly related to the carbon cycle: basic soil organic carbon (SOC) content, microbial characteristics, and physicochemical characteristics.

[0159] Among these, the basic soil organic carbon (SOC) content (unit: g / kg) serves as the background value for carbon cycling (e.g., 20 g / kg). Microbial characteristics include microbial activity (PLFA value, unit: nmol / g, reflecting community abundance); respiration rate (unit: mgCO2 / kg•h, reflecting carbon decomposition intensity). Physicochemical characteristics include: pH value (affecting the microbial living environment, such as neutral soil being more suitable); bulk density (unit: g / cm³, affecting carbon migration resistance, high bulk density indicates poor aeration).

[0160] Step S3035: Extract features from environmental monitoring data to obtain temperature features, moisture features, dynamic features, and radiation features.

[0161] Specifically, electronic devices can extract features from environmental monitoring data, identifying four types of meteorological features that influence carbon decomposition and migration: temperature, moisture, dynamics, and radiation. Temperature features include daily average temperature (°C) and accumulated temperature (°C•d), which drive microbial enzyme activity; moisture features include rainfall (mm / month) and evaporation (mm / month), which affect soil moisture; dynamic features include wind speed (m / s, affecting carbon blowback) and runoff depth (mm / month, affecting carbon leaching); and radiation features include sunshine duration (h / d), which affects carbon production through plant photosynthesis.

[0162] Step S3036: Determine the carbon input and carbon output items corresponding to the target region based on the extracted features.

[0163] Specifically, the electronic device can calculate the carbon input of the plant residue based on the formula: Carbon input of plant residue = (Aboveground biomass × 0.3 + Belowground biomass × 0.7) × 0.45 × Return-to-field rate; 0.45 is used to characterize the easier retention of carbon in belowground biomass (weight 0.7), and 0.45 is the carbon content coefficient (the carbon content of plant dry matter is about 45%). For example, with aboveground biomass of 5000 kg / mu (≈750 g / m²), belowground biomass of 225 g / m², and a return-to-field rate of 0.6, then the carbon input of plant residue = (750 × 0.3 + 225 × 0.7) × 0.45 × 0.6 ≈ 87 g / m²•year.

[0164] Electronic devices can calculate the carbon input of root exudates based on the formula: Carbon input of root exudates = Concentration of root exudates × Percentage of root coverage area × 0.02; For example, if the concentration of root exudates is 50 mg / L and the coverage area is 80%, then the carbon input of root exudates is ≈ 50 × 0.8 × 0.02 ≈ 0.8 g / m²•year.

[0165] Electronic devices can calculate microbial carbon sequestration input based on the formula: Microbial carbon sequestration input = PLFA value (microbial activity) × soil weight × 0.1 × moisture suitability. Wherein, the PLFA value reflects the biomass of microorganisms, and each nmol PLFA corresponds to 0.1 g of carbon assimilation; moisture suitability is taken as 1.0 when W > 0.6, otherwise 0.5.

[0166] Initial carbon input = residual carbon input + root exudate carbon input + microbial carbon fixation input.

[0167] Then, the electronic device corrects the initial carbon input based on the nutrient carrying capacity sub-index to obtain the final carbon input.

[0168] For example, if the nutrient carrying capacity sub-index is greater than a first preset index threshold, the initial carbon input is multiplied by a first weight, where the first weight is greater than 1. If the nutrient carrying capacity sub-index is less than a second preset index threshold, the initial carbon input is multiplied by a second weight, where the second weight is less than 1. The first preset index threshold is greater than the second preset index threshold.

[0169] For example, if the nutrient carrying capacity sub-index is >0.7 (nutrients are sufficient): total carbon input ×1.2 (promotes plant growth); if the nutrient carrying capacity sub-index is <0.3 (nutrients are deficient): total carbon input ×0.8 (inhibits growth).

[0170] Electronic devices can calculate the carbon output based on the formula: Carbon output = Base soil organic carbon content × 0.02 × Temperature sensitivity coefficient × Moisture sensitivity coefficient; where 0.02 is the base annual decomposition rate; Temperature sensitivity coefficient = e^0.06 × (Daily average temperature - 15) (15℃ is the baseline); Moisture sensitivity coefficient (take 1.0 when W = 0.4-0.6, and 0.5 for excessively wet / dry conditions); For example, if the base soil organic carbon content = 20g / kg (≈4000g / m²), the daily average temperature is 25℃, and W = 0.5, then the carbon output ≈ 4000 × 0.02 × e^0.6 × 1.0 ≈ 80 × 1.82 ≈ 145.6g / m²•year.

[0171] In one optional implementation, the electronic device can adjust the carbon decomposition output based on a moisture carrying capacity sub-index. For example, if the moisture carrying capacity sub-index is greater than a third preset index threshold, the carbon decomposition output is multiplied by a third weight, where the third weight is less than 1. If the moisture carrying capacity sub-index is within a fourth preset index range, the carbon decomposition output is multiplied by a fourth weight, where the fourth weight is greater than 1. The third preset index threshold is greater than the maximum value within the fourth preset index range.

[0172] For example, if the water carrying capacity sub-index is >0.8 (overly humid): carbon output from decomposition ×0.8 (anaerobic deficiency inhibits microorganisms); if the water carrying capacity sub-index is =0.4-0.6 (suitable): carbon output from decomposition ×1.2 (humidity promotes decomposition).

[0173] Electronic devices can calculate runoff erosion carbon output based on the formula: Runoff erosion carbon output = Basic soil organic carbon content × 0.005 × Slope coefficient × Rainfall coefficient; where, Slope coefficient = 1 + 0.05 × Slope (°), and Rainfall coefficient = 1 + 0.01 × Average monthly rainfall (mm); For example, with a slope of 10° and an average monthly rainfall of 100 mm, the runoff carbon output ≈ 4000 × 0.005 × 1.5 × 2 = 60 g / m²•year.

[0174] Electronic devices can calculate carbon output from wind erosion based on the formula: Carbon output from wind erosion = Basic soil organic carbon content × 0.003 × Wind speed coefficient × Sand particle ratio; where, Wind speed coefficient = 1 + 0.1 × Wind speed (m / s). For example, if the sand content is 30% and the wind speed is 3m / s, then the carbon output from wind erosion is approximately 4000×0.003×1.3×0.3≈4.68g / m²•year.

[0175] Initial carbon output = decomposition carbon output + runoff erosion carbon output + wind erosion carbon output.

[0176] Step S3037: Calculate the initial change in soil organic carbon in the target area based on carbon input and carbon output terms.

[0177] Specifically, electronic devices can calculate the initial change in soil organic carbon in the target area by subtracting carbon output from carbon input.

[0178] Step S3038: Input the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, environmental monitoring data, and initial soil organic carbon change into the preset intelligent prediction model for dynamic change trend of soil organic carbon.

[0179] Specifically, step S3038 above may include the following steps: Step c1: The preset intelligent prediction model for dynamic changes in soil organic carbon extracts features from soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data to obtain root-microbe interaction features, microbial activity thermodynamics, topographic slope gradient, wind drying index, and slope aspect correction coefficient.

[0180] Specifically, the electronic device can integrate crop root exudate concentration and soil microbial PLFA value based on a preset intelligent prediction model of dynamic changes in soil organic carbon, quantifying the synergistic effect of root carbon input and microbial carbon sequestration (e.g., when both exudate concentration and PLFA value are high, the interaction characteristic value approaches 1). Root-microbe interaction characteristic = root exudate concentration (red-edge index inversion) × PLFA activity index (PLFA × 0.01). For example, with an exudate concentration of 50 mg / L (normalized to 1.0) and PLFA = 30 nmol / g (activity index 0.3), the root-microbe interaction characteristic = 1.0 × 0.3 = 0.3.

[0181] The electronic device calculates the microbial activity thermodynamics based on soil respiration rate, temperature characteristics, and a water carrying capacity sub-index (W), reflecting the intensity of microbial carbon decomposition activity (range 0-1, higher values ​​indicate stronger decomposition ability). For example, weights are assigned according to the influence of each parameter on microbial activity: soil respiration rate (R... n ): Weight 0.5 (directly reflects current metabolic intensity); Temperature standardized value (T) n ): Weight 0.3 (key environmental factors affecting enzyme activity); Water carrying capacity sub-index (W): Weight 0.2 (water is a basic condition for microbial activity).

[0182] The calculation formula is: Microbial activity heat = 0.5 × R n +0.3×T n +0.2×W.

[0183] The electronic device extracts and standardizes the terrain data (0-1). The greater the slope, the closer the gradient value is to 1 (e.g., 15° slope corresponds to 0.8, 25° corresponds to 1.0), thus obtaining the terrain slope gradient.

[0184] Electronic devices combine wind speed, evaporation and soil bulk density to calculate and quantify the soil carbon erosion potential caused by wind (formula: dryness index = 0.3 × wind speed coefficient + 0.7 × evaporation coefficient, range 0-1), thus obtaining the dryness index.

[0185] The electronic device determines the slope correction factor based on the slope aspect corresponding to the target area. For example, a factor of 1.2 is used for sunny slopes (to accelerate carbon decomposition), 0.8 for shady slopes (to slow down decomposition), and 1.0 for level slopes, to correct for the microenvironmental impact of topography on the carbon cycle.

[0186] Step c2 involves inputting the root-microbe interaction characteristics, microbial activity thermodynamics, topographic slope gradient, desiccation index, aspect correction coefficient, and initial soil organic carbon change into the preset convolutional network of the preset intelligent prediction model for dynamic changes in soil organic carbon.

[0187] Specifically, electronic devices can input root-microbe interaction characteristics, microbial activity thermodynamics, topographic slope gradient, desiccation index, aspect correction coefficient, and initial soil organic carbon change into a preset convolutional network in a preset intelligent prediction model for dynamic changes in soil organic carbon.

[0188] Step c3: The pre-set convolutional network performs convolution processing on the input features and outputs a carbon fixation potential feature map and a carbon loss risk feature map corresponding to the target region.

[0189] Specifically, the preset convolutional network includes two convolutional layers, and the target region includes multiple grid units. Step c3 above may include the following steps: Step c31: For each target grid cell, the first layer of the pre-defined convolutional network is used to extract convolutional features of the root-microbe interaction features, microbial activity thermodynamics, and initial soil organic carbon change of the target grid cell and its corresponding neighboring grid cells based on the first attention convolutional kernel of the first pre-defined size, so as to obtain the sub-carbon fixation potential feature map corresponding to each grid cell.

[0190] Specifically, for each target grid cell, the first convolutional layer in the preset convolutional network uses a first attention convolutional kernel of a first preset size (wherein the first preset size can be a 3×3 window, focusing on the central grid cell and its 8 neighboring grid cells) to extract features of root-microbe interaction characteristics, microbial activity thermodynamics, and initial SOC change. The attention weights are tilted towards regions with high interaction feature values ​​(e.g., the weight of grid cells with strong root-microbe synergy is increased by 20%). The sub-carbon fixation potential feature value of each target grid cell is output (range 0-1, representing the carbon fixation potential of the grid cell and its surroundings).

[0191] For example, the target grid is centered on the target grid, and the window covers the eight adjacent grids around it (forming a 3×3 square area; if the coordinates of the central grid are (i,j), then the window contains nine grids from (i-1,j-1) to (i+1,j+1)). This captures the spatial relationships between grids (e.g., root exudates from the central grid may affect the microbial activity of adjacent grids).

[0192] Then, three types of input features are extracted for each grid cell within the 3×3 window, forming a 9×3 feature matrix: root-microbe interaction features (I): range 0-1, higher values ​​indicate stronger synergistic carbon fixation ability between root exudates and microorganisms (e.g., 0.8 represents strong synergy); microbial activity thermodynamics (A): range 0-1, reflecting microbial carbon fixation activity (e.g., 0.6 represents moderate activity); initial SOC change ( : Unit: g / kg•year, standardized to 0-1 (positive value indicates carbon accumulation, such as 0.7 representing strong initial accumulation).

[0193] Next, the first convolutional network assigns basic spatial weights to the grid cells within the 3×3 window, highlighting the dominant role of the central grid cell. Specifically, the central grid cell (i,j) has a basic weight of 0.3; the eight adjacent grid cells each have a basic weight of 0.0875 (totaling 0.7).

[0194] The first layer of the convolutional network can adjust the base weights based on the root-microbe interaction feature (I) to enhance the contribution of highly collaborative regions. The modified formula is: Attention Weight = Base Weight × (1 + 0.2 × I). The higher the interaction feature I, the more significant the weight increase, up to a maximum of 20%.

[0195] Example calculation: Center grid I=0.9 → Attention weight=0.3×(1+0.2×0.9)=0.3×1.18=0.354; A neighboring grid I=0.5 → Attention weight=0.0875×(1+0.2×0.5)=0.0875×1.1=0.096.

[0196] To ensure that the sum of the attention weights of the nine grids is 1 (eliminating the instability of convolution calculations), the electronic device can normalize the corrected attention weights.

[0197] Finally, the electronic device can perform weighted summation on the three types of features within the 3×3 window: Interaction feature fusion value: ; Active thermal fusion value: ; initial Fusion value: .

[0198] Electronic devices can be weighted and summed according to carbon fixation mechanisms based on fusion characteristics.

[0199] For example, weight allocation: Interacting features (0.4, core driver) > Initial (0.3, basic trend) > active heat (0.3, carbon fixation capacity); Calculation formula: Sub-carbon fixation potential characteristic value = 0.4 × I_fusion + 0.3 × S_fusion + 0.3 × A_fusion.

[0200] The electronic device outputs a sub-carbon sequestration potential characteristic value (e.g., 0.76) for the target grid (i,j), reflecting the combined carbon sequestration potential of the grid and its neighborhood.

[0201] The above method enhances the areas with significant root-microbe synergy through the attention mechanism, while integrating the influence of the surrounding environment, so that the feature values ​​reflect both local advantages and spatial correlation, laying the foundation for subsequent assembly of a complete carbon fixation potential feature map.

[0202] Step c32: Slide the window to traverse the target region and stitch together the feature maps of each sub-carbon fixation potential to obtain the carbon fixation potential feature map.

[0203] Specifically, the electronic device can use a sliding 3×3 window to traverse the entire target area, and stitch all the sub-feature values ​​together according to their spatial positions to form a carbon fixation potential feature map of the same size as the target area (e.g., 500×500 pixels, with each pixel value being 0-1).

[0204] Step c33: For each target grid cell, the second convolutional layer in the preset convolutional network is used to extract features of the terrain slope gradient, wind dryness index and slope aspect correction coefficient corresponding to the target grid cell and its corresponding neighboring grid cells based on the second attention convolutional kernel of the second preset size, so as to obtain the unidirectional feature value corresponding to the target grid cell.

[0205] The second attention convolution kernel of the second preset size can be 5×5 in size, centered on the target grid, covering 24 adjacent grids around it (forming a 5×5 square area; if the coordinates of the center grid are (i,j), then the window contains 25 grids from (i-2,j-2) to (i+2,j+2)).

[0206] Specifically, the electronic device can extract three types of input features for each grid cell within a 5×5 window, forming a 25×3 feature matrix. These include: terrain slope gradient (G): ranging from 0 to 1, with higher values ​​indicating steeper slopes (e.g., 0.8 represents a steep slope with a high risk of carbon loss); wind dryness index (W): ranging from 0 to 1, reflecting the carbon erosion potential caused by wind (e.g., 0.7 represents strong wind drying); and aspect correction coefficient (D): 1.2 for sunny slopes, 0.8 for shady slopes, and 1.0 for level slopes, correcting the impact of slope aspect on carbon loss (sunny slopes have higher temperatures, accelerating carbon decomposition and loss).

[0207] Then, the electronic device assigns basic spatial weights to the grids within the 5×5 window, highlighting the influence of the central grid and its neighbors. Specifically, the central grid (i,j) has a basic weight of 0.1; the four adjacent grids (top, bottom, left, right) each have a basic weight of 0.08; the four diagonal grids (e.g., (i-1,j-1)) each have a basic weight of 0.05; and the 16 edge grids each have a basic weight of 0.03. The total checksum is: 0.1 + 4 × 0.08 + 4 × 0.05 + 16 × 0.03 = 0.1 + 0.32 + 0.2 + 0.48 = 1.1.

[0208] Next, the electronic device can adjust the basic weights corresponding to each grid based on the terrain slope gradient (G), with a more significant increase in weight as the slope steeper.

[0209] Correction rules: If G > 0.7 (steep slope), weight = base weight × 1.5 (increase by 50%); if 0.3 ≤ G ≤ 0.7 (gentle slope), weight = base weight × 1.0 (no correction); if G < 0.3 (flat ground), weight = base weight × 0.8 (decrease by 20%).

[0210] The electronic device can adjust the sum of the corrected weights of the 25 grids to 1, ensuring the stability of the convolution calculation.

[0211] The electronic device performs a weighted summation of the three types of features within a 5×5 window (based on normalized weights): Slope gradient fusion value: ; Dryness index composite value: ; Aspect correction blending value: ; Then, the electronic devices assign weights based on the influence intensity of the fusion features and sum them. The weight assignment is as follows: slope gradient (0.5, dominant leaching) > aridity index (0.3, wind-driven) > aspect correction (0.2, microenvironment regulation).

[0212] Calculation formula: Unidirectional eigenvalue = 0.5 × G 融合 +0.3×W 融合 +0.2×D 融合。

[0213] Finally, the electronic device outputs the unidirectional characteristic value of the target grid (i,j), reflecting the overall carbon loss risk of the grid and its 5×5 neighborhood.

[0214] By capturing the spatial correlation of terrain through a 5×5 large window and strengthening the steep slope signal through an attention mechanism, the feature values ​​can reflect both the loss potential of local terrain and the diffusion impact of the surrounding environment, providing accurate local data for subsequent stitching of a complete carbon loss risk feature map.

[0215] Step c34: The sliding window traverses the target area and splices the feature values ​​of each unidirectional direction to obtain a carbon loss risk feature map.

[0216] Specifically, electronic devices can use a sliding 5×5 window to traverse the region, stitch together all unidirectional feature values, and form a carbon loss risk feature map (the same size as the region, with higher values ​​indicating greater risk).

[0217] Step c4: Based on the carbon fixation potential feature map and the carbon loss risk feature map, output the dynamic change trend of soil organic carbon.

[0218] Specifically, the electronic device obtains the carbon fixation potential value (F, e.g., 0.75) and the carbon loss risk value (L, e.g., 0.2) from the carbon fixation potential characteristic map and the carbon loss risk characteristic map, and obtains the initial SOC value (S0, e.g., 20 g / kg, from soil monitoring data).

[0219] Then, the electronic device calculates the carbon fixation and loss. Wherein, carbon fixation: C fix =S0×F×k1; where k1 is a fixed coefficient, such as 0.05 / year for regional calibration. Example: 20×0.75×0.05=0.75g / kg•year.

[0220] Carbon loss: C loss =S0×L×k2; where k2 is the leaching coefficient, with a regional calibration value of 0.04 / year. Example: 20×0.2×0.04=0.16g / kg•year.

[0221] Annual change in grid SOC: ΔS=C fix -C loss Example: 0.75 - 0.16 = 0.59 g / kg per year (showing an increasing trend).

[0222] Electronic devices can extend single-grid results to the spatiotemporal dimension, capturing the heterogeneity of SOC changes.

[0223] Time series forecast: Short-term (1-3 years): Superimposed seasonal corrections (e.g., rainy season loss × 1.2, plant growth period fixed amount × 1.3); For example, the total change over 3 years ≈ 0.59 × 3 + seasonal adjustment ≈ 1.80 g / kg.

[0224] Medium term (5-10 years): Introduce trend decay (fixed potential decreases as SOC increases, F′=F×e-0.1×t); Example: after 5 years F′≈0.45, ΔS≈0.29g / kg•year.

[0225] Long-term (10 years or more): Combined with land use change (e.g., crop rotation of leguminous plants can increase F by 30%).

[0226] Electronic devices can also integrate all grid ΔS to divide trend regions: Growth zone (ΔS > 0.3 g / kg•year): Field center, high fixation potential, low risk of loss; Stable zone (0≤ΔS≤0.3g / kg•year): Gentle slope area, with basic balance between income and expenditure; The descending zone (ΔS < 0): the edge of a steep slope, dominated by the risk of water loss.

[0227] Step S3039: The preset intelligent prediction model for dynamic change trend of soil organic carbon outputs the dynamic change trend of soil organic carbon.

[0228] Specifically, the pre-set intelligent prediction model for dynamic changes in soil organic carbon integrates all the above analysis results and finally outputs a standardized dynamic change trend of SOC, including: grid-level annual change prediction; regional-scale growth / decline zoning; contribution of key driving factors (such as root-microbe interaction and slope) to the trend; and trend warnings for different future periods (such as "a significant decrease in SOC may occur in steep slope areas in 2028").

[0229] Step S304: Based on the dynamic change trend of soil organic carbon, determine the target measures corresponding to the target area.

[0230] Please refer to the above description of step S204 for details on this step, which will not be repeated here.

[0231] The intelligent prediction method for dynamic changes in soil organic carbon provided in this application extracts biomass characteristics, root exudate concentration, and current growth stage from crop data. It captures the core carriers of carbon exchange between plants and soil, while the growth stage reflects the temporal rhythm of carbon input, providing key parameters for carbon input calculation.

[0232] Slope gradient and aspect are extracted from topographic data. This reflects the driving role of topography in carbon loss, laying a spatial foundation for subsequent carbon loss risk assessment. The soil and water carrying capacity index is decomposed into water carrying capacity sub-indices and nutrient carrying capacity sub-indices. This separates the independent effects of water and nutrients on the carbon cycle, making carbon calculations more accurate. Basic organic carbon content, microbial characteristics, and physicochemical characteristics are extracted from soil data. Obtaining the background value and key driving factors of the carbon cycle is the core basis for carbon quantification. Temperature, moisture, dynamics, and radiation characteristics are extracted from environmental data. This captures the regulatory role of the external environment on the carbon cycle, improving the environmental adaptability of carbon calculations. Carbon inputs and outputs are identified based on these characteristics. Key pathways of the carbon cycle are identified, and multi-source characteristics are transformed into quantifiable carbon balance items, providing clear indicators for carbon balance analysis. The initial change in soil organic carbon is obtained by the difference between carbon inputs and outputs. The basic trend (accumulation or loss) of regional carbon balance is quantified as a benchmark value for subsequent dynamic prediction. The soil and water carrying capacity index, multi-source monitoring data, and the initial change in soil organic carbon are input into the prediction model. This model integrates all factors influencing soil organic carbon dynamics to provide comprehensive input for accurate prediction. It extracts root-microbe interaction features, microbial activity thermodynamics, topographic slope gradient, aridity index, and aspect correction coefficients from the input data. Focusing on key driving features of the carbon cycle, it reduces redundant information and improves model efficiency. These core features, along with the initial soil organic carbon variation, are input into a convolutional network. Utilizing the spatial awareness of the convolutional network, it captures spatial correlations of features (such as carbon migration between adjacent grids) to prepare for generating a spatial feature map. The first layer of the convolutional network uses a first attention convolutional kernel (e.g., 3×3) to extract carbon fixation-related features of the center and its neighborhood, stitching them together to form a potential map. It strengthens areas with strong root-microbe interactions (high carbon fixation potential) to generate a refined spatial distribution map, visually demonstrating spatial differences in carbon fixation. The second layer of the convolutional network uses a second attention convolutional kernel (e.g., 5×5) to extract topographically related loss features, stitching them together to form a risk map. It covers a wider range of topographic correlations (e.g., the impact of steep slopes on the surrounding area), highlighting high-risk areas and providing spatial guidance for loss prevention and control. By combining carbon sequestration potential maps and loss risk maps, the dynamic trend of soil organic carbon is output. Integrating the spatial patterns of carbon sequestration and loss, the final output is a complete result that includes both temporal trends (increase / decrease) and spatial distribution (hotspot areas), providing decision support for precise carbon management.

[0233] This embodiment also provides an intelligent prediction device for the dynamic change trend of soil organic carbon. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0234] This embodiment provides an intelligent prediction device for the dynamic change trend of soil organic carbon, such as... Figure 4 As shown, it includes: The acquisition module 401 is used to acquire soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data and environmental monitoring data corresponding to the target area; The first determining module 402 is used to extract features from soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area. The prediction module 403 is used to predict the dynamic change trend of soil organic carbon in the target area based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data and environmental monitoring data. The second determination module 404 is used to determine the target measures corresponding to the target area based on the dynamic change trend of soil organic carbon.

[0235] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0236] The intelligent prediction device for dynamic changes in soil organic carbon in this embodiment is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0237] This invention also provides an electronic device having the above-described features. Figure 4 The device shown is an intelligent prediction device for the dynamic change trend of soil organic carbon.

[0238] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an optional embodiment of the present invention, such as... Figure 5As shown, the electronic device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise as required. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0239] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for intelligent prediction of dynamic changes in soil organic carbon, characterized in that, The method includes: Acquire soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area; Feature extraction is performed on the soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area; Based on the soil and water carrying capacity index, the soil monitoring data, the topography monitoring data, the crop monitoring data, and the environmental monitoring data, predict the dynamic change trend of soil organic carbon in the target area; Based on the dynamic change trend of soil organic carbon, the target measures corresponding to the target area are determined.

2. The method according to claim 1, characterized in that, The step of extracting features from the soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area includes: Static data and time-series dynamic data are determined from the soil monitoring data, the topography monitoring data, the crop monitoring data, the facility monitoring data, and the environmental monitoring data; The target region is divided into at least one grid cell; For each of the grid cells, target static data is filtered from the static data; Based on the target static data, construct a static vector; The static vectors corresponding to each of the grid units are concatenated to generate a static matrix; Feature extraction is performed on the static matrix to obtain static fusion features; Based on the duration, dimension, and number of grid cells corresponding to the time series dynamic data, a dynamic input tensor is generated. Feature extraction is performed on the dynamic input tensor to obtain dynamic fusion features; The static fusion feature and the dynamic fusion feature are fused together to generate the target fusion feature; The target fusion features are input into a preset soil and water bearing capacity assessment model, and the soil and water bearing capacity index corresponding to the target area is output.

3. The method according to claim 2, characterized in that, The step of extracting features from the static matrix to obtain static fusion features includes: The correlation between each sub-static feature in the static matrix and the soil and water bearing capacity was calculated using the Pearson correlation coefficient: Convert the correlation degree into initial feature weights; The static matrix and the initial feature weights are input into a preset random forest model; The preset random forest model binds the initial weights of the features to the column vectors of the static matrix to form a sub-static feature-initial weight mapping table; For each grid cell in the static matrix, based on the sub-static feature-initial weight mapping table, each sub-static feature in the static matrix is ​​modified, and a weighted static matrix is ​​obtained based on the modified sub-static features. Feature extraction is performed on the weighted static matrix to obtain the splitting features of each target; For each of the target splitting features, a first weighting coefficient corresponding to the target splitting feature is calculated based on the initial weight of the feature corresponding to the target splitting feature; Based on the first weighting coefficient corresponding to each of the target split features, the target split features are fused to obtain the static fused features.

4. The method according to claim 2, characterized in that, The step of extracting features from the dynamic input tensor to obtain dynamic fusion features includes: The dynamic input tensor is input into the input gate of the preset temporal processing model; The input gate identifies the dynamic input tensor, filters valid information using the sigmoid function, and transmits the valid information to the forget gate; The forget gate identifies the valid information, discards invalid information, and transmits the remaining valid information to the output gate; The output gate integrates the output information of the forget gate and outputs the regular temporal characteristics of each grid cell; Obtain the current growth stage of the crop; Based on the current reproductive period, each of the conventional time-series features is modified to obtain the modified time-series features; Obtain the resource demand threshold for the current reproductive period corresponding to each of the modified time-series features and the modified sub-dynamic features corresponding to the current reproductive period; Calculate the matching degree between each of the modified sub-dynamic features corresponding to each of the modified time-series features and the corresponding reproductive period resource demand threshold; The matching degree of each of the modified sub-dynamic features corresponding to each of the modified temporal features is integrated to generate a reproductive period adaptation feature vector; The dynamic fusion features are obtained based on the reproductive period adaptation feature vector.

5. The method according to claim 4, characterized in that, The dynamic fusion features obtained based on the reproductive period adaptation feature vector include: The environmental monitoring data is identified to determine whether the current environment meets the preset emergency threshold. If the current environment does not meet the preset emergency threshold, then extract the sub-dynamic feature mutation signal corresponding to the extreme time period from the dynamic input tensor; The sub-dynamic feature mutation signal is quantized into emergency feature values; Based on the emergency feature value, the reproductive period adaptation feature vector is modified in an emergency manner to obtain the dynamic fusion feature.

6. The method according to claim 1, characterized in that, The method of predicting the dynamic change trend of soil organic carbon in the target area based on the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data includes: Feature extraction was performed on the crop monitoring data to obtain biomass characteristics, root exudate concentration, and current growth stage; Feature extraction is performed on the terrain monitoring data to obtain the terrain slope gradient and terrain aspect; The water and soil bearing capacity index is decomposed into a water bearing capacity sub-index and a nutrient bearing capacity sub-index; Feature extraction was performed on the soil monitoring data to obtain basic soil organic carbon content, microbial characteristics, and physicochemical characteristics; Feature extraction is performed on the environmental monitoring data to obtain temperature features, moisture features, dynamic features, and radiation features; Based on the extracted features, determine the carbon input and carbon output items corresponding to the target region; Based on the carbon input and carbon output terms, the initial change in soil organic carbon in the target area is calculated. The water and soil carrying capacity index, the soil monitoring data, the topography monitoring data, the crop monitoring data, the environmental monitoring data, and the initial soil organic carbon change are input into the preset intelligent prediction model for the dynamic change trend of soil organic carbon. The preset intelligent prediction model for dynamic changes in soil organic carbon outputs the dynamic changes in soil organic carbon.

7. The method according to claim 6, characterized in that, The preset intelligent prediction model for dynamic changes in soil organic carbon outputs the dynamic changes in soil organic carbon, including: The preset intelligent prediction model for dynamic changes in soil organic carbon extracts features from the soil and water carrying capacity index, soil monitoring data, topographic monitoring data, crop monitoring data, and environmental monitoring data to obtain root-microbe interaction features, microbial activity thermodynamics, topographic slope gradient, aridity index, and aspect correction coefficient. The root-microbe interaction characteristics, the microbial activity thermodynamics, the topographic slope gradient, the aridity index, the aspect correction coefficient, and the initial soil organic carbon change are input into the preset convolutional network of the preset intelligent prediction model for dynamic change trend of soil organic carbon. The preset convolutional network performs convolution processing on the input features and outputs a carbon fixation potential feature map and a carbon loss risk feature map corresponding to the target region. Based on the carbon fixation potential feature map and the carbon loss risk feature map, the dynamic change trend of soil organic carbon is output.

8. The method according to claim 7, characterized in that, The preset convolutional network includes two layers, and the target region includes multiple grid units. The preset convolutional network performs convolution processing on the input features and outputs a carbon fixation potential feature map and a carbon loss risk feature map corresponding to the target region, including: For each target grid cell, the first layer of the preset convolutional network extracts convolutional features based on the first attention convolutional kernel of the first preset size for the root-microbe interaction features, the microbial activity thermodynamics, and the initial soil organic carbon change of the target grid cell and its corresponding neighboring grid cells, thereby obtaining the sub-carbon fixation potential feature map corresponding to each grid cell. The sliding window traverses the target region and stitches together the feature maps of each sub-carbon fixation potential to obtain the carbon fixation potential feature map. For each target grid cell, the second convolutional layer in the preset convolutional network extracts features from the terrain slope gradient, the wind dryness index, and the slope aspect correction coefficient corresponding to the target grid cell and its corresponding neighboring grid cells based on the second attention convolutional kernel of the second preset size, thereby obtaining the unidirectional feature value corresponding to the target grid cell. The sliding window traverses the target area and splices together the unidirectional feature values ​​to obtain the carbon loss risk feature map.

9. A smart predictive device for the dynamic change trend of soil organic carbon, characterized in that, The device includes: The acquisition module is used to acquire soil monitoring data, topographic monitoring data, crop monitoring data, facility monitoring data, and environmental monitoring data corresponding to the target area. The first determining module is used to extract features from the soil monitoring data, the topography monitoring data, the crop monitoring data, the facility monitoring data, and the environmental monitoring data to determine the water and soil carrying capacity index corresponding to the target area. The prediction module is used to predict the dynamic change trend of soil organic carbon in the target area based on the soil and water carrying capacity index, the soil monitoring data, the topography monitoring data, the crop monitoring data, and the environmental monitoring data. The second determining module is used to determine the target measures corresponding to the target area based on the dynamic change trend of soil organic carbon.

10. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent prediction method for dynamic changes in soil organic carbon as described in any one of claims 1 to 8.

Images