An artificial intelligence-based soil carbon sequestration and increase optimization method

Abstract

The application discloses a kind of soil carbon sequestration and carbon increase optimization method based on artificial intelligence, and the application relates to soil carbon sequestration and carbon increase technical field.The steps of the method include: obtaining the soil internal data set and remote sensing image data set of the target area soil;Based on LSTM network, a time prediction model is constructed, the soil internal data set is input, and the future soil carbon content change rate is output;A spatial distribution prediction model is constructed, the remote sensing data set is input, and the soil carbon density spatial distribution is output;The change rate and spatial distribution are combined using a weighted fusion algorithm to calculate the predicted value of soil carbon storage, and then the soil carbon sequestration capacity is obtained;Determine the microbial respiration rate, calculate the carbon content natural loss combined with the soil oxygen content, and then obtain the soil carbon sink potential combined with the carbon sequestration capacity;According to the carbon sink potential, the target area is classified, and carbon sequestration and carbon increase measures are implemented for different potential level areas.The application integrates multiple data sources to realize soil carbon content coupling analysis, and then optimizes the pertinence of carbon sequestration and carbon increase measures.

Description

Technical Field

[0001] This invention relates to the field of soil carbon sequestration and enhancement technology, specifically to an artificial intelligence-based optimization method for soil carbon sequestration and enhancement. Background Technology

[0002] With the intensification of global climate change, the continuous rise in carbon dioxide concentration has become one of the main factors leading to the greenhouse effect and ecosystem degradation. As one of the Earth's largest carbon sinks, soil's carbon sequestration capacity is directly related to the carbon cycle balance and sustainable agricultural development.

[0003] Currently, in the field of soil carbon sequestration and enhancement, existing technologies generally adopt a method that combines traditional ecological measures with simple model analysis. Simple linear regression models are used to analyze the correlation between soil carbon storage and single environmental factors to obtain soil carbon content. Then, based on experience, measures such as vegetation restoration, straw return to the field, and organic fertilizer application are used to increase soil organic carbon input and achieve soil carbon sequestration and enhancement. However, simple models are difficult to integrate multi-source heterogeneous data, such as real-time data from soil sensors and spatial information from remote sensing images, and cannot achieve coupled analysis of soil carbon content, which leads to insufficient targeting and precision of carbon sequestration and enhancement measures. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based optimization method for soil carbon sequestration and enhancement. This method solves the problem that simple models are unable to integrate multi-source heterogeneous data, thus failing to achieve coupled analysis of soil carbon content, which in turn leads to insufficient targeting and precision of carbon sequestration and enhancement measures.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based method for optimizing soil carbon sequestration and enhancement, comprising the following steps:

[0006] Step S1: Collect and preprocess the soil internal data of the target carbon sequestration and carbon enrichment area to obtain the soil internal dataset; at the same time, acquire and calibrate the remote sensing image data of the target carbon sequestration and carbon enrichment area to obtain the remote sensing image dataset.

[0007] Step S2: Based on the Long Short-Term Memory (LSTM) network model, construct a soil carbon content change rate time prediction model, input the soil internal dataset into the soil carbon content change rate time prediction model, and output the soil carbon content change rate for future time periods.

[0008] Step S3: Based on the convolutional neural network (CNN) model, construct a soil carbon density spatial distribution prediction model, input the remote sensing image dataset into the soil carbon density spatial distribution prediction model, and output the soil carbon density spatial distribution.

[0009] Step S4: Using a weighted fusion algorithm, combine the rate of change of soil carbon content and the spatial distribution of soil carbon density in the future time period to calculate the predicted value of soil carbon storage. Based on the predicted value of soil carbon storage, calculate the soil carbon sequestration capacity.

[0010] Step S5: Determine the microbial respiration rate through a respiration measurement experiment. Calculate the natural loss of soil carbon content based on the microbial respiration rate and the soil oxygen content in the soil internal dataset. Combine the natural loss of soil carbon content with the soil carbon sequestration capacity to calculate the soil carbon sink potential.

[0011] Step S6: Based on the soil carbon sequestration potential, the target carbon sequestration and carbon enhancement areas are classified into potential-level areas. Based on the potential-level areas, carbon sequestration and carbon enhancement measures are implemented in the target carbon sequestration and carbon enhancement areas.

[0012] Preferably, the soil internal data of the target carbon sequestration and carbon enhancement area is collected and preprocessed to obtain a soil internal dataset including:

[0013] Within the target carbon sequestration and enhancement area, a multi-layered, integrated soil sensor IoT network is deployed according to the grid layout principle of one point per 10 mu (approximately 1.65 acres). To ensure the consistency and comparability of data in the vertical space, all soil sensors are uniformly deployed in the core soil tillage layer of 0-30cm for synchronous measurement to obtain internal soil data.

[0014] Soil internal data processing is as follows:

[0015] First, physical outliers, data that exceed the sensor's measurement range or do not conform to actual physical laws, are removed. Then, missing values ​​are filled using cubic spline interpolation. Finally, Z-score standardization is used to unify the dimensions of the soil internal data. Standardized data facilitates subsequent data analysis and model processing, avoiding the influence of different dimensions on the analysis results, thus obtaining the soil internal dataset.

[0016] Preferably, the remote sensing image data of the target carbon sequestration and carbon enrichment area is acquired and calibrated to obtain a remote sensing image dataset including:

[0017] Data was collected using drones equipped with multispectral cameras, and a strategy of seasonal aerial surveys and weekly inspections of key areas was adopted to capture changes in soil and vegetation in different growing seasons and key areas.

[0018] Pix4D software is used to process images acquired by the drone to generate orthophotos to eliminate distortions caused by terrain undulations and shooting angles.

[0019] The calibration process is as follows:

[0020] Using a standard reflectivity plate with known radiance values, images were taken under the same lighting conditions as the image acquisition. Based on the actual reflectance value of the standard reflectivity plate and the DN value in the image, a linear regression equation was established to radiometrically calibrate each band of the acquired multispectral image, converting the DN value into radiance or reflectance value.

[0021] Geometric calibration is as follows:

[0022] Ground control points (GCPs) are evenly distributed within the study area. The precise geographical coordinates of these GCPs are measured using high-precision GPS. Image data with GCP coordinates is imported into Pix4D software. The software algorithm matches the GCPs in the image with the actual measured coordinates and performs geometric correction on the image to reduce geometric errors caused by flight attitude and lens distortion.

[0023] After the above processing, the calibrated remote sensing image dataset is finally obtained.

[0024] Preferably, the soil carbon content change rate time prediction model based on the Long Short-Term Memory (LSTM) network model includes:

[0025] A soil carbon content change rate time prediction model based on the Long Short-Term Memory (LSTM) network model is constructed, including the model architecture and mathematical principles, input layer, core gating mechanism, output layer, and LSTM model evaluation metrics.

[0026] The model architecture and mathematical principles are as follows:

[0027] Long Short-Term Memory (LSTM) networks capture long-term dependencies in time series data through a gating mechanism, making them suitable for dynamic prediction of soil carbon content.

[0028] The input layer is a 4-dimensional feature vector. , Soil moisture For temperature Organic matter content Soil pH value;

[0029] The core gating mechanisms include: forget gate, input gate, cell state update, and output gate;

[0030] Output layer: 3D vector , representing the rate of change of soil carbon content in the next 1 month, 3 months, and 6 months, respectively;

[0031] LSTM model evaluation metrics include the coefficient of determination and the mean absolute error.

[0032] Coefficient of determination The calculation formula is:

[0033]

[0034] in, It is a predicted value. This is the actual value. It is the average of the actual values; The closer the value is to 1, the better the model fits; it measures the proportion of data variance that the model can explain. This indicates that the model perfectly fits the data;

[0035] The formula for calculating the mean absolute error (MAE) is:

[0036]

[0037] in, It is a predicted value. This is the actual value. It represents the sample size; MAE reflects the average deviation between the predicted and actual values. The smaller its value, the smaller the prediction deviation of the model, that is, the more accurate the prediction result.

[0038] Preferably, the soil carbon density spatial distribution prediction model based on the convolutional neural network (CNN) model includes:

[0039] A soil carbon density spatial distribution prediction model based on a convolutional neural network (CNN) is constructed, including the model architecture and mathematical principles, input layer, convolutional layer, output layer, pooling layer, and CNN model evaluation metrics.

[0040] The architecture and mathematical principles of the CNN model are as follows:

[0041] Convolutional Neural Networks (CNNs) extract spatial features through local convolution operations, making them suitable for resolving the spatial distribution of carbon density from remote sensing images.

[0042] The model input layer uses a multispectral image I with a spatial resolution of 256 pixels × 256 pixels and containing 5 spectral bands;

[0043] The formula for extracting local spectral features using the first convolutional layer is as follows:

[0044]

[0045] in, It is the output of the k-th feature map of the first convolutional layer at position (i, j); These are the weights of the first convolutional kernel, and their dimension is... ,in It is a spatial dimension. It is the channel dimension. is the index of the feature map; i and j are the spatial indices of the output feature map; p and q are the offsets within the pooling window; Is the input image in Position, number The value of the channel; It is the first floor. The bias of each feature map; ReLU is the activation function;

[0046] The pooling layer uses 2×2 max pooling with a stride of 2 to reduce dimensionality while preserving key features. The calculation formula is as follows:

[0047]

[0048] in, It is the first pooling layer after the second pooling layer. Each feature map in Position output; It is the first convolutional layer after the first convolutional layer. Each feature map in Position output;

[0049] The second convolutional and fully connected layers use 64 3×3×5 convolutional kernels, with the same structure as the first convolutional layer, to further extract abstract features;

[0050] Finally, the output is passed through a fully connected layer:

[0051]

[0052] in, It is the predicted regional carbon density, i.e., the spatial distribution of soil carbon density, in tons per hectare; Linear indicates a linear transformation operation; This represents the result of flattening the feature map output from the second convolutional layer.

[0053] Preferably, by using a weighted fusion algorithm, the soil carbon content change rate and soil carbon density spatial distribution over the future time period are combined to obtain the predicted soil carbon storage value, including:

[0054] The rate of change of soil carbon content and the spatial distribution of soil carbon density for the future time period are combined using the following formula:

[0055]

[0056] in, It is the predicted value of soil carbon storage at spatial location s at time t, representing the predicted soil carbon storage at a specific spatial location and time. It was predicted by CNN. Initial carbon density; It is predicted by the Long Short-Term Memory Network. At the The monthly carbon content change rate, where k is the monthly index, is calculated by interpolation using only the 1-month, 3-month, and 6-month change rates output by LSTM. It is the baseline scenario prediction, which represents the predicted value of soil carbon storage under the condition of natural soil carbon content without predictive intervention. It is a spatial weighting coefficient used to dynamically balance the contribution ratio of the model predictions to the baseline scenario predictions in the final result. Its value is determined based on historical data for that location.

[0057] Preferably, the soil carbon sequestration capacity is calculated based on the predicted soil carbon storage value, including:

[0058] The following formula can be used to predict and calculate soil carbon sequestration capacity based on spatiotemporal coupled carbon content:

[0059]

[0060] in, This indicates the soil carbon sequestration capacity at spatial location s; It is the predicted value of soil carbon storage at time t obtained by spatiotemporal coupling in the step; This represents the maximization operator. The subscript t indicates that this operation is performed on the time variable t. This operator will traverse and compare all values ​​within the brackets {} and take the maximum value.

[0061] Preferably, the microbial respiration rate is determined through a respiration measurement experiment, and the natural loss of soil carbon content is calculated based on the microbial respiration rate and the soil oxygen content in the soil internal dataset, including:

[0062] The collected soil samples were placed in a breathing chamber, sealed, and the measurement was started. The CO2 concentration was recorded at regular intervals for 24-48 hours.

[0063] Calculate the microbial respiration rate based on the change in CO2 concentration over time. The calculation formula is:

[0064]

[0065] in, It measures the change in CO2 concentration over a time interval, where Δt is the time interval and V is the volume of the breathing chamber. This is the dry weight of the soil sample;

[0066] Calculate the carbon loss factor. The calculation formula is:

[0067]

[0068] in, Indicates the carbon loss factor. Indicates the rate of microbial respiration. Indicates soil oxygen content;

[0069] Soil in location Place, from time arrive Total natural carbon loss experienced It can be calculated using the following integral formula:

[0070]

[0071] in, and It calculates the start and end times of the loss, where t is a continuous time variable within the integration interval. The internal continuous change represents every instant from the beginning to the end.

[0072] Preferably, the soil carbon sequestration potential is calculated by combining the natural loss of soil carbon content and the soil carbon sequestration capacity, including:

[0073] The current carbon content is as follows:

[0074] Current carbon content Part of the calculation involved collecting soil and vegetation samples from the target area, measuring them in the laboratory, and correcting the results using spatiotemporally coupled carbon content predictions. The calculation formula is as follows:

[0075]

[0076] in, This is the carbon content measured in the laboratory. The carbon content is predicted through spatiotemporal coupling. The weighting coefficients are determined based on the reliability of the measurement data and the accuracy of the prediction data.

[0077] Combining the natural loss of soil carbon content, soil carbon sequestration capacity, and current carbon content, the soil carbon sequestration potential is calculated as follows:

[0078]

[0079] in, It is the potential for soil carbon sequestration. It refers to the soil's carbon sequestration capacity. This is the natural loss of soil carbon content.

[0080] Preferably, step S6 includes:

[0081] Based on the soil carbon sequestration potential The target carbon sequestration and carbon enrichment areas are classified into potential-level areas:

[0082] when When the concentration is 1 tonnes per square kilometer, it is considered a high-potential area.

[0083] when tons / square kilometer At a rate of tons per square kilometer, it is considered a medium-potential area;

[0084] when At a rate of tons per square kilometer, it is considered a low-potential area;

[0085] Based on the aforementioned potential areas, carbon sequestration and enhancement measures will be implemented in the target carbon sequestration and enhancement areas:

[0086] In high-potential areas, investment can be increased, more advanced technologies and management methods can be adopted to improve carbon sequestration efficiency, precision agriculture technologies can be introduced, and soil carbon content processes can be optimized through variable fertilization and irrigation measures to further improve the soil's carbon sequestration capacity.

[0087] In areas with medium potential, optimize existing planting and management methods, adjust the planting structure, increase the proportion of perennial crops, reduce soil disturbance, and promote the accumulation of soil organic carbon.

[0088] In low-potential areas, basic soil improvement work should be carried out first to gradually improve the soil's carbon sequestration capacity. This can be achieved by adding organic materials to improve soil structure, increase soil microbial activity, and thus enhance the soil's carbon sequestration potential.

[0089] This invention provides an artificial intelligence-based method for optimizing soil carbon sequestration and enhancement, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0090] (1) The artificial intelligence-based soil carbon sequestration and carbon enhancement optimization method uses LSTM to capture time series dynamics and CNN to analyze spatial heterogeneity. It uses a spatiotemporal coupling model to integrate the change rate of time dimension with spatial distribution characteristics, so that carbon storage prediction can simultaneously reflect "when it changes" and "where it differs". Compared with a single model, it significantly improves the prediction accuracy and provides a reliable basis for the quantification of carbon sequestration capacity.

[0091] (2) This artificial intelligence-based soil carbon sequestration and carbon enhancement optimization method quantifies carbon sink potential and achieves precise policy implementation. It calculates natural carbon loss based on respiration experiments and soil oxygen content, and obtains carbon sink potential by combining carbon sequestration capacity and current carbon content. After classifying the region according to the potential value, it adopts precision agriculture technology, planting structure optimization and soil improvement measures for high, medium and low potential areas respectively, to avoid resource waste and make carbon sequestration and carbon enhancement measures more targeted and efficient.

[0092] (3) This artificial intelligence-based soil carbon sequestration and carbon enhancement optimization method integrates multi-source data-driven and mechanism analysis. On the one hand, it acquires multi-dimensional data on soil moisture, temperature, and spectrum through sensor networks and remote sensing images, and realizes automated data-driven analysis with the help of artificial intelligence models. On the other hand, it quantifies carbon loss factors through experimental mechanism research on microbial respiration rate, combining data-driven and ecological mechanisms. This ensures prediction efficiency and improves the model's ability to explain the actual carbon content process, providing scientific support for agricultural carbon sink management. Attached Figure Description

[0093] Figure 1 This is a flowchart of an artificial intelligence-based optimization method for soil carbon sequestration and enhancement proposed in this invention.

[0094] Figure 2 This invention proposes an artificial intelligence-based method for optimizing soil carbon sequestration and enhancement, which yields a hierarchical map of soil carbon sequestration capacity.

[0095] Figure 3 This is a hierarchical diagram of soil carbon sequestration and enhancement measures obtained in an artificial intelligence-based soil carbon sequestration and enhancement optimization method proposed in this invention. Detailed Implementation

[0096] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0097] Please see Figures 1-3 This invention provides a technical solution: an artificial intelligence-based method for optimizing soil carbon sequestration and enhancement. Specifically, the following artificial intelligence-based method for optimizing soil carbon sequestration and enhancement is provided; please refer to [link / reference]. Figure 1 The method includes the following steps:

[0098] Step S1: Collect and preprocess the soil internal data of the target carbon sequestration and carbon enrichment area to obtain the soil internal dataset; at the same time, acquire and calibrate the remote sensing image data of the target carbon sequestration and carbon enrichment area to obtain the remote sensing image dataset.

[0099] Within the target carbon sequestration and enhancement area, a multi-layered, integrated soil sensor IoT network is deployed according to a grid layout principle of one monitoring point per 10 mu (approximately 1.65 acres). Each monitoring point in the soil sensor IoT network integrates a complete sensor unit, containing one of each of the following sensor types. The data acquired by the complete sensor unit at each monitoring point represents the soil condition of the 10-mu grid area where that point is located. To ensure data consistency and comparability in vertical space, all soil sensors are uniformly deployed in the 0-30cm core soil tillage layer for synchronous measurement to acquire internal soil data. This network integrates the following sensor units to achieve real-time acquisition of key soil parameters:

[0100] Soil moisture sensor: Used to accurately measure soil volumetric water content with an accuracy of ±3%. This data provides a direct basis for assessing soil conditions and developing precise irrigation strategies, effectively avoiding soil carbon loss caused by water stress or over-irrigation;

[0101] Soil temperature sensor: with a resolution of 0.1℃, it can accurately monitor changes in soil temperature. Soil temperature directly affects microbial activity and the rate of organic matter decomposition, and is a key environmental factor for assessing soil carbon cycle processes;

[0102] Soil organic matter spectral sensor: Based on near-infrared spectroscopy (wavelength range 350-2500nm) analysis technology, it can non-destructively and rapidly retrieve soil organic carbon content. Organic matter content is a core indicator characterizing soil carbon sequestration potential.

[0103] Soil pH sensor: Utilizing electrochemical sensing principles, this sensor uses a metal probe to react with hydrogen ions in the soil solution via a redox reaction, generating a current signal related to pH. This allows for in-situ measurement of pH, typically within the pH range, with an accuracy of ±0.1 pH. Soil pH significantly influences soil microbial community structure, enzyme activity, and carbon stabilization processes (e.g., acidic soils are more conducive to the physicochemical protection of organic carbon), making it a crucial parameter for assessing soil carbon sequestration capacity and stability.

[0104] Soil internal data processing is as follows:

[0105] First, outliers—data that exceeds the sensor's measurement range or does not conform to actual physical laws—are removed. Then, cubic spline interpolation is used to fill in the missing values. During data acquisition, data loss may occur due to sensor malfunction or other reasons. Cubic spline interpolation can recover these missing data to some extent, making the data more complete.

[0106] Z-score standardization was used to unify the dimensions of soil internal data. Standardized data facilitates subsequent data analysis and model processing, avoids the influence of different dimensions on the analysis results, and yields a soil internal dataset.

[0107] The remote sensing image data acquisition process is as follows:

[0108] Data was collected using drones equipped with multispectral cameras (MicaSenseRedEdge, 5 bands: blue, green, red, red edge, and near-infrared). A strategy of seasonal aerial surveys (one flight each in spring, summer, and autumn, at an altitude of 120m, with an overlap rate of 75%) and weekly inspections of key areas was adopted to capture changes in soil and vegetation in different growing seasons and key areas.

[0109] Remote sensing image data processing: Pix4D software is used to process the images acquired by the UAV to generate orthophotos to eliminate distortions caused by terrain undulations and shooting angles.

[0110] The calibration process is as follows:

[0111] Using standard reflectivity plates with known radiance values ​​(standard plates with reflectance of 20%, 50%, and 80%), images were taken under the same lighting conditions as the image acquisition. Based on the actual reflectance values ​​of the standard reflectivity plates and the DN values ​​(digital quantization values) in the images, a linear regression equation was established to radiometrically calibrate each band of the acquired multispectral images, converting the DN values ​​into radiance values ​​or reflectance values.

[0112] Geometric calibration is as follows:

[0113] Ground control points (GCPs) are evenly distributed within the study area. The number of GCPs is determined based on the size of the study area and the accuracy requirements, generally no less than four per square kilometer. The precise geographical coordinates of these GCPs are measured using high-precision GPS. Image data containing the GCP coordinates is imported into Pix4D software. Software algorithms are used to match the GCPs in the image with their actual measured coordinates, performing geometric correction on the image to reduce geometric errors caused by flight attitude and lens distortion.

[0114] After the above processing, the calibrated remote sensing image dataset is finally obtained.

[0115] This step uses soil data from the 0-30cm core soil tillage layer obtained through a soil sensor IoT network. After removing outliers, performing cubic spline interpolation, and standardizing with Z-score, a soil internal dataset is formed. Simultaneously, remote sensing images are collected using a drone equipped with a multispectral camera. These images are then radiometrically calibrated and geometrically calibrated using Pix4D software to generate a remote sensing image dataset. This provides standardized input data for the LSTM temporal prediction model in step S2 and the CNN spatial prediction model in step S3, ensuring the accuracy and reliability of model training.

[0116] Step S2: Based on the Long Short-Term Memory (LSTM) network model, construct a soil carbon content change rate time prediction model, input the soil internal dataset into the soil carbon content change rate time prediction model, and output the soil carbon content change rate for future time periods.

[0117] A time-based prediction model for soil carbon content change rate based on the Long Short-Term Memory (LSTM) network model is constructed, including the model architecture and mathematical principles, input layer, core gating mechanism, output layer, and LSTM model evaluation metrics.

[0118] The model architecture and mathematical principles are as follows:

[0119] Long Short-Term Memory (LSTM) networks capture long-term dependencies in time series through gating mechanisms, making them suitable for dynamic prediction of soil carbon content (such as the impact of seasonal temperature changes on carbon storage).

[0120] The input layer consists of 4-dimensional feature vectors. , For soil moisture, For temperature, For organic matter content, This refers to the soil pH value.

[0121] The core gating mechanisms include: forget gate, input gate, cell state update, and output gate.

[0122] The forgetting gate determines how many historical cell states are retained, calculated using the following formula:

[0123]

[0124] in, Indicates the forget gate at time The output of the cell determines which information to discard from the cell state. It is the Sigmoid activation function, whose output ranges from 0 to 1; when the output is 0, it means that the corresponding information is completely discarded; when the output is 1, it means that the corresponding information is completely retained. It is the weight matrix of the forget gate, which determines the degree of influence of the input data and the hidden state at the previous time step on the output of the forget gate; It is the hidden state of the previous moment, which contains information about the previous time step; It is the input data at the current moment. It is a bias term of the forget gate, used to adjust the output of the activation function; for example, in the cold season, the forget gate will "forget" the active state of microorganisms during the high temperature of summer to avoid interfering with the prediction of the current carbon content.

[0125] The input gate determines how much new information is updated to the cell state; the calculation formula is as follows:

[0126] ;

[0127] ;

[0128] in, This indicates the input gate at time [time]. The output of this function has a value between 0 and 1. It determines how much information from the current input can be updated in the cell state. It is the weight matrix of the input gate, used to weight the input data and the hidden state of the previous time step; It is the bias term of the input gate, used to adjust the output of the Sigmoid activation function; This indicates the state of the candidate cell at time t. The value of is limited to between 1 and 1 by the tanh function. It is a candidate value used to calculate the cell state update at the current moment; It is the hyperbolic tangent activation function, used to map the input to a range of 1 to 1; It is a weight matrix for candidate cell states, used to weight the input data and the hidden state of the previous time step; It is a bias term for candidate cell states, used to adjust the output of the tanh activation function; when spring litter increases, the input gate strengthens the weight of the "carbon input rate" feature and updates the cell state to reflect the carbon accumulation trend.

[0129] Cell state update, calculated using the following formula:

[0130]

[0131] in, Indicates the current time Cellular state (long-term memory); The output of the forget gate determines how much information to discard from the cell state; Indicates the previous moment The cellular state. It is element multiplication; the physical meaning of this formula is to integrate historical carbon content (such as carbon storage in the same period last year) with current environmental characteristics (such as precipitation this month) to form a continuous time-dependent relationship.

[0132] Output gate: determines the current hidden state (short-term memory);

[0133]

[0134] in, Is the output gate at any time? The output of has a value range between 0 and 1; It is the weight matrix of the output gate, used to weight the input data and the hidden state of the previous time step. It is the bias term of the output gate, used to adjust the output of the Sigmoid activation function.

[0135] Then, the current hidden state (short-term memory). The calculation formula is:

[0136]

[0137] in, It is the current moment. The hidden states are used for output layer prediction. It is the current moment. Cellular state (long-term memory).

[0138] The significance of the input gate is to screen key features related to changes in carbon storage (such as the output gate amplifying the effect of "microbial respiration rate" at humidity).

[0139] Output layer: 3D vector , representing the percentage change (%) of soil carbon content over the next 1 month, 3 months, and 6 months, respectively.

[0140] The calculation method for the fully connected layer is as follows: The output formula is as follows:

[0141]

[0142] in , These are the output layer weights and biases.

[0143] LSTM model evaluation metrics include the coefficient of determination and the mean absolute error.

[0144] Coefficient of determination The calculation formula is:

[0145]

[0146] in, It is a predicted value. This is the actual value. It is the average of the actual values; The closer the value is to 1, the better the model fits the data. It measures the proportion of data variance that the model can explain. This indicates that the model perfectly fits the data.

[0147] The formula for calculating the mean absolute error (MAE) is:

[0148]

[0149] in, It is a predicted value. This is the actual value. It represents the sample size; MAE reflects the average deviation between predicted and actual values. The smaller its value, the smaller the prediction bias of the model, meaning the more accurate the prediction result.

[0150] A model is constructed based on an LSTM gating mechanism (forget gate, input gate, output gate). The input consists of a 4-dimensional feature vector of soil moisture, temperature, organic matter content, and CO2 concentration. The output is a time series of carbon content change rates for the next 1, 3, and 6 months. The model is then analyzed using the coefficient of determination R0. 2 The mean absolute error (MAE) is used to assess model accuracy. This provides a time-dimensional prediction of carbon content changes for the spatiotemporal coupling in step S4, capturing seasonal carbon content dynamics (such as changes in microbial activity with temperature), enabling subsequent predictions to reflect time-series carbon storage fluctuations.

[0151] Step S3: Based on the convolutional neural network (CNN) model, construct a soil carbon density spatial distribution prediction model, input the remote sensing image dataset into the soil carbon density spatial distribution prediction model, and output the soil carbon density spatial distribution.

[0152] A soil carbon density spatial distribution prediction model based on a convolutional neural network (CNN) is constructed, including the model architecture and mathematical principles, input layer, convolutional layer, output layer, pooling layer, and CNN model evaluation metrics.

[0153] The architecture and mathematical principles of the CNN model are as follows:

[0154] CNN (Convolutional Neural Network) extracts spatial features through local convolution operations, making it suitable for resolving the spatial distribution of carbon density from remote sensing images (e.g., areas with high vegetation cover have higher carbon density).

[0155] The model input layer uses a multispectral image I with a spatial resolution of 256 pixels × 256 pixels, containing 5 spectral bands. These 5 bands correspond to the blue, green, red, red-edge, and near-infrared channels, respectively. This image originates from a pre-calibrated radiometric and geometrically calibrated remote sensing image dataset.

[0156] The first convolutional layer uses 32 3×3×5 convolutional kernels to extract local spectral features (such as the reflectance of plant leaves). The formula is as follows:

[0157]

[0158] in, It is the output of the k-th feature map of the first convolutional layer at position (i, j); These are the weights of the first convolutional kernel, and their dimension is... ,in It is spatial dimension. It is the channel dimension. is the index of the feature map; i and j are the spatial indices of the output feature map; p and q are the offsets within the pooling window; Is the input image in Position, number The value of the channel; It is the first floor. The bias of each feature map; ReLU is the activation function, which introduces nonlinearity to simulate the "nonlinear relationship between vegetation spectrum and carbon density"; the convolution kernel captures the features of "high reflectance in near-infrared light + low reflectance in red light" (corresponding to dense vegetation), providing a basis for subsequent carbon density prediction.

[0159] The pooling layer uses 2×2 max pooling with a stride of 2 to reduce dimensionality while preserving key features. The calculation formula is as follows:

[0160]

[0161] in, It is the first pooling layer after the second pooling layer. Each feature map in Position output; It is the first convolutional layer after the first convolutional layer. Each feature map in Position output.

[0162] The strongest responses in local areas (such as pixels with the densest vegetation) are preserved; the amount of data is reduced by 75%, while the salient features related to carbon density are enhanced.

[0163] The second convolutional and fully connected layers use 64 3×3×5 convolutional kernels, with the same structure as the first convolutional layer (except for the number of convolutional kernels), to further extract abstract features (such as the association pattern of "vegetation community type soil carbon density").

[0164] Finally, the output is passed through a fully connected layer:

[0165]

[0166] in, It is the predicted regional carbon density, i.e., the spatial distribution of soil carbon density, in tons per hectare; Linear indicates a linear transformation operation; This represents the result of flattening the feature map output from the second convolutional layer.

[0167] CNN model evaluation metrics:

[0168] The formula for calculating the root mean square error (RMSE) is as follows:

[0169]

[0170] in, It is the predicted regional carbon density; It is the actual regional carbon density; It represents the number of samples in the CNN model; RMSE is sensitive to large errors and is used to ensure the stability of carbon density prediction; the smaller the RMSE value, the closer the model's prediction is to the actual value, and the better the model's performance.

[0171] The formula for calculating the correlation coefficient r is as follows:

[0172]

[0173] in, It is the predicted regional average carbon density; This is the actual regional average carbon density; the correlation coefficient is used to measure the linear correlation between the predicted and actual values. The value of r is between 1 and 1. The closer r is to 1, the stronger the linear relationship between the predicted and actual values, and the better the prediction effect of the model; the closer r is to 1, the stronger the negative correlation; r close to 0 indicates that there is almost no linear relationship between the predicted and actual values.

[0174] Multispectral remote sensing images are input into the soil carbon density spatial distribution prediction model. Through convolutional layers (extracting spectral features), pooling layers (reducing dimensions to preserve features), and fully connected layers, the model outputs the spatial distribution prediction of the regional average carbon density. The root mean square error (RMSE) and correlation coefficient (r) are used to evaluate the model. This provides the spatial dimension of carbon density distribution (such as the difference in carbon density between forest land and farmland) for step S4, enabling spatiotemporal coupling to combine "regional heterogeneity" and providing spatial data support for subsequent carbon sequestration capacity calculation and regional classification.

[0175] Step S4: Using a weighted fusion algorithm, combine the rate of change of soil carbon content and the spatial distribution of soil carbon density in the future time period to obtain the predicted value of soil carbon storage. Based on the predicted value of soil carbon storage, calculate the soil carbon sequestration capacity.

[0176] The rate of change of soil carbon content and the spatial distribution of soil carbon density for the future time period are combined using the following formula:

[0177]

[0178] in, It is the predicted value of soil carbon storage at spatial location s at time t, representing the predicted soil carbon storage at a specific spatial location and time. It was predicted by CNN. Initial carbon density (tons / hectare); It was predicted by the Long Short-Term Memory (LSTM) network. At the The monthly carbon content change rate, where k is the monthly index, is calculated by interpolation using only the 1-month, 3-month, and 6-month change rates output by LSTM. It is the baseline scenario prediction (natural carbon content without intervention), which represents the predicted value of soil carbon storage under the condition of natural soil carbon content without predictive intervention. It is the spatial weighting coefficient (ranging from 0 to 1), used to balance the contribution ratio of the dynamic model prediction and the baseline scenario prediction in the final result. Its value is determined based on historical data for that location.

[0179] The physical meaning of spatiotemporal coupling is as follows:

[0180] Time dimension (LSTM): Memorize "seasonal patterns" through gating mechanisms, such as predicting that farmland carbon content will decrease in summer (t=3) (due to increased microbial respiration) and increase in autumn (t=6) (due to decreased respiration and straw return to the field).

[0181] Spatial dimension (CNN): Extracts “regional heterogeneity” through convolution, for example: predicting that the carbon density of forest s1 (15 tons / hectare) is higher than that of farmland s2 (8 tons / hectare) due to differences in litter input.

[0182] The blending effect, the final output This not only reflects the temporal dynamics of "the carbon storage of forest land s1 decreased from 15 tons / hectare to 14.5 tons three months later due to increased summer respiration," but also reflects the spatial differences of "the carbon storage of farmland s2 increased from 8 tons / hectare to 8.3 tons during the same period due to straw return to the field," thus achieving accurate prediction of "when, where, and how carbon changes."

[0183] The following formula can be used to predict and calculate soil carbon sequestration capacity based on spatiotemporal coupled carbon content:

[0184]

[0185] in, This indicates the soil carbon sequestration capacity at spatial location s; It is the predicted value of soil carbon storage at time t obtained by spatiotemporal coupling in the step; This represents the maximization operator. The subscript t indicates that this operation is performed on the time variable t. This operator iterates through and compares all values ​​within the brackets {} and takes the maximum value. The meaning of this formula is that the maximum soil carbon storage value among all predicted time points is considered the soil carbon sequestration capacity at that location.

[0186] By combining the time series prediction of LSTM with the spatial distribution prediction of CNN through a weighted fusion algorithm, the formula integrates the initial carbon density, the monthly change rate and the baseline scenario to obtain the spatiotemporally coupled soil carbon storage prediction value, and takes the maximum value of the time dimension as the soil carbon sequestration capacity.

[0187] For subsequent applications: quantifying the dynamic process of "when, where, and how carbon changes" (such as the decline in forest carbon storage in summer), providing key parameters for the carbon sink potential calculation in step S5, namely the maximum carbon storage that the soil can achieve.

[0188] Step S5: Determine the microbial respiration rate through a respiration measurement experiment. Calculate the natural loss of soil carbon content based on the microbial respiration rate and the soil oxygen content in the soil internal dataset. Combine the natural loss of soil carbon content with the soil carbon sequestration capacity to calculate the soil carbon sink potential.

[0189] The respiratory measurement experiment was designed as follows:

[0190] First, sample collection is conducted. Soil samples are collected within the study area according to a specific grid layout (e.g., one sampling point per 10×10 square meters). The sampling depth is determined based on research needs, generally ranging from 0 to 30 centimeters.

[0191] Next, the experimental setup was prepared using an airtight breathing chamber (a transparent plexiglass container with gas inlet and outlet and sensor interface) to ensure that the soil sample was relatively isolated from the external environment during the measurement process.

[0192] Environmental control is then implemented by installing temperature and humidity sensors in the respiration chamber. The temperature is controlled at the actual ambient temperature of the soil (e.g., 20-30 degrees Celsius, determined based on actual measurements), and the humidity is controlled at 60-80% of the soil's field capacity (the soil's field capacity is determined through preliminary experiments). A carbon dioxide (CO2) gas analyzer (such as a high-precision infrared CO2 analyzer) is connected to the gas outlet of the respiration chamber to monitor the CO2 concentration released by soil microbial respiration in real time.

[0193] The final microbial respiration rate was measured as follows:

[0194] The collected soil samples were placed in a breathing chamber, sealed, and measurements were taken. The CO2 concentration was recorded at regular intervals (e.g., every hour) for 24-48 hours.

[0195] Calculate the microbial respiration rate based on the change in CO2 concentration over time. The calculation formula is:

[0196]

[0197] in, Δt is the change in CO2 concentration (ppm) within the measurement time interval, Δt is the time interval (h), and V is the volume of the breathing chamber (L). This is the dry weight (kg) of the soil sample.

[0198] Calculate the carbon loss factor. The calculation formula is:

[0199]

[0200] in, Indicates the carbon loss factor. Indicates the rate of microbial respiration. This formula represents the soil oxygen content; it signifies that the carbon loss factor is the product of microbial respiration rate and soil oxygen content. Microbial respiration rate and soil oxygen content are two key factors influencing soil carbon loss, and this formula can quantify the effects of microbial respiration on soil at a specific location. and time The carbon loss situation.

[0201] Soil in location Place, from time arrive Total natural carbon loss experienced It can be calculated using the following integral formula:

[0202]

[0203] in, and It calculates the start and end times of the loss, where t is a continuous time variable within the integration interval. The continuous change within the formula represents every instant from beginning to end; the significance of this formula is that it adjusts the carbon loss factor... In the time period ( By integrating within a certain time period, we can obtain the natural loss of soil carbon content during that time period.

[0204] The current carbon content is as follows:

[0205] Current carbon content Part of the analysis involved collecting soil and vegetation samples from the target area, measuring them in the laboratory, and correcting the results using spatiotemporally coupled carbon content predictions. The calculation formula is as follows:

[0206]

[0207] in, This is the carbon content measured in the laboratory. The carbon content is predicted through spatiotemporal coupling. The weighting coefficients are determined based on the reliability of the measurement data and the accuracy of the prediction data.

[0208] Combining the natural loss of soil carbon content, soil carbon sequestration capacity, and current carbon content, the soil carbon sequestration potential is calculated as follows:

[0209]

[0210] in, It is the potential for soil carbon sequestration. It refers to the soil's carbon sequestration capacity. This refers to the natural depletion of soil carbon content. By considering the soil's maximum carbon sequestration capacity, current carbon content, and natural depletion, the carbon sink potential of the soil at a specific location can be calculated. This value can help assess the soil's potential for carbon storage and provide a basis for developing relevant carbon sequestration and enhancement strategies.

[0211] Microbial respiration rate was calculated through a respiration measurement experiment (monitoring CO2 concentration changes in an airtight respiration chamber), and carbon loss factor was obtained by combining it with soil oxygen content. Natural loss was calculated by integration. The current carbon content was corrected by laboratory measurements and spatiotemporal prediction. Finally, the carbon sink potential was calculated from the natural loss of soil carbon content, soil carbon sequestration capacity, and current carbon content. The actual potential for increasing carbon storage in the soil was quantified, providing quantitative indicators for regional classification in step S6, so that subsequent measures can be formulated for carbon loss and carbon sequestration space in different potential areas.

[0212] Step S6: Based on the soil carbon sequestration potential, the target carbon sequestration and carbon enhancement areas are classified into potential-level areas. Based on the potential-level areas, carbon sequestration and carbon enhancement measures are implemented in the target carbon sequestration and carbon enhancement areas.

[0213] Based on the soil carbon sequestration potential The target carbon sequestration and carbon enrichment areas are classified into potential-level areas:

[0214] when When the concentration is 1 tonnes per square kilometer, it is considered a high-potential area.

[0215] when tons / square kilometer At a rate of tons per square kilometer, it is considered a medium-potential area;

[0216] when When the concentration is 1 tonnes per square kilometer, it is considered a low-potential area.

[0217] Based on the aforementioned potential areas, carbon sequestration and enhancement measures will be implemented in the target carbon sequestration and enhancement areas:

[0218] In high-potential areas, increased investment and the adoption of more advanced technologies and management methods can enhance carbon sequestration efficiency. For example, introducing precision agriculture technologies and optimizing soil carbon content processes through variable fertilization and irrigation measures can further improve the soil's carbon sequestration capacity.

[0219] In areas with medium potential for development, optimize existing planting and management methods. For example, adjust the planting structure, increase the proportion of perennial crops, reduce soil disturbance, and promote the accumulation of soil organic carbon.

[0220] In low-potential areas, basic soil improvement work should be carried out first to gradually improve the soil's carbon sequestration capacity. For example, adding organic materials (such as compost and green manure) can improve soil structure, increase soil microbial activity, and thus enhance the soil's carbon sequestration potential.

[0221] This technical solution addresses the imbalance of soil carbon content under global climate change by proposing an AI-based optimization method for soil carbon sequestration and enhancement. It involves deploying a soil sensor network in the target area to collect and preprocess internal data on moisture, temperature, organic matter spectra, and atmospheric CO2 concentration. Simultaneously, it utilizes a UAV equipped with a multispectral camera to acquire and calibrate remote sensing image data. Two models are then constructed: a time-based LSTM-based model predicting the rate of change of soil carbon content and a spatial prediction model based on CNN. The former takes a 4D feature vector as input and outputs a time series of future carbon content change rates, while the latter takes multispectral images as input and outputs the spatial distribution of carbon density. A weighted fusion algorithm combines these two models to obtain a spatiotemporally coupled carbon content prediction and calculate soil carbon sequestration capacity. Next, a respiration measurement experiment is used to determine the microbial respiration rate, and combined with soil oxygen content, the natural carbon loss is calculated. This, along with carbon sequestration capacity and current carbon content, yields the carbon sink potential. Finally, based on the carbon sink potential, the region is divided into high, medium, and low potential zones, and precision agriculture, optimized planting structure, and soil improvement measures for carbon sequestration and enhancement are implemented accordingly. This solution achieves spatiotemporal dynamic prediction through multi-model fusion. By combining data-driven and mechanistic analysis, it enables precise tiered policy implementation, facilitating carbon sink assessment.

[0222] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0223] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. An artificial intelligence-based method for optimizing soil carbon sequestration and enhancement, characterized in that, Includes the following steps: Step S1: Collect and preprocess the soil internal data of the target carbon sequestration and carbon enrichment area to obtain the soil internal dataset; at the same time, acquire and calibrate the remote sensing image data of the target carbon sequestration and carbon enrichment area to obtain the remote sensing image dataset. Step S2: Based on the Long Short-Term Memory (LSTM) network model, construct a soil carbon content change rate time prediction model, input the soil internal dataset into the soil carbon content change rate time prediction model, and output the soil carbon content change rate for future time periods. Step S3: Based on the convolutional neural network (CNN) model, construct a soil carbon density spatial distribution prediction model, input the remote sensing image dataset into the soil carbon density spatial distribution prediction model, and output the soil carbon density spatial distribution. Step S4: Using a weighted fusion algorithm, combine the rate of change of soil carbon content and the spatial distribution of soil carbon density in the future time period to calculate the predicted value of soil carbon storage. Based on the predicted value of soil carbon storage, calculate the soil carbon sequestration capacity. Step S5: Determine the microbial respiration rate through a respiration measurement experiment. Calculate the natural loss of soil carbon content based on the microbial respiration rate and the soil oxygen content in the soil internal dataset. Combine the natural loss of soil carbon content with the soil carbon sequestration capacity to calculate the soil carbon sink potential. Step S6: Based on the soil carbon sequestration potential, the target carbon sequestration and carbon enhancement areas are classified into potential-level areas. Based on the potential-level areas, carbon sequestration and carbon enhancement measures are implemented in the target carbon sequestration and carbon enhancement areas.

2. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 1, characterized in that, Soil internal data from the target carbon sequestration and carbon enrichment area were collected and preprocessed to obtain a soil internal dataset, including: Within the target carbon sequestration and enhancement area, a multi-layered, integrated soil sensor IoT network is deployed according to the grid layout principle of one point per 10 mu (approximately 1.65 acres). To ensure the consistency and comparability of data in the vertical space, all soil sensors are uniformly deployed in the core soil tillage layer of 0-30cm for synchronous measurement to obtain internal soil data. Soil internal data processing is as follows: First, physical outliers, data that exceed the sensor's measurement range or do not conform to actual physical laws, are removed. Then, missing values ​​are filled using cubic spline interpolation. Finally, Z-score standardization is used to unify the dimensions of the soil internal data. Standardized data facilitates subsequent data analysis and model processing, avoiding the influence of different dimensions on the analysis results, thus obtaining the soil internal dataset.

3. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 2, characterized in that, Remote sensing image data of the target carbon sequestration and carbon enrichment area were acquired and calibrated to obtain a remote sensing image dataset, including: Data was collected using drones equipped with multispectral cameras, and a strategy of seasonal aerial surveys and weekly inspections of key areas was adopted to capture changes in soil and vegetation in different growing seasons and key areas. Pix4D software is used to process images acquired by the drone to generate orthophotos to eliminate distortions caused by terrain undulations and shooting angles. The calibration process is as follows: Using a standard reflectivity plate with known radiance values, images were taken under the same lighting conditions as the image acquisition. Based on the actual reflectance value of the standard reflectivity plate and the DN value in the image, a linear regression equation was established to radiometrically calibrate each band of the acquired multispectral image, converting the DN value into radiance or reflectance value. Geometric calibration is as follows: Ground control points (GCPs) are evenly distributed within the study area. The precise geographical coordinates of these GCPs are measured using high-precision GPS. Image data with GCP coordinates is imported into Pix4D software. The software algorithm matches the GCPs in the image with the actual measured coordinates and performs geometric correction on the image to reduce geometric errors caused by flight attitude and lens distortion. After the above processing, the calibrated remote sensing image dataset is finally obtained.

4. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 3, characterized in that, A time-based prediction model for the rate of change of soil carbon content was constructed based on the Long Short-Term Memory (LSTM) network model, including: A soil carbon content change rate time prediction model based on the Long Short-Term Memory (LSTM) network model is constructed, including the model architecture and mathematical principles, input layer, core gating mechanism, output layer, and LSTM model evaluation metrics. The model architecture and mathematical principles are as follows: Long Short-Term Memory (LSTM) networks capture long-term dependencies in time series data through a gating mechanism, making them suitable for dynamic prediction of soil carbon content. The input layer consists of 4-dimensional feature vectors. , Soil moisture For temperature Organic matter content Soil pH value; The core gating mechanisms include: forget gate, input gate, cell state update, and output gate; Output layer: 3D vector , representing the rate of change of soil carbon content in the next 1 month, 3 months, and 6 months, respectively; LSTM model evaluation metrics include the coefficient of determination and the mean absolute error. Coefficient of determination The calculation formula is: ； in, It is a predicted value. This is the actual value. It is the average of the actual values; The closer the value is to 1, the better the model fits; it measures the proportion of data variance that the model can explain. This indicates that the model perfectly fits the data; The formula for calculating the mean absolute error (MAE) is: ； in, It is a predicted value. This is the actual value. It represents the sample size; MAE reflects the average deviation between the predicted and actual values. The smaller its value, the smaller the prediction deviation of the model, that is, the more accurate the prediction result.

5. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 4, characterized in that, A spatial distribution prediction model for soil carbon density is constructed based on a convolutional neural network (CNN) model, including: A soil carbon density spatial distribution prediction model based on a convolutional neural network (CNN) is constructed, including the model architecture and mathematical principles, input layer, convolutional layer, output layer, pooling layer, and CNN model evaluation metrics. The architecture and mathematical principles of the CNN model are as follows: Convolutional Neural Networks (CNNs) extract spatial features through local convolution operations, making them suitable for resolving the spatial distribution of carbon density from remote sensing images. The model input layer uses a multispectral image I with a spatial resolution of 256 pixels × 256 pixels and containing 5 spectral bands; The formula for extracting local spectral features using the first convolutional layer is as follows: ； in, It is the output of the k-th feature map of the first convolutional layer at position (i, j); These are the weights of the first convolutional kernel, and their dimension is... ,in It is spatial dimension. It is the channel dimension. is the index of the feature map; i and j are the spatial indices of the output feature map; p and q are the offsets within the pooling window; Is the input image in Position, number The value of the channel; It is the first floor. The bias of each feature map; ReLU is the activation function; The pooling layer uses 2×2 max pooling with a stride of 2 to reduce dimensionality while preserving key features. The calculation formula is as follows: ； in, It is the first pooling layer after the second pooling layer. Each feature map in Position output; It is the first convolutional layer after the first convolutional layer. Each feature map in Position output; The second convolutional and fully connected layers use 64 3×3×5 convolutional kernels, with the same structure as the first convolutional layer, to further extract abstract features; Finally, the output is passed through a fully connected layer: ； in, It is the predicted regional carbon density, i.e., the spatial distribution of soil carbon density, in tons per hectare; Linear indicates a linear transformation operation; This represents the result of flattening the feature map output from the second convolutional layer.

6. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 5, characterized in that, By combining the rate of change of soil carbon content and the spatial distribution of soil carbon density over the future time period using a weighted fusion algorithm, a predicted value of soil carbon storage is obtained, including: The rate of change of soil carbon content and the spatial distribution of soil carbon density for the future time period are combined using the following formula: ； in, It is the predicted value of soil carbon storage at spatial location s at time t; It was predicted by CNN. Initial carbon density; It is predicted by the Long Short-Term Memory Network. Located in the The monthly carbon content change rate, where k is the monthly index, is calculated by interpolation using only the 1-month, 3-month, and 6-month change rates output by LSTM. It is the baseline scenario prediction, which represents the predicted value of soil carbon storage under the condition of natural soil carbon content without predictive intervention. It is a spatial weighting coefficient used to dynamically balance the contribution ratio of the model predictions to the baseline scenario predictions in the final result. Its value is determined based on historical data for that location.

7. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 6, characterized in that, Based on the predicted soil carbon storage value, the soil carbon sequestration capacity is calculated, including: The following formula can be used to predict and calculate soil carbon sequestration capacity based on spatiotemporal coupled carbon content: ； in, This indicates the soil carbon sequestration capacity at spatial location s; It is the predicted value of soil carbon storage at time t obtained by spatiotemporal coupling in the step; This represents the maximization operator. The subscript t indicates that this operation is performed on the time variable t. This operator will traverse and compare all values ​​within the brackets {} and take the maximum value.

8. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 7, characterized in that, Microbial respiration rate was determined through respiration measurement experiments. Based on the microbial respiration rate and soil oxygen content in the soil internal dataset, the natural loss of soil carbon content was calculated, including: The collected soil samples were placed in a breathing chamber, sealed, and the measurement was started. The CO2 concentration was recorded at regular intervals for 24-48 hours. Calculate the microbial respiration rate based on the change in CO2 concentration over time. The calculation formula is: ； in, It measures the change in CO2 concentration over a time interval, where Δt is the time interval and V is the volume of the breathing chamber. This is the dry weight of the soil sample; Calculate the carbon loss factor. The calculation formula is: ； in, Indicates the carbon loss factor. Indicates the rate of microbial respiration. Indicates soil oxygen content; Soil in location Place, from time arrive Total natural carbon loss experienced It can be calculated using the following integral formula: ； in, and It calculates the start and end times of the loss, where t is a continuous time variable within the integration interval. The internal continuous change represents every instant from the beginning to the end.

9. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 8, characterized in that, Combining the natural loss of soil carbon content and soil carbon sequestration capacity, the soil carbon sequestration potential is calculated, including: The current carbon content is as follows: Current carbon content Part of the analysis involved collecting soil and vegetation samples from the target area, measuring them in the laboratory, and correcting the results using spatiotemporally coupled carbon content predictions. The calculation formula is as follows: ； in, This is the carbon content measured in the laboratory. The carbon content is predicted through spatiotemporal coupling. The weighting coefficients are determined based on the reliability of the measurement data and the accuracy of the prediction data. Combining the natural loss of soil carbon content, soil carbon sequestration capacity, and current carbon content, the soil carbon sequestration potential is calculated as follows: ； in, It is the potential for soil carbon sequestration. It refers to the soil's carbon sequestration capacity. This is the natural loss of soil carbon content.

10. The method for optimizing soil carbon sequestration and enhancement based on artificial intelligence according to claim 9, characterized in that, Step S6 includes: Based on the soil carbon sequestration potential The target carbon sequestration and carbon enrichment areas are classified into potential-level areas: when When the concentration is 1 tonnes per square kilometer, it is considered a high-potential area. when tons / square kilometer At a rate of tons per square kilometer, it is considered a medium-potential area; when At a rate of tons per square kilometer, it is considered a low-potential area; Based on the aforementioned potential areas, carbon sequestration and enhancement measures will be implemented in the target carbon sequestration and enhancement areas: In high-potential areas, investment can be increased, more advanced technologies and management methods can be adopted to improve carbon sequestration efficiency, precision agriculture technologies can be introduced, and soil carbon content processes can be optimized through variable fertilization and irrigation measures to further improve the soil's carbon sequestration capacity. In areas with medium potential, optimize existing planting and management methods, adjust the planting structure, increase the proportion of perennial crops, reduce soil disturbance, and promote the accumulation of soil organic carbon. In low-potential areas, basic soil improvement work should be carried out first to gradually improve the soil's carbon sequestration capacity. This can be achieved by adding organic materials to improve soil structure, increase soil microbial activity, and thus enhance the soil's carbon sequestration potential.

Images