A farmland net carbon emission prediction method, system and device

Abstract

The application provides a farmland net carbon emission prediction method, system and device, and the method comprises the following steps: obtaining crop related data of different crops in different planting areas, nitrous oxide emission measured data, methane emission measured data and carbon sequestration efficiency measured data; constructing a nitrous oxide emission prediction model according to the crop related data and the nitrous oxide emission measured data; constructing a methane emission prediction model according to the crop related data and the methane emission measured data; constructing a carbon sequestration efficiency prediction model according to the crop related data and the carbon sequestration efficiency measured data; calculating a methane emission prediction value, a nitrous oxide emission prediction value and a carbon sequestration efficiency prediction value; and calculating a net carbon emission prediction value according to the nitrous oxide emission prediction value, the methane emission prediction value, the carbon sequestration efficiency prediction value and a preset net carbon emission model. The application can realize high-precision and regional prediction and evaluation of farmland net carbon emission.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of agricultural carbon emission prediction, and particularly relates to a farmland net carbon emission prediction method, system and device. BACKGROUND

[0002] With the increasing attention on the problem of greenhouse gas emissions of agricultural systems, the research on the farmland carbon emission assessment system has become a technical focus. The system generally includes two core links: one is the simulation of the greenhouse gas emission process, especially the high spatiotemporal resolution estimation of key gases such as N2O and CH4; the other is the quantification of the carbon fixation process, mainly involving the contribution of measures such as straw returning, organic fertilizer or biochar application to the increment of soil organic carbon. Due to the high complexity of the ecological process of farmland, the emission and carbon fixation processes are influenced by the interaction of various management measures (such as planting density, fertilization mode and type, tillage mode) and environmental factors (soil type, climate conditions, etc.), forming a nonlinear, multivariate coupled feedback mechanism, which puts forward higher requirements for the model structure design, variable processing ability and dynamic response ability. The prediction means in the prior art has the following deficiencies or defects:

[0003] I. In the modeling of greenhouse gas emissions, traditional methods mostly rely on the default emission factors of IPCC or empirical statistical models, which can achieve rough estimation, but are difficult to reflect the complex nonlinear interaction between farmland management measures and environmental factors at a fine scale. Taking nitrous oxide (N2O) as an example, its emission is not only affected by the amount of fertilizer and crop type, but also by the coupling regulation between management measures such as tillage mode, fertilization period and climate and soil properties. There is a high degree of nonlinearity and multi-level interaction between the above variables, and it is difficult to achieve accurate modeling by using fixed coefficients or single variable functions.

[0004] In addition, the multi-source heterogeneity (classification variables and continuous variables coexist) of variables, the scale difference (field operation and long-term average of climate coexist) and the time sequence dynamics further increase the modeling difficulty. In recent years, although some researches have tried to introduce neural networks, random forests and other machine learning methods to improve the prediction accuracy of the model, most modeling schemes still exist as "black box tools", which cannot effectively integrate the mechanism of agronomic processes, lack the generalization ability and management response explanation force across regions, and are difficult to serve the goals of carbon emission reduction and precise regulation of farmland.

[0005] II. In the modeling of carbon fixation, most existing researches use empirical parameters or fixed coefficient methods, ignoring the interactive regulation of farmland management measures (such as tillage mode, fertilization intensity, water regulation, crop type) and environmental and soil climate factors. For example, the carbon stability and carbon fixation potential of the same straw returning measure may be completely different in the high temperature and high humidity areas in the south and the dry and cold areas in the north. Traditional methods generally lack the modeling ability of the "management x environment" interaction, and cannot realize the simulation of the carbon fixation efficiency according to local conditions.

[0006] In addition, although patent CN113513269A proposes a farmland carbon emission prediction system based on "resource estimation", integrating multi-source data and modeling methods, its core is still based on the static application of emission factors, without establishing a coupling learning mechanism between resource input and carbon emission. The "prediction model" claimed by the system is actually a combination of traditional models, which is difficult to respond to the dynamic response of emission processes to management measures, and the overall is closer to a data reporting type list tool, rather than an intelligent system with generalization ability and decision support function.

[0007] In summary, the existing farmland carbon emission and carbon sequestration evaluation technology generally has problems such as low estimation accuracy, insufficient response ability and poor system integration. The emission model relies on fixed emission factors, which is difficult to reflect the complex interaction between management measures and environmental factors; the carbon sequestration evaluation often uses a unified efficiency coefficient, ignoring the differences between different management measures. In addition, although some data-driven methods have improved prediction performance, they generally lack agronomic mechanism support, and the model has limited interpretability and regional generalization ability, making it difficult to meet the current demand for intelligent and refined development of agricultural carbon accounting. SUMMARY

[0008] The technical problem to be solved by the present application is to provide a farmland net carbon emission prediction method, system and device, which can realize high-precision and regional prediction and evaluation of farmland net carbon emission.

[0009] The present application provides a farmland net carbon emission prediction method, comprising the following steps:

[0010] S1, obtaining crop-related data, nitrous oxide emission measured data, methane emission measured data and carbon sequestration efficiency measured data of different crops in different planting areas, and dividing them into a training set and a validation set; the crop-related data includes crop type data, management measure data, soil data and climate data;

[0011] S2, constructing a nitrous oxide emission prediction model based on machine learning according to the crop-related data and the nitrous oxide emission measured data;

[0012] S3, constructing a methane emission prediction model based on machine learning according to the crop-related data and the methane emission measured data;

[0013] S4, constructing a carbon sequestration efficiency prediction model based on machine learning according to the crop-related data and the carbon sequestration efficiency measured data;

[0014] S5, calculating methane emission prediction value, nitrous oxide emission prediction value and carbon sequestration efficiency prediction value respectively according to the crop-related data, nitrous oxide emission prediction model, methane emission prediction model and carbon sequestration efficiency prediction model;

[0015] S6. Calculate the net carbon emission prediction value based on the predicted nitrous oxide emission value, the predicted methane emission value, the predicted carbon sequestration efficiency value, and the preset net carbon emission model.

[0016] Furthermore, S1 also includes: preprocessing the crop-related data, encoding the categorical variables into integers, normalizing all variables, and using the processed data as the basis for dividing the training set and validation set.

[0017] Furthermore, step S2 includes the following steps:

[0018] S201. Construct a nonlinear emission response function based on nitrogen input, N2O emission value, natural N2O emission coefficient under nitrogen-free conditions, and the sensitivity factor of N2O emission to nitrogen application rate.

[0019] S202. Construct a twin neural network and use crop-related data and measured nitrous oxide emission data from the training and validation sets to determine the values ​​of the natural N2O emission coefficient and sensitivity factor.

[0020] S203. Substitute the values ​​of the natural N2O emission coefficient and the sensitivity factor into the nonlinear emission response function to obtain the nitrous oxide emission prediction model.

[0021] Furthermore, the expression for the nonlinear emission response function is: ;

[0022] in, Indicates N2O emission value, denoted by α, where α represents the nitrogen input value, α represents the natural N2O emission coefficient under conditions of no nitrogen input, and β represents the sensitivity factor of N2O emissions to nitrogen application rate.

[0023] Furthermore, step S3 also includes the following steps:

[0024] S301. Set the candidate algorithm for methane emissions and its key parameters;

[0025] S302. Based on the crop-related data and measured methane emission data in the training set, the methane emission candidate algorithm is cross-validated and trained respectively.

[0026] S303. The optimal candidate algorithm for methane emission prediction is determined by using preset evaluation indicators.

[0027] Furthermore, step S4 includes the following steps:

[0028] S401. Set the candidate algorithms for carbon fixation efficiency and their key parameters;

[0029] S402. Based on the crop-related data and measured carbon sequestration efficiency data in the training set, the candidate carbon sequestration efficiency algorithm is cross-validated and trained.

[0030] S403. The best candidate algorithm is determined by using preset evaluation indicators as the carbon sequestration efficiency prediction model.

[0031] Furthermore, step S6 includes the following steps:

[0032] S601. Calculate the predicted greenhouse gas carbon emissions based on the predicted nitrous oxide emissions and methane emissions.

[0033] S602. Calculate the predicted carbon fixation value based on the predicted carbon fixation efficiency value and the carbon input value.

[0034] S603. Calculate the net carbon emission forecast based on the predicted greenhouse gas carbon emissions and carbon sequestration.

[0035] Furthermore, the formula for calculating the net carbon emission forecast is as follows:

[0036] ;

[0037] The formula for calculating the predicted greenhouse gas carbon emissions is as follows: ;

[0038] The formula for calculating the predicted carbon fixation value is: ;

[0039] in, This represents the predicted value of nitrous oxide emissions. This represents the predicted methane emissions. This represents the predicted carbon sequestration efficiency. This indicates the carbon input value.

[0040] Furthermore, it also includes: S7, generating a recommended scheme based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, net carbon emissions, and their corresponding weights, including the following steps:

[0041] S701. For any variety in any planting area, construct a scenario matrix based on continuous variables, categorical variables, predicted values ​​of nitrous oxide emissions, predicted values ​​of methane emissions, predicted values ​​of carbon sequestration, and predicted values ​​of net carbon emissions; the continuous variables include carbon input and nitrogen input, and the categorical variables include tillage method, fertilization frequency, fertilization method, fertilizer type, and planting density.

[0042] S702. Normalize the predicted values ​​of nitrous oxide emissions, methane emissions, carbon fixation, and net carbon emissions for any scenario in the scenario matrix using positive indicators to obtain four indicators.

[0043] S703. Construct a comprehensive scoring model based on the four indicators and their weights;

[0044] S704. Generate a recommendation scheme with corresponding weight combinations by using the maximum value of the comprehensive score.

[0045] The present invention also provides a farmland net carbon emission prediction system, comprising:

[0046] The acquisition module is used to acquire crop-related data, measured data on nitrous oxide emissions, measured data on methane emissions, and measured data on carbon sequestration efficiency for different crops in different planting areas, and divide them into training sets and validation sets; the crop-related data includes crop type data, management measure data, soil data, and climate data;

[0047] The first model building module is used to build a machine learning-based nitrous oxide emission prediction model based on the crop-related data and measured nitrous oxide emission data.

[0048] The second model building module is used to build a machine learning-based methane emission prediction model based on the crop-related data and measured methane emission data.

[0049] The third model building module is used to build a machine learning-based carbon sequestration efficiency prediction model based on the crop-related data and measured carbon sequestration efficiency data.

[0050] The prediction module is used to calculate the predicted values ​​of methane emissions, nitrous oxide emissions, and carbon sequestration efficiency based on the crop-related data, the nitrous oxide emission prediction model, the methane emission prediction model, and the carbon sequestration efficiency prediction model, respectively.

[0051] The calculation module is used to calculate the net carbon emission prediction value based on the predicted nitrous oxide emission value, the predicted methane emission value, the predicted carbon sequestration efficiency value, and the preset net carbon emission model.

[0052] Furthermore, it also includes:

[0053] The recommendation module is used to generate a recommendation scheme based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, net carbon emissions, and their corresponding weights.

[0054] The present invention also provides a farmland net carbon emission prediction device, including a processor and a memory, wherein the memory is used to store a computer program and the processor is used to execute the computer program to implement the steps of the above method.

[0055] The beneficial effects of this invention are:

[0056] 1. Universality and Adaptability of Crop- and Region-Specific Multi-Agent Modeling: This invention proposes a crop- and region-specific multi-agent modeling method. It constructs corresponding agent models based on the growth patterns, phenological characteristics, and regional differences in climate and soil conditions of different crops, achieving collaborative simulation at the regional scale. This method effectively captures the impact of spatial heterogeneity on carbon emissions and carbon sequestration processes, ensuring high universality and regional adaptability in cross-crop and cross-regional applications. It overcomes the limitation of single models experiencing decreased prediction accuracy under different environmental conditions.

[0057] 2. High-precision prediction capability through multi-dimensional input data fusion: In the modeling process, this invention introduces multi-dimensional input data covering soil physicochemical properties, climate factors, carbon and nitrogen inputs, tillage management, crop management, and fertilization management. It comprehensively utilizes the interpretability of farmland process mechanisms and the nonlinear fitting advantages of data-driven models to form a hybrid prediction framework that integrates mechanisms and data. This framework can significantly improve prediction accuracy and stability in multi-crop systems and multi-management scenarios, ensuring that carbon budget assessment results meet decision-making requirements in terms of both spatiotemporal resolution and numerical reliability.

[0058] 3. Precision Management and Policy Support Based on Model Output: Leveraging the high accuracy and broad adaptability of the aforementioned prediction results, this invention enables the recommendation of intelligent farmland management solutions based on model output. Users and government management departments can quickly screen and optimize combinations of management measures such as cropping systems, fertilization strategies, and carbon and nitrogen inputs, based on different carbon reduction and carbon sequestration targets, achieving precise management tailored to local conditions and feasible for implementation. Simultaneously, this function provides quantitative evidence for agricultural carbon management policy formulation, supporting the promotion of differentiated and precise carbon reduction and carbon sequestration strategies in different regions. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method in this invention;

[0060] Figure 2 This is a schematic diagram of the method in S2 of the present invention;

[0061] Figure 3 This is a comparison chart of the output results of the nitrous oxide emission prediction model in this invention and existing machine learning methods;

[0062] Figure 4 This is a schematic diagram of the output results of the methane emission prediction model in this invention;

[0063] Figure 5 This is a schematic diagram of the output results of the carbon fixation efficiency prediction model in this invention. Detailed Implementation

[0064] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0065] As 1 to Figure 5 As shown, taking a specific planting area as an example, this invention provides a method for predicting net carbon emissions from farmland, comprising the following steps:

[0066] S1. Obtain crop-related data, measured nitrous oxide emission data, measured methane emission data, and measured carbon sequestration efficiency data for different crops in different planting areas, and divide them into training and validation sets at an 8:2 ratio. Crop-related data includes crop type data, management practice data, soil data, and climate data.

[0067] Crop type data includes corn, wheat, and rice.

[0068] Management data include the number of fertilizations (single or multiple), fertilization method (surface or deep), fertilizer type (urea, inorganic fertilizer, nitrogen fertilizer with enhanced efficiency, organic fertilizer, green manure, biochar, straw), tillage method (no-till or conventional tillage), planting density (low, medium, or high density), nitrogen input, and carbon input.

[0069] Soil data include soil pH, bulk density, organic carbon content, total nitrogen content, clay content, and cation exchange capacity.

[0070] Climate data includes annual average temperature and annual average rainfall.

[0071] The above data includes categorical variables. To use this data as input for model training, the crop-related data underwent preprocessing. Specifically, the categorical variables were encoded as integers, and the results are shown in the table below:

[0072]

[0073] After all variables were numerically represented, Min-Max normalization was uniformly applied to normalize all numerical variables, scaling them to the [0,1] interval. This process was performed uniformly across the entire dataset to ensure that each feature dimension was within a relatively balanced numerical range in the model input. Subsequently, the processed samples were divided into training and validation sets, with an 80% training and 20% validation ratio, ensuring that the model could achieve effective learning results while maintaining its generalization ability.

[0074] Considering the significant spatial heterogeneity of farmland carbon cycling processes across different natural regions and agricultural ecosystems, this invention introduces a sub-model construction strategy based on crop-region combinations during the data modeling stage to further enhance the model's regional adaptability and predictive representativeness. Specifically, for maize planting areas, based on their spatial distribution characteristics in China, the data is divided into four sub-regions: Northeast China, Northwest China, the Huang-Huai-Hai Plain, and Southern China. For wheat planting systems, considering different cropping systems and soil and climate conditions in different ecological zones, they are further subdivided into the Northern Dryland Area, the Huang-Huai-Hai Winter Wheat Main Production Area, the Southwest Hilly Transitional Area, and the Yangtze River Mid-Lower Reaches Irrigation Area. The rice system is divided into the Northern Rice Area, the Yangtze River Basin Rice Area, the Southwest Mountainous Area, and the South China Tropical Rice Area based on differences in hydrothermal conditions and planting patterns. For each of these crop planting areas, a separate predictive model is constructed, effectively capturing the differences in the impact of key influencing factors (such as climate, soil, and management practices) on carbon emissions and carbon sequestration. This provides more regionally targeted intelligent management recommendations, enhancing the operability and practical value of the model output in actual agricultural decision-making.

[0075] S2. Construct a machine learning-based nitrous oxide emission prediction model based on crop-related data and measured nitrous oxide emission data. Specifically, such as... Figure 2 As shown, S2 includes the following steps:

[0076] S201. Construct a nonlinear emission response function based on nitrogen input, N2O emission value, natural N2O emission coefficient under nitrogen-free conditions, and the sensitivity factor of N2O emission to nitrogen application rate. The expression for the nonlinear emission response function is: ;

[0077] in, Indicates N2O emission value, Let denot a, where 'a' represents the nitrogen input value, 'b' represents the natural N2O emission coefficient under nitrogen-free conditions, and 'b' represents the sensitivity factor of N2O emissions to nitrogen application rate. This function effectively characterizes the nonlinear variation pattern of relatively stable emissions at low nitrogen application levels and rapidly increasing emissions at high nitrogen application levels.

[0078] S202. Since both a and b are unknowns, two parallel fully connected feedforward neural networks are constructed: sub-network A is used to estimate a, and sub-network B is used to estimate b. The values ​​of the natural N2O emission coefficient and sensitivity factor are determined using crop-related data and measured nitrous oxide emission data from the training and validation sets.

[0079] Specifically, the input data for subnetwork A consists of soil and climate data. Since background emissions are less affected by management measures, they are not included as management-related variables. Subnetwork A comprises five layers with the following neuron counts: 64, 128, 256, 128, and 64, progressively increasing the network's ability to fit the nonlinear relationship between environmental variables and background emissions. The input data for subnetwork B consists of soil, climate, and management measure data. Subnetwork B maintains the same network structure as subnetwork A, also a five-layer fully connected neural network with the same neuron counts of 64, 128, 256, 128, and 64, ensuring consistency in structural comparison and comparability of model responses.

[0080] To optimize model performance, this system employs a unified loss function for end-to-end training, with the objective of minimizing the mean squared error (MSE) between predicted and measured emissions.

[0081] ;

[0082] Where y i This represents the measured value of N2O emissions. Let be the predicted N2O emission value, and ...

[0083] S203. Substitute the values ​​of the natural N2O emission coefficient and the sensitivity factor into the nonlinear emission response function to obtain the nitrous oxide emission prediction model.

[0084] like Figure 3 As shown, the model (KDL-N2O) constructed in this step is compared with three traditional machine learning methods—Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Standard Neural Network (NN). The results show that the model outperforms these methods in both prediction accuracy and non-linear trend capture, with the overall prediction R... 2 The R value reached 0.79, significantly higher than that of other methods. 2 The range is 0.65 to 0.76. The model also shows stronger consistency and stability in terms of emission curve fitting.

[0085] As can be seen, this step, through a coupled modeling structure of "climate-soil-management input to parameter estimation and then to nonlinear emission prediction," achieves accurate and highly interpretable modeling of nitrous oxide emissions from farmland. This makes the model possess both the interpretability of agronomic formulas and the generalization ability of neural networks. This structure overcomes the technical limitations of fixed parameters and rigid responses in traditional models, achieving functional enhancement of the model structure rather than being limited to variable stacking. The model input system covers management measures such as fertilizer application rate, tillage method, fertilization time, and planting density, as well as environmental factors such as soil type and climate conditions. Through a unified feature space, it achieves fusion modeling, improving the model's response accuracy to changes in management combinations and solving the problem of insufficient adaptability in previous models that only addressed single variables or fixed scenarios.

[0086] S3. Construct a machine learning-based methane emission prediction model based on crop-related data and measured methane emission data. Specifically, S3 also includes the following steps:

[0087] S301. Define candidate algorithms for methane emission and their key parameters. Candidate algorithms for methane emission include Ridge Regression, Support Vector Regression, Feedforward Neural Network, Residual Neural Network, Gradient Boosting Tree, and Random Forest.

[0088] The main hyperparameter of the ridge regression algorithm is the strength of the L2 regularization term, i.e., the regularization coefficient, which ranges from 0.1 to 10, with a step size of 0.1, controlling the degree of penalty imposed on the parameter amplitude by the model. Since ridge regression uses a closed-form solution approach and does not require iterative training, it does not involve setting stopping conditions.

[0089] Key parameters of the support vector regression algorithm include: a penalty coefficient search range of 1 to 100, with a step size of 10; and an error tolerance interval of 0.01 to 0.2, with a step size of 0.01. To control training time, the maximum number of iterations is set to 1000. If the loss function changes by less than 1 × 10⁻⁶ during iterations... 3 Then stop early.

[0090] The parameter space of the feedforward neural network algorithm includes: hidden layer structure, set as a single hidden layer (32, 64, 128) or a double hidden layer ([64, 32], [128, 64]); learning rate range of 0.001 to 0.01, step size of 0.001. The training process is executed for a maximum of 100 rounds, and an early stopping mechanism is introduced. If the validation error does not decrease significantly within 10 consecutive rounds (the change is less than 1 × 10⁻⁶), the algorithm will stop the training. 4 This means terminating training prematurely.

[0091] To adapt to training deeper models, the residual neural network algorithm has the following main parameters: number of residual blocks ∈ [2,3,4]; number of neurons in each block ∈ [32,64,128]; learning rate from 0.001 to 0.01, step size 0.001. The maximum number of training epochs is 150, and an early stopping mechanism is also enabled, with the judgment criteria consistent with those of the feedforward neural network.

[0092] The gradient boosting tree algorithm, as an ensemble learning model, has the following parameters: number of weak learners ∈ [100, 200, ..., 1000], step size 100; maximum tree depth ∈ [3, 5, ..., 10], step size 1; subsampling ratio ∈ [0.6, 0.7, ..., 1.0], step size 0.1. An early stopping mechanism is used during training; if the validation set error does not decrease within 20 consecutive iterations, the iteration is terminated.

[0093] The parameters of the random forest algorithm include the number of decision trees, ranging from 100 to 1000, with a step size of 100; and the size of the feature subset, ranging from 1 to 20, with a step size of 1.

[0094] S302. Based on crop-related data and measured methane emission data in the training set, cross-validation training was performed on the methane emission candidate algorithms. Specifically, during the hyperparameter optimization process, to uniformly evaluate the comprehensive performance of the model under different parameter combinations, a ten-fold cross-validation strategy was adopted, and key performance indicators were normalized and weighted. The specific process is as follows: First, for each set of candidate parameter combinations, ten-fold cross-validation was performed: the training set was divided into ten equal sets, one of which was used as the validation set in turn, and the other nine were used for model training. After ten rounds, the average was taken to obtain three evaluation indicators: root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 ).

[0095] S303. The optimal candidate algorithm for methane emission prediction is determined using preset evaluation indicators. Since the three indicators have different measurement scales and different optimization directions, normalization is required to ensure comparability. RMSE and MAE are normalized using inverse normalization. 2 Using positive normalization, the following results were obtained:

[0096] RMSE_norm=(RMSE_max-RMSE) / (RMSE_max-RMSE_min);

[0097] MAE_norm=(MAE_max-MAE) / (MAE_max-MAE_min);

[0098] R 2 _norm=(R 2 -R2 _min) / (R 2 _max-R 2 _min).

[0099] The weighting ratio of each indicator is set to 1:1:1, and the comprehensive scoring function is constructed as follows:

[0100] Score = RMSE_norm + MAE_norm + R 2 _norm.

[0101] Based on the cross-validation evaluation results and model complexity, the algorithm with the highest comprehensive score under multiple indicators was selected from the candidate methane emission algorithms as the final methane emission prediction model.

[0102] like Figure 4 As shown, in this embodiment, the Random Forest algorithm was chosen as the final methane emission prediction model due to its strong nonlinear fitting ability, robustness to feature selection, and overall stability. The final parameters selected after multiple rounds of validation were: 100 decision trees and a feature subset size of 12. Other parameters remained at their default configurations, and the model directly entered the prediction and evaluation process after training. It not only exhibited a low error level during the training phase but also demonstrated good generalization ability on the independent test set.

[0103] The final model can output the methane emission intensity per unit area of ​​paddy field under different combinations of management measures, realize a high-precision estimate of the methane contribution of rice areas in the carbon emission system, and provide technical support for intelligent carbon accounting and optimization of field management measures.

[0104] This example employs a nonlinear modeling approach centered on random forests, integrating collaborative modeling of farmland management practices and environmental variables. By systematically constructing an input feature system encompassing key factors such as crop systems, fertilization strategies, tillage methods, carbon and nitrogen inputs, soil properties, and climatic conditions, the model can effectively identify the nonlinear regulatory mechanisms of management and environmental variables on CH4 emissions under different combinations. Furthermore, by extracting the model's influence weights on each input variable and combining this with visualization analysis of variable-emission response trends, the driving direction and degree of influence of key management measures on CH4 emissions can be identified.

[0105] In summary, this example overcomes the problems of low variable fusion and weak interaction processing capabilities in traditional methods for CH4 emission modeling. It constructs a random forest model that integrates key management measures and environmental variables, which significantly improves the model's prediction accuracy, application adaptability, and management interpretability. It has good scalability and practical application value.

[0106] S4. Construct a machine learning-based carbon sequestration efficiency prediction model based on crop-related data and measured carbon sequestration efficiency data. Further, S4 includes the following steps:

[0107] S401. Set the candidate algorithms for carbon sequestration efficiency and their key parameters. The candidate algorithms for carbon sequestration efficiency in this step are the same as those in S301 in terms of type and key parameters, so they will not be described again.

[0108] S402. Cross-validation training of candidate carbon sequestration efficiency algorithms is performed based on crop-related data and measured carbon sequestration efficiency data in the training set. The training method in this step is the same as that in S302, so it will not be described in detail.

[0109] S403. The optimal candidate algorithm is determined as the carbon sequestration efficiency prediction model using preset evaluation indicators. The evaluation indicators in this step are the same as those in S303, so they will not be repeated. Figure 5 As shown, in this embodiment, the random forest algorithm was chosen as the final carbon sequestration efficiency prediction model due to its strong nonlinear fitting ability, robustness to feature selection, and overall stability. The final parameters selected after multiple rounds of validation were: 100 decision trees and a feature subset size of 8. Other parameters remained at their default configurations, and the model directly entered the prediction and evaluation process after training. It not only exhibited a low error level during the training phase but also demonstrated good generalization ability on the independent test set.

[0110] The final model can output the carbon sequestration efficiency per unit area of ​​farmland under different combinations of management measures, providing technical support for intelligent carbon sequestration accounting and optimization of field management measures.

[0111] This example integrates a collaborative modeling system for farmland greenhouse gas (CH4, N2O) emissions and soil organic carbon (SOC) sequestration processes. During the construction phase, data connectivity and structural alignment between models were achieved, and interaction mechanisms between management measures and environmental factors were incorporated into each submodule, significantly improving the estimation accuracy and regional adaptability of each indicator. The system ultimately outputs the net carbon emission intensity per unit area of ​​farmland (emissions minus carbon sequestration) under a given management combination, supporting differentiated carbon footprint accounting refined to the crop, field, region, and management strategy level.

[0112] S5. Calculate the predicted values ​​of methane emissions, nitrous oxide emissions, and carbon sequestration efficiency based on crop-related data, nitrous oxide emission prediction models, methane emission prediction models, and carbon sequestration efficiency prediction models, respectively.

[0113] S6. Calculate the predicted net carbon emissions based on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration efficiency, and the pre-set net carbon emission model. S6 includes the following steps:

[0114] S601. Calculate the predicted greenhouse gas carbon emissions based on the predicted nitrous oxide and methane emissions. The formula for calculating the predicted greenhouse gas carbon emissions is as follows: .

[0115] in, This represents the predicted value of nitrous oxide emissions. This represents the predicted methane emissions.

[0116] S602. Calculate the predicted carbon sequestration value based on the predicted carbon sequestration efficiency and carbon input value. The formula for calculating the predicted carbon sequestration value is: .in, This represents the predicted carbon sequestration efficiency. This represents the predicted value of carbon input.

[0117] S603. Calculate the net carbon emission forecast based on the predicted greenhouse gas carbon emissions and carbon sequestration. The formula for calculating the net carbon emission forecast is: .

[0118] This step ultimately outputs a comprehensive assessment result including CH4 emissions, N2O emissions, carbon sequestration, and net carbon emissions. It can also support agricultural carbon management and visualization at the county, watershed, and even national scales, assisting in the formulation of carbon reduction policies and the optimization of implementation pathways.

[0119] S7. Generate recommended solutions based on predicted nitrous oxide emissions, methane emissions, carbon sequestration, and net carbon emissions, along with their corresponding weights. This step uses farmland as the analysis unit to systematically generate a scenario combination space for carbon and nitrogen inputs and management measures. Combined with the output of a regionally differentiated prediction model, and through a multi-objective standardization and user preference function-driven scoring mechanism, personalized low-carbon farmland management optimization is achieved. Specifically, S7 includes the following steps:

[0120] S701. For any variety in any planting area, construct a scenario matrix based on continuous variables, categorical variables, predicted values ​​of nitrous oxide emissions, predicted values ​​of methane emissions, predicted values ​​of carbon fixation, and predicted values ​​of net carbon emissions, to comprehensively simulate carbon and nitrogen responses under different input and management strategies.

[0121] Continuous variables include carbon input and nitrogen input, while categorical variables include farming practices, fertilization frequency, fertilization method, fertilizer type, and planting density.

[0122] In this example, the carbon input variable is set to 200–4000 kg·C·ha⁻¹·yr⁻¹, with a step size of 200 kg, for a total of 20 levels. The nitrogen fertilizer input is set to 10–300 kg·N·ha⁻¹·yr⁻¹, with a step size of 10 kg, for a total of 30 gradients. For management measures, the system defines five typical controllable variables: tillage method (no-till / conventional tillage), fertilization frequency (single, multiple), fertilization method (surface application, deep application), fertilizer type (urea, inorganic fertilizer, enhanced nitrogen fertilizer, organic fertilizer, green manure, biochar, straw), and planting density (low density, medium density, high density). Through multidimensional Cartesian product combinations, a total of 168 combined management scenarios are generated. Based on the above settings, the system forms a complete input variable combination space, containing 168 × 30 × 20 = 100,800 management scenarios. All management scenario simulations are based on the climate, soil, and crop type of user-specified plots, fixing external natural variables to ensure that the simulation results reflect the impact of management behavior itself on indicators.

[0123] Based on the aforementioned scenario combination space, the system invokes regionally customized carbon and nitrogen index prediction models to simulate all indices for each scenario. Considering the significant differences in climate conditions, soil types, and cropping systems across different geographical locations, this system has established regionally differentiated model libraries for four categories of indices—nitrous oxide emissions, methane emissions, carbon sequestration efficiency, and net carbon emissions—during the modeling phase. Specifically, for the same prediction index, the model structure, the importance of input variables, and parameter settings used in different agricultural regions have all been independently trained and optimized, fully reflecting the heterogeneity of regional ecological responses.

[0124] For example, when constructing the nitrous oxide emission prediction model, the system distinguishes between multiple ecological types, such as the Northeast maize region, the Huang-Huai wheat-maize rotation region, and the Southern rice-growing region. An independent neural network model is constructed for each region, with its structural depth and weight settings trained based on the actual measured data input to that region. Similarly, the methane emission prediction model also adopts an independent random forest model structure within the region based on the distribution of rice paddies, significantly improving its adaptability to methane emissions from rice-growing areas. Through the system's built-in model scheduling mechanism, the system can automatically match the appropriate indicator model according to the region to which the plot belongs, ensuring that all scenario simulation results have regional adaptability and realistic reliability.

[0125] S702. For any scenario in the scenario matrix, the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, and net carbon emissions are normalized using positive indices to obtain four indices. Since the four indices—N2O, CH4, Carbon_Sequestration, and Net_GHG—differ significantly in dimensions, direction, and scale, the system normalizes all simulation outputs, mapping them uniformly to the [0,1] interval. To adapt to the optimization algorithm, all indices are standardized as positive indices where "larger values ​​are better." In practice, reverse normalization is used for N2O, CH4, and Net_GHG, while forward normalization is directly applied to Carbon_Sequestration. The standardization formula for the Carbon_Sequestration index is: Carbon_Sequestration norm =(XX min ) / (Xmax-X min For the remaining three indicators (taking N2O as an example), the standardized formula is: N2O norm =(X max -X) / (X max -X min Where X is the predicted value of the current indicator, X min X is the minimum of all predicted values ​​for the current indicator. max This represents the maximum value of all predicted values ​​for the current indicator. After normalization, the normalized values ​​for the four indicators are obtained: N2O norm CH 4norm Carbon_Sequestration norm Net_GHG norm Through this standardization mechanism, the system achieves comparability and weighting of multiple objective indicators under a unified evaluation system, laying the foundation for subsequent personalized optimization.

[0126] S703. Construct a comprehensive scoring model based on four indicators and their weights. After completing all scenario simulations and indicator normalization, a user-driven objective preference function is further introduced to achieve personalized optimization recommendations under multiple objectives. Users can set the weights of each indicator according to actual management needs, including N2O emission weight w1, CH4 emission weight w2, Carbon_Sequestration weight w3, and Net_GHG weight w4. The weight values ​​range from [0,1] and satisfy the condition that the sum is 1. Therefore, the expression of the comprehensive scoring model is:

[0127] .

[0128] S704. A recommended solution with corresponding weight combinations is generated based on the maximum comprehensive score. Each management scenario is scored according to the above scoring function, with higher scores indicating better performance under the current preference settings. Finally, the management scenario with the highest score is selected from all 100,800 management scenarios as the optimal management recommendation solution for the current plot and objective, achieving an organic integration of regional adaptability and user orientation.

[0129] It should be noted that for different planting areas, due to different relevant data, the relevant models and their parameters will also be different. However, since the prediction process is the same, it will not be elaborated on.

[0130] Based on the same inventive concept, the present invention also provides a farmland net carbon emission prediction system, comprising:

[0131] The acquisition module is used to acquire crop-related data, measured data on nitrous oxide emissions, measured data on methane emissions, and measured data on carbon sequestration efficiency for different crops in different planting areas, and divides them into training sets and validation sets; the crop-related data includes crop type data, management practice data, soil data, and climate data.

[0132] The first model building module is used to build a nitrous oxide emission prediction model based on crop-related data and measured nitrous oxide emission data.

[0133] The second model building module is used to build a machine learning-based methane emission prediction model based on crop-related data and measured methane emission data.

[0134] The third model building module is used to build a machine learning-based carbon sequestration efficiency prediction model based on crop-related data and measured carbon sequestration efficiency data.

[0135] The prediction module is used to calculate the predicted values ​​of methane emissions, nitrous oxide emissions, and carbon sequestration efficiency based on crop-related data, nitrous oxide emission prediction models, methane emission prediction models, and carbon sequestration efficiency prediction models, respectively.

[0136] The calculation module is used to calculate the net carbon emission prediction value based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration efficiency, and a preset net carbon emission model.

[0137] The recommendation module is used to generate recommended solutions based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, net carbon emissions, and their corresponding weights.

[0138] Based on the same inventive concept, the present invention also provides a farmland net carbon emission prediction device, including a processor and a memory, wherein the memory is used to store a computer program and the processor is used to execute the computer program to implement the steps of the above method.

[0139] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting net carbon emissions from farmland, characterized in that, Includes the following steps: S1. Obtain crop-related data, measured data on nitrous oxide emissions, measured data on methane emissions, and measured data on carbon sequestration efficiency for different crops in different planting areas, and divide them into training sets and validation sets; the crop-related data includes crop type data, management measure data, soil data, and climate data; S2. Construct a machine learning-based nitrous oxide emission prediction model based on the crop-related data and measured nitrous oxide emission data; S2 includes the following steps: S201. Construct a nonlinear emission response function based on nitrogen input, N2O emission value, natural N2O emission coefficient under nitrogen-free conditions, and the sensitivity factor of N2O emission to nitrogen application rate. S202. Construct a twin neural network and use crop-related data and measured nitrous oxide emission data from the training and validation sets to determine the values ​​of the natural N2O emission coefficient and sensitivity factor. S203. Substitute the values ​​of the natural N2O emission coefficient and the sensitivity factor into the nonlinear emission response function to obtain the nitrous oxide emission prediction model; S3. Construct a machine learning-based methane emission prediction model based on the crop-related data and measured methane emission data; S4. Construct a carbon sequestration efficiency prediction model based on machine learning based on the crop-related data and measured carbon sequestration efficiency data. S5. Calculate the predicted values ​​of methane emissions, nitrous oxide emissions, and carbon sequestration efficiency based on the crop-related data, the nitrous oxide emission prediction model, the methane emission prediction model, and the carbon sequestration efficiency prediction model, respectively. S6. Calculate the net carbon emission prediction value based on the predicted nitrous oxide emission value, the predicted methane emission value, the predicted carbon sequestration efficiency value, and the preset net carbon emission model.

2. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, S1 further includes: preprocessing the crop-related data, encoding the categorical variables into integers, normalizing all variables, and using the processed data as the basis for dividing the training set and validation set.

3. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, The expression for the nonlinear emission response function is: ; in, Indicates N2O emission value, denoted by α, where α represents the nitrogen input value, α represents the natural N2O emission coefficient under conditions of no nitrogen input, and β represents the sensitivity factor of N2O emissions to nitrogen application rate.

4. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, S3 further includes the following steps: S301. Set the candidate algorithm for methane emissions and its key parameters; S302. Based on the crop-related data and measured methane emission data in the training set, the methane emission candidate algorithm is cross-validated and trained respectively. S303. The optimal candidate algorithm for methane emission prediction is determined by using preset evaluation indicators.

5. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, S4 includes the following steps: S401. Set the candidate algorithms for carbon fixation efficiency and their key parameters; S402. Based on the crop-related data and measured carbon sequestration efficiency data in the training set, the candidate carbon sequestration efficiency algorithm is cross-validated and trained. S403. The best candidate algorithm is determined by using preset evaluation indicators as the carbon sequestration efficiency prediction model.

6. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, S6 includes the following steps: S601. Calculate the predicted greenhouse gas carbon emissions based on the predicted nitrous oxide emissions and methane emissions. S602. Calculate the predicted carbon fixation value based on the predicted carbon fixation efficiency value and the carbon input value. S603. Calculate the net carbon emission forecast based on the predicted greenhouse gas carbon emissions and carbon sequestration.

7. The method for predicting net carbon emissions from farmland according to claim 6, characterized in that, The formula for calculating the net carbon emission forecast is as follows: ； The formula for calculating the predicted greenhouse gas carbon emissions is as follows: ; The formula for calculating the predicted carbon fixation value is: ; in, This represents the predicted value of nitrous oxide emissions. This represents the predicted methane emissions. This represents the predicted carbon sequestration efficiency. This indicates the carbon input value.

8. The method for predicting net carbon emissions from farmland according to claim 1, characterized in that, Also includes: S7. Generate a recommended scheme based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, net carbon emissions, and their corresponding weights, including the following steps: S701. For any variety in any planting area, construct a scenario matrix based on continuous variables, categorical variables, predicted values ​​of nitrous oxide emissions, predicted values ​​of methane emissions, predicted values ​​of carbon sequestration, and predicted values ​​of net carbon emissions; the continuous variables include carbon input and nitrogen input, and the categorical variables include tillage method, fertilization frequency, fertilization method, fertilizer type, and planting density. S702. Normalize the predicted values ​​of nitrous oxide emissions, methane emissions, carbon fixation, and net carbon emissions for any scenario in the scenario matrix using positive indicators to obtain four indicators. S703. Construct a comprehensive scoring model based on the four indicators and their weights; S704. Generate a recommendation scheme with corresponding weight combinations by using the maximum value of the comprehensive score.

9. A farmland net carbon emission prediction system, characterized in that, include: The acquisition module is used to acquire crop-related data, measured data on nitrous oxide emissions, measured data on methane emissions, and measured data on carbon sequestration efficiency for different crops in different planting areas, and divide them into training sets and validation sets; the crop-related data includes crop type data, management measure data, soil data, and climate data; The first model building module is used to build a machine learning-based nitrous oxide emission prediction model based on the crop-related data and measured nitrous oxide emission data; this module performs the following steps: Construct a nonlinear emission response function based on nitrogen input, N2O emission value, natural N2O emission coefficient under nitrogen-free conditions, and sensitivity factors of N2O emission to nitrogen application rate; A twin neural network is constructed, and the values ​​of the natural N2O emission coefficient and sensitivity factor are determined using crop-related data and measured nitrous oxide emission data in the training and validation sets. Substituting the values ​​of the natural N2O emission coefficient and the sensitivity factor into the nonlinear emission response function yields the nitrous oxide emission prediction model. The second model building module is used to build a machine learning-based methane emission prediction model based on the crop-related data and measured methane emission data. The third model building module is used to build a machine learning-based carbon sequestration efficiency prediction model based on the crop-related data and measured carbon sequestration efficiency data. The prediction module is used to calculate the predicted values ​​of methane emissions, nitrous oxide emissions, and carbon sequestration efficiency based on the crop-related data, the nitrous oxide emission prediction model, the methane emission prediction model, and the carbon sequestration efficiency prediction model, respectively. The calculation module is used to calculate the net carbon emission prediction value based on the predicted nitrous oxide emission value, the predicted methane emission value, the predicted carbon sequestration efficiency value, and the preset net carbon emission model.

10. A farmland net carbon emission prediction system according to claim 9, characterized in that, Also includes: The recommendation module is used to generate a recommendation scheme based on the predicted values ​​of nitrous oxide emissions, methane emissions, carbon sequestration, net carbon emissions, and their corresponding weights.

11. A farmland net carbon emission prediction device, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the steps of the method as described in claim 1.

Images