Method, system, and apparatus for predicting net carbon emissions from agricultural land

Abstract

This invention provides a method, system, and apparatus for predicting net carbon emissions from agricultural land, enabling highly accurate and localized prediction and evaluation of net carbon emissions from agricultural land. [Solution] The method involves acquiring crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas; constructing a nitrous oxide emission prediction model based on the crop-related data and nitrous oxide emission measurement data; constructing a methane emission prediction model based on the crop-related data and methane emission measurement data; constructing a carbon sequestration efficiency prediction model based on the crop-related data and carbon sequestration efficiency measurement data; calculating predicted methane emission values, predicted nitrous oxide emission values, and predicted carbon sequestration efficiency values; and calculating predicted net carbon emission values ​​based on predicted nitrous oxide emission values, predicted methane emission values, predicted carbon sequestration efficiency values, and a predetermined net carbon emission model.

Description

[Technical Field]

[0001] This invention belongs to the technical field of agricultural carbon emission prediction, and more specifically, relates to a method, system, and apparatus for predicting net carbon emissions from agricultural land. [Background technology]

[0002] As the issue of greenhouse gas emissions from agricultural systems gains increasing attention, research into agricultural land carbon emission assessment systems has become a technical focus. These systems typically include two core stages: firstly, simulation of greenhouse gas emission processes, particularly high spatiotemporal resolution estimation for key gases such as N2O and CH4; and secondly, quantification of carbon sequestration processes, primarily concerning the contribution of measures such as straw repatriation, organic fertilizer application, or biochar application to the increase in soil organic carbon. Because the ecological processes of agricultural land are highly complex, emission and carbon sequestration processes are influenced by various management measures (e.g., planting density, fertilization methods and types, cultivation methods) and environmental factors (soil type, climatic conditions, etc.), forming a nonlinear, multivariable coupling feedback mechanism, which presents relatively high requirements for model structure design, variable processing capability, and dynamic response capability. Conventional prediction methods have the following shortcomings or deficiencies:

[0003] 1. In terms of greenhouse gas emission modeling, conventional methods often rely on IPCC (Intergovernmental Panel on Climate Change) default emission factors or empirical statistical models. While these can provide approximations, they struggle to reflect the complex nonlinear interaction relationships between agricultural land management measures and environmental factors at a precise scale. Taking nitrous oxide (N2O) as an example, its emissions are influenced not only by fertilizer application rates and crop types, but also by coupling adjustments between management measures such as cultivation methods and fertilization timing, and climate and soil properties. A high degree of nonlinearity and multi-layered interaction exists between these variables, making accurate modeling difficult with only fixed coefficients or single-variable functions.

[0004] Furthermore, the multi-source heterogeneity between variables (coexistence of taxonomic and continuous variables), scale differences (coexistence of farmland operations and long-term climate mean values), and their time-series dynamics make modeling even more difficult. In recent years, there have been studies attempting to improve model prediction accuracy by introducing machine learning methods such as neural networks and random forests. However, many modeling strategies still exist as "black-box tools," failing to effectively integrate agricultural process mechanisms, lacking cross-regional generalization capabilities and the ability to interpret management responses, and thus failing to serve the goals of reducing farmland carbon emissions and precise adjustment control.

[0005] Secondly, in terms of carbon sequestration modeling, conventional studies often employ empirical parameters or sequestration coefficient methods, neglecting the interaction and control between farmland management measures (e.g., cultivation methods, fertilization intensity, moisture control, crop types) and environmental soil-climate factors. For example, the carbon stability and carbon sequestration potential of the same straw return measure to farmland can be completely different in the hot, humid southern regions of China and the dry, cold northern regions. Conventional methods generally lack the ability to model the "management × environment" interaction and cannot realize simulations of carbon sequestration efficiency that take measures appropriate to the local conditions.

[0006] Furthermore, Chinese patent CN113513269A proposes a farmland carbon emission prediction system based on "resource prediction," integrating multi-source data and modeling methods. However, its core remains based on the static application of emission factors, and a coupling learning mechanism between resource input and carbon emissions has not been established. The "predictive model" referred to in the system is in fact a combination of conventional models, making it difficult to respond to the dynamic response of changes in control measures to the emission process. It is not an intelligent system with generalization capabilities and decision-making support functions, but rather more akin to a data entry-based list tool overall.

[0007] In summary, conventional agricultural carbon emission and carbon sequestration assessment technologies generally suffer from problems such as low estimation accuracy, insufficient response capability, and poor system integration. Emission models often depend on sequestration factors and fail to reflect the complex interaction between control measures and environmental factors. Carbon sequestration assessments invariably employ a unified efficiency coefficient, ignoring differences between different control measures. Furthermore, while some data-driven methods improve predictive performance, they generally lack support for agricultural mechanisms, have limited model interpretability and regional generalization capabilities, and fail to meet the current needs for intelligent and refined agricultural carbon calculations. [Overview of the project] [Problems that the invention aims to solve]

[0008] This invention provides a method, system, and apparatus for predicting net carbon emissions from agricultural land that can achieve high accuracy and localized prediction and evaluation of net carbon emissions from agricultural land. [Means for solving the problem]

[0009] This invention provides a method for predicting net carbon emissions from agricultural land. Step S1 involves acquiring crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas, and dividing them into a training set and a validation set, wherein the crop-related data includes crop type data, management measures data, soil data, and climate data. Step S2 involves constructing a machine learning-based nitrous oxide emission prediction model based on the aforementioned crop-related data and actual nitrous oxide emission measurement data. Step S3 involves constructing a machine learning-based methane emission prediction model based on the aforementioned crop-related data and actual methane emission measurement data. Step S4 involves constructing a machine learning-based carbon sequestration efficiency prediction model based on the aforementioned crop-related data and measured carbon sequestration efficiency data. Step S5 of calculating predicted methane emissions, predicted nitrous oxide emissions, and predicted carbon sequestration efficiency values respectively based on the crop-related data, nitrous oxide emission prediction model, methane emission prediction model, and carbon sequestration efficiency prediction model; Step S6 of calculating a predicted net carbon emission value based on the predicted nitrous oxide emission value, predicted methane emission value, predicted carbon sequestration efficiency value, and a predetermined net carbon emission model, is included.

[0010] Furthermore, S1 further includes the step of preprocessing the crop-related data, performing integer encoding on categorical variables, performing normalization on all variables, and using the processed data as basic data for dividing into a training set and a validation set.

[0011] Furthermore, S2 Step S201 of constructing a non-linear emission response function based on nitrogen input, N2O emission value, natural N2O emission coefficient under the condition of no nitrogen input, and sensitivity factor of N2O emission to nitrogen application rate; Step S202 of constructing a twin neural network and determining the values of the natural N2O emission coefficient and sensitivity factor using the crop-related data and measured nitrous oxide emission data in the training set and validation set; Step S203 of substituting the values of the natural N2O emission coefficient and sensitivity factor into the non-linear emission response function to obtain the nitrous oxide emission prediction model, is included.

[0012] Furthermore, the expression formula of the non-linear emission response function is

Equation

[0013] Furthermore, S3 Step S301 involves setting a methane emission candidate algorithm and its important parameters, Step S302 involves performing cross-validation training on the candidate methane emission algorithms based on the crop-related data and actual methane emission data in the training set, respectively. The method further includes step S303, which determines the optimal methane emission candidate algorithm as a methane emission prediction model using predetermined evaluation indicators.

[0014] Furthermore, S4 is, Step S401 involves setting a candidate algorithm for carbon sequestration efficiency and its important parameters, Step S402 involves performing cross-validation training on the carbon sequestration efficiency candidate algorithms based on the crop-related data and measured carbon sequestration efficiency data in the training set, respectively. The process includes step S403, which determines the optimal carbon sequestration efficiency candidate algorithm as a carbon sequestration efficiency prediction model using predetermined evaluation indicators.

[0015] Furthermore, S6 is, Step S601: Calculate the predicted greenhouse gas carbon emission value based on the predicted nitrous oxide emission value and the predicted methane emission value. Step S602: Calculates a predicted carbon sequestration value based on the predicted carbon sequestration efficiency value and the carbon input value. The method includes step S603, which calculates a net carbon emission forecast based on the greenhouse gas carbon emission forecast and carbon sequestration forecast.

[0016] Furthermore, the formula for calculating the predicted net carbon emissions is:

number

number

number

[0017] Furthermore, step S7 generates a recommended plan based on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, predicted net carbon emissions, and corresponding weights, Step S701 involves constructing a scenario matrix for any variety in any cultivation area based on continuous variables, classification variables, predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, and predicted net carbon emissions, wherein the continuous variables include carbon input and nitrogen input, and the classification variables include cultivation method, number of fertilization applications, fertilization method, fertilizer type, and planting density. Step S702 involves performing a positive index normalization process on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, and predicted net carbon emissions for any scenario in the aforementioned scenario matrix to obtain four indicators. Step S703 involves constructing a comprehensive scoring model based on the four indicators and their weights, Step S7 further includes step S704, which generates a recommended strategy for the corresponding weight combination based on the maximum overall score.

[0018] The present invention further provides a net carbon emission prediction system for agricultural land. An acquisition module used to acquire crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas, and to divide them into training sets and validation sets, wherein the crop-related data includes crop type data, management measures data, soil data, and climate data. A first model building module used to construct a machine learning-based nitrous oxide emission prediction model based on the aforementioned crop-related data and actual nitrous oxide emission measurement data, A second model building module used to construct a machine learning-based methane emission prediction model based on the aforementioned crop-related data and actual methane emission measurement data, A third model building module used to construct a machine learning-based carbon sequestration efficiency prediction model based on the aforementioned crop-related data and measured carbon sequestration efficiency data, A prediction module used to calculate predicted methane emissions, predicted nitrous oxide emissions, and predicted carbon sequestration efficiency based on the aforementioned crop-related data, nitrous oxide emission prediction model, methane emission prediction model, and carbon sequestration efficiency prediction model, respectively. The system includes a calculation module used to calculate the net carbon emission forecast based on the nitrous oxide emission forecast, methane emission forecast, carbon sequestration efficiency forecast, and a predetermined net carbon emission model.

[0019] moreover, The system further includes a recommendation module used to generate a recommended plan based on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, predicted net carbon emissions, and corresponding weights.

[0020] The present invention further provides a farmland net carbon emission prediction device, comprising a processor and memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to realize the steps of the above method. [Effects of the Invention]

[0021] 1. Versatility and adaptability of multi-intelligent modeling by crop and region: The present invention provides a multi-intelligent modeling method for each crop and region, constructing corresponding intelligent models for differences in growth rules, meteorological characteristics, regional climate, and soil conditions of different crops, and enabling collaborative simulation at a regional scale. By effectively capturing the impact of spatial heterogeneity on carbon emissions and carbon sequestration processes, the model can maintain high versatility and regional adaptability in both cross-crop and cross-regional applications, overcoming the limitation that the predictive accuracy of a single model decreases under different environmental conditions.

[0022] 2. High-precision predictive capability of multidimensional input data fusion: In the modeling process, the present invention introduces multidimensional input data including soil physicochemical attributes, climatic factors, carbon and nitrogen inputs, cultivation management, crop management, and fertilization management. By comprehensively utilizing the interpretability of agricultural process mechanisms and the nonlinear fitting advantages of data-driven models, an integrated mechanism-data fusion predictive framework is formed. This framework significantly improves predictive accuracy and stability in multi-crop systems and multi-management scenarios, enabling carbon sequestration and emission assessment results to meet decision-making needs in terms of spatiotemporal resolution and numerical reliability.

[0023] 3. Accurate Management and Policy Support Based on Model Output: Based on the highly accurate and broadly adaptable prediction results described above, the present invention can realize the recommendation of intelligent farmland management plans based on model output. Users and government management departments can combine different carbon emission reduction and carbon sequestration targets, quickly select and optimize combinations of management measures such as cultivation systems, fertilization strategies, and carbon nitrogen inputs, and take measures appropriate to local conditions to achieve highly accurate management that can be reliably implemented. At the same time, this function provides a quantitative basis for formulating agricultural carbon management policies and supports differentiated, highly accurate carbon emission reduction and carbon sequestration strategies in different regions. [Brief explanation of the drawing]

[0024] [Figure 1] This is a flowchart of the method of the present invention. [Figure 2] This is a diagram illustrating the principle of method S2 of the present invention. [Figure 3] This is a comparison diagram of the output results of the nitrous oxide emission prediction model of the present invention and conventional machine learning. [Figure 4] This is a schematic diagram of the output results of the methane emission prediction model of the present invention. [Figure 5] This is a schematic diagram of the output results of the carbon fixation efficiency prediction model of the present invention. [Modes for carrying out the invention]

[0025] The present invention will be further described below with reference to the drawings. The following embodiments are merely for the purpose of more clearly illustrating the technical concept of the present invention and do not limit the scope of protection of the present invention.

[0026] As shown in Figures 1 to 5, using a certain cultivation area as an example, the present invention provides a method for predicting net carbon emissions from agricultural land, which includes the following steps S1 to S6.

[0027] S1 acquires crop-related data, actual nitrous oxide emission data, actual methane emission data, and actual carbon sequestration efficiency data for different crops in different cultivation areas, and divides them into training and validation sets according to an 8:2 ratio. Crop-related data includes crop type data, management measures data, soil data, and climate data.

[0028] Crop type data includes corn, wheat, and rice.

[0029] Management data includes the number of fertilization applications (single application, multiple applications), fertilization method (surface application, deep application), fertilizer type (urea, inorganic fertilizer, efficiency-enhancing nitrogen fertilizer, organic fertilizer, green manure, biochar, straw), cultivation method (no-till, conventional cultivation), planting density (low density, medium density, high density), nitrogen input, and carbon input.

[0030] Soil data includes soil pH, volumetric weight, organic carbon content, total nitrogen content, clay content, and cation exchange rate.

[0031] Climate data includes average annual temperature and average annual rainfall.

[0032] The above data includes classification variable data and is preprocessed to allow the model to be trained as input data. Specifically, integer encoding is performed on the classification variables, and the processing results are shown in the table below. [Table 1]

[0033] After quantifying and displaying all variables, all numerical variables are normalized using the Min-Max normalization method and scaled to the [0,1] interval. This process is integrated based on the entire dataset to ensure that each feature dimension is within a relatively balanced numerical range in the model input. Subsequently, the processed samples are divided into a training set and a validation set, with the training set accounting for 80% and the validation set for 20%, ensuring that the model can achieve effective learning while maintaining its generalization ability.

[0034] Considering that the carbon cycling processes in agricultural land exhibit significant spatial heterogeneity in different natural regions and agricultural ecosystems, the present invention introduces a sub-model construction strategy divided according to crop-region combinations during the data modeling stage to further improve the regional adaptability and predictive representativeness of the model. Specifically, for maize-growing regions, the data is classified into four sub-regions—Northeast China, Northwest China, Huanghuaihai Plain, and Southern China—based on their spatial distribution characteristics in China. For wheat cultivation systems, the data is subdivided into Northern Drought-Producing Areas, Huanghuaihai Winter Wheat-Producing Areas, Southwest Hill Transition Areas, and Yangtze River Middle and Lower Irrigation Areas, combining different cultivation systems and soil-climate conditions in different ecological zones. For rice cultivation systems, the data is divided into Northern Rice-Grown Areas, Yangtze River Basin Rice-Grown Areas, Southwest Mountain Areas, and South China Tropical Rice-Grown Areas, based on differences in hydrothermal conditions and cultivation modes. By constructing predictive models for each of the above crop cultivation regions, it is possible to effectively capture the differential impacts on carbon emissions and carbon sequestration of key regional differences (e.g., climate, soil, management measures), thereby providing recommendations for more region-oriented intelligent management measures in subsequent steps and improving the operability and implementation value of the model output in actual agricultural decisions.

[0035] S2 constructs a machine learning-based nitrous oxide emission prediction model based on crop-related data and actual nitrous oxide emission measurement data. Specifically, as shown in Figure 2, S2 includes the following steps S201 to S203.

[0036] A nonlinear emission response function is constructed based on S201, nitrogen input, N2O emission values, the natural N2O emission coefficient under conditions without nitrogen input, and the sensitivity factor of N2O emissions to nitrogen application. The expression for the nonlinear emission response function is:

number

[0037] Since both a and b are unknowns in S202, two parallel fully connected feedforward neural networks are constructed. Subnetwork A is used to estimate a, and subnetwork 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.

[0038] Specifically, the input data for subnetwork A consists of soil data and climate data, and background emissions are not included in the control variables because their influence from control measures is relatively small. Subnetwork A has a total of five layers, with the number of neurons being 64, 128, 256, 128, and 64, respectively, deepening the network's ability to fit nonlinear relationships between environment variables and background emissions with each layer. The inputs for subnetwork B are soil data, climate data, and control measure data. The network structure of subnetwork B is the same as that of subnetwork A, and is similarly a fully connected neural network with five layers, and the number of neurons is also 64, 128, 256, 128, and 64, ensuring consistency in structural comparison and comparability of model responses.

[0039] To optimize model performance, this system employs a unified loss function and performs end-to-end training, with the goal of minimizing the mean squared error (MSE) between predicted and actual emissions.

number

Number

[0040] Substitute the values of S203, the natural N2O emission coefficient and the sensitivity factor into the non-linear emission response function to obtain a nitrous oxide emission prediction model.

[0041] As shown in Figure 3, it can be seen from the comparison between the model (KDL-N2O) constructed in this step and three conventional machine learning methods - Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Standard Neural Network (NN) that the model is superior to the above methods in terms of prediction accuracy and capturing non-linear trends. Among them, the overall prediction R 2 reaches 0.79, which is significantly higher than the range of R 2 of 0.65 - 0.76 by other methods. In terms of emission curve fitting, the model also shows higher consistency and stability.

[0042] As will be seen, this step achieves accurate and highly interpretable modeling of nitrous oxide emissions from agricultural land through a coupling modeling structure from climate-soil-management inputs to parameter estimation and nonlinear emission prediction. This ensures that the model not only has the interpretability of agricultural formulas but also the generalization capabilities of a neural network. This structure overcomes the technical limitations of conventional models, such as fixed parameters and rigid responses, and enhances the functionality of the model structure without being limited to simply adding variables. The model input system covers management measures such as fertilizer application rate, cultivation method, fertilization time, and planting density, as well as environmental factors such as soil type and climate conditions. Furthermore, it achieves fused modeling through a unified feature space, improving the accuracy of the model's response to variations in combinations of management measures and solving the problem of insufficient adaptability to single-variable or fixed scenarios in conventional models.

[0043] S3 constructs a machine learning-based methane emission prediction model based on crop-related data and actual methane emission measurement data. Specifically, S3 further includes the following steps S301 to S303.

[0044] S301 sets the methane emission candidate algorithms and their important parameters. The methane emission candidate algorithms include ridge regression algorithms, support vector regression algorithms, feedforward neural network algorithms, residual neural network algorithms, gradient boosting tree algorithms, and random forest algorithms.

[0045] The main hyperparameter of the ridge regression algorithm is the strength of the L2 regularization term, i.e., the regularization coefficient, which has a range of 0.1 to 10 and a step size of 0.1, controlling the degree of penalty to the model's parameter range. The ridge regression algorithm employs a closed-form solution-finding method and does not require iterative training, thus not involving the setting of stopping conditions.

[0046] Key parameters of the support vector regression algorithm include a penalty coefficient with a search range of 1 to 100 and a step size of 10, and a set error tolerance with a range of 0.01 to 0.2 and a step size of 0.01. To control the training time, the maximum number of iterations is set to 1000, and the change in the loss function during iterations is set to 1 × 10⁻¹⁶. -3 If the value is smaller, it will stop early.

[0047] The parameter space of the feedforward neural network algorithm includes a hidden layer structure, which can be set to a single hidden layer (32, 64, 128) or a double hidden layer ([64,32], [128,64]), and a learning rate with a step size of 0.001 and a range of 0.001 to 0.01. During the training process, it is run up to 100 times, and an early stopping mechanism is introduced, and the validation error must not decrease significantly within 10 consecutive runs (change of 1 × 10⁻¹⁰). -4 (Smaller), end training early.

[0048] The residual neural network algorithm, adapted for training relatively deep models, has as its main parameters the number of residual blocks ∈[2, 3, 4], the number of neurons within each block ∈[32, 64, 128], and a learning rate of 0.001 to 0.01 with a step size of 0.001. The upper limit of training iterations is 150, and an early termination mechanism is also enabled, with decision criteria consistent with feedforward neural networks.

[0049] The gradient boosting tree algorithm is an integrated learning model whose parameters include the number of weak learners with a step size of 100 ∈[100, 200, ..., 1000], the maximum tree depth with a step size of 1 ∈[3, 5, ..., 10], and the subsampling ratio with a step size of 0.1 ∈[0.6, 0.7, ..., 1.0]. An early stopping mechanism is employed during the training process, and iterations are terminated if the error on the validation set does not decrease within 20 consecutive iterations.

[0050] The parameters of the random forest algorithm include the number of decision trees, each with a step size of 100 and a range of 100 to 1000, and the size of the feature subset, each with a step size of 1 and a range of 1 to 20.

[0051] S302, based on crop-related data and actual methane emission data in the training set, cross-validation training is performed on candidate methane emission algorithms. Specifically, in the hyperparameter optimization process, a 10-fold cross-validation strategy is adopted to uniformly evaluate the overall performance of the model with different parameter combinations, and normalization and weight scoring are performed on important performance indicators. The specific flow is to first perform 10-fold cross-validation on each set of candidate parameter combinations. In 10-fold cross-validation, the training set is divided into 10 equal subsets, one of which is used sequentially as the validation set, and the remaining nine are used to train the model. After completing 10 iterations, the average value is taken, and the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R) are calculated. 2 The goal is to obtain these three evaluation metrics.

[0052] S303 determines the optimal methane emission candidate algorithm as a methane emission prediction model using predetermined evaluation indicators. Since the metric scales of the three indicators are different and the optimization directions do not coincide, it is necessary to perform normalization first in order to ensure comparability. For RMSE and MAE, an inverse normalization method is adopted, and R 2 The normalization method is used to obtain the following results. RMSE_norm=(RMSE_max-RMSE) / (RMSE_max-RMSE_min), MAE_norm=(MAE_max-MAE) / (MAE_max-MAE_min), R 2 _norm=(R 2 -R 2 _min) / (R 2 _max-R 2 (min).

[0053] The weight parameter ratios for each indicator are set to 1:1:1, and the following overall scoring function is constructed. Score = RMSE_norm + MAE_norm + R 2 _norm.

[0054] By combining the cross-validation evaluation results and model complexity, the algorithm with the highest overall score across multiple indicators is selected from the candidate methane emission algorithms as the final methane emission prediction model.

[0055] As shown in Figure 4, in this embodiment, the random forest algorithm is selected as the final methane emission prediction model due to its relatively high nonlinear fitting capability, robustness to feature selection, and overall stability. The final parameters selected through multiple validations are a number of decision trees of 100 and a feature subset size of 12. Other parameters are left at their default settings, and after the model training is complete, the prediction and evaluation flow proceeds directly. It not only exhibits a low error level during the training phase but also has good generalization ability on an independent test set.

[0056] The final model outputs methane emission intensity for different combinations of management measures for a unit area of ​​rice paddies, enabling highly accurate estimation of the methane contribution of the rice sector in carbon emission systems, and providing technical support for intelligent carbon calculation and optimization of rice paddy management measures.

[0057] This example employs a nonlinear modeling method with a random forest as its core, integrating collaborative modeling of farmland management measures and environmental variables. The system constructs an input feature framework covering key factors such as crop systems, fertilization strategies, cultivation methods, carbon and nitrogen inputs, soil properties, and climatic conditions. The model can effectively identify and manage nonlinear adjustment control mechanisms for CH4 emissions in different combinations with environmental variables. Simultaneously, by extracting the weights of the influence of each input variable on the model and combining this with a visualization analysis of variable-emission response trends, the driving direction and extent of influence of key management measures on CH4 emissions can be identified.

[0058] In summary, this example overcomes the problems of low variable integration and poor interoperability in conventional methods for CH4 emission modeling. It constructs a random forest model that integrates critical control measures and environmental variables, significantly improving the model's predictive accuracy, applicability, and interpretability of management, thus possessing good dissemination potential and practical application value.

[0059] S4 constructs a machine learning-based carbon sequestration efficiency prediction model based on crop-related data and measured carbon sequestration efficiency data. Furthermore, S4 includes the following steps S401 to S403.

[0060] S401 sets the candidate carbon sequestration efficiency algorithm and its important parameters. The candidate carbon sequestration efficiency algorithm in this step is the same as the candidate algorithm type and important parameters in S301, so its explanation is omitted.

[0061] In S402, cross-validation training is performed on candidate carbon sequestration efficiency algorithms based on crop-related data and measured carbon sequestration efficiency data from the training set. The training method in this step is the same as the method in S302, so its explanation is omitted.

[0062] In S403, the optimal carbon sequestration efficiency candidate algorithm is determined as the carbon sequestration efficiency prediction model using predetermined evaluation metrics. Since the evaluation metrics in this step are the same as those in S303, their explanation is omitted. As shown in Figure 5, in this embodiment, the random forest algorithm is selected as the final carbon sequestration efficiency prediction model due to its relatively high nonlinear fitting ability, robustness to feature selection, and overall stability. The final parameters selected through multiple validations are a number of decision trees of 100 and a feature subset size of 8. Other parameters remain at their default settings, and after the model training is complete, the process proceeds directly to the prediction and evaluation flow. It not only exhibits a low error level during the training phase but also possesses good generalization ability on an independent test set.

[0063] The final model can output carbon sequestration efficiency for different combinations of management measures per unit area of ​​farmland, and can provide technical support for intelligent carbon calculation and optimization of farmland management measures.

[0064] This example integrates a collaborative modeling system for agricultural greenhouse gas (CH4, N2O) emissions and soil organic carbon (SOC) sequestration processes. During the construction phase, data consistency and structural alignment between models are achieved, and interaction mechanisms between management measures and environmental factors are introduced into each submodule, significantly improving the estimation accuracy and regional adaptability of each indicator. The system ultimately outputs the net carbon emission intensity (emissions minus carbon sequestration) for a given management combination per unit area of ​​agricultural land, supporting differentiated carbon footprint calculations subdivided by crop, field mass, region, and management strategy.

[0065] S5 calculates predicted methane emissions, predicted nitrous oxide emissions, and predicted carbon sequestration efficiency based on crop-related data, a nitrous oxide emission prediction model, a methane emission prediction model, and a carbon sequestration efficiency prediction model, respectively.

[0066] S6 calculates the net carbon emission forecast based on the nitrous oxide emission forecast, methane emission forecast, carbon sequestration efficiency forecast, and a predetermined net carbon emission model. S6 includes the following steps S601 to S603.

[0067] S601 calculates greenhouse gas carbon emission forecasts based on nitrous oxide emission forecasts and methane emission forecasts. The formula for calculating greenhouse gas carbon emission forecasts is:

number

[0068] S602 calculates the predicted carbon sequestration value based on the predicted carbon sequestration efficiency and carbon input values. The formula for calculating the predicted carbon sequestration value is:

number

[0069] S603 calculates net carbon emission forecasts based on greenhouse gas carbon emission forecasts and carbon sequestration forecasts. The formula for calculating net carbon emission forecasts is:

number

[0070] This step ultimately outputs a comprehensive assessment result including CH4 emissions, N2O emissions, carbon sequestration, and net carbon emissions, supporting agricultural carbon management and visualization of results at the provincial, watershed, and even national scales, and assisting in the formulation and optimization of carbon emission reduction policies and implementation pathways.

[0071] S7 generates recommended strategies based on predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, predicted net carbon emissions, and their corresponding weights. This step systematically generates a scenario combination space of carbon, nitrogen inputs, and management measures using farmland in the cultivated area as the analysis unit, combines it with the output of a regional differentiation prediction model, and optimizes personalized low-carbon farmland management through a multi-objective standardization and user preference function-driven scoring mechanism. Specifically, S7 includes the following steps S701 to S704.

[0072] S701 constructs a scenario matrix for any variety in any growing region, based on continuous variables, taxonomic variables, predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, and predicted net carbon emissions. The scenario matrix is ​​used to comprehensively simulate the carbon-nitrogen response under different inputs and management strategies.

[0073] Continuous variables include carbon input and nitrogen input, while classification variables include cultivation method, number of fertilization applications, fertilization method, fertilizer type, and planting density.

[0074] In this example, the carbon input is 200-4000 kg·C·ha -1 ·yr -1 The settings are configured with a step size of 200 kg and a total of 20 levels. The amount of nitrogen fertilizer input is 10-300 kg·N·ha. -1 ·yr -1The settings are configured with a step size of 10 kg and a total of 30 gradients. For the management measures portion, the system defines five typical controllable variables, including cultivation method (no-till / conventional cultivation), number of fertilization applications (single application, multiple applications), fertilization method (surface application, deep application), fertilizer type (urea, inorganic fertilizer, efficiency-enhancing nitrogen fertilizer, organic fertilizer, green manure, biochar, straw), and planting density (low density, medium density, high density). A total of 168 combination management scenarios are generated by combining multidimensional Cartesian products. Based on the above settings, the system forms a complete input variable combination space containing a total of 168 × 30 × 20 = 100,800 management scenarios. All management scenario simulations are performed based on the climate, soil, and crop type of the plot specified by the user, and external natural variables are fixed to ensure that the simulation results reflect the impact of the management actions themselves on the indicators.

[0075] Based on the above scenario combination space, the system invokes a regionally adapted carbon-nitrogen index prediction model and performs simulations of all indicators for each scenario. Considering the significant differences in climatic conditions, soil types, and cultivation systems across different geographical regions of agricultural ecosystems, the system constructs regionally differentiated model libraries for four indicators—nitrous oxide emissions, methane emissions, carbon sequestration efficiency, and net carbon emissions—during the modeling phase. Specifically, the same prediction indicator is trained and optimized independently within different agricultural regions, with the adopted model structure, the importance of input variables, and parameter settings all independently adjusted to fully reflect the heterogeneity of regional ecological responses.

[0076] For example, when constructing a nitrous oxide emission prediction model, the system distinguishes between multiple ecological types, such as the northeastern maize plot, the yellow wheat-maize rotation plot, and the southern rice plot. For each region, an independent neural network model is constructed, and its structural depth and weight settings are obtained by training based on actual measured data input from that region. Similarly, the methane emission prediction model also employs an independent random forest model structure within each region according to the distribution of rice fields, significantly improving the adaptability of rice plots to methane emissions. The model scheduling mechanism built into the system allows the system to automatically match the corresponding index model according to the region to which a plot belongs, ensuring that all scenario simulation results have regional adaptability and realistic reliability.

[0077] In S702, four indicators are obtained by performing positive normalization on the predicted nitrous oxide emissions, methane emissions, carbon sequestration, and net carbon emissions for any scenario in the scenario matrix. Since the four indicators, N2O, CH4, Carbon_Sequestration, and Net_GHG, have significant differences in terms of dimension, direction, and scale, the system normalizes all simulation outputs and maps them uniformly to the [0,1] interval. To fit the optimization algorithm, all indicators are standardized to positive indicators where "higher values ​​are preferable." During the specific operation, inverse normalization is applied to N2O, CH4, and Net_GHG, and direct normalization is applied to Carbon_Sequestration. For the Carbon_Sequestration indicator, the standardization formula is: Carbon_Sequestration norm =(XX min ) / (X max -X min ) is the standardization formula for the other three indicators (using N2O as an example). norm =( X max -X) / (X max -X min ) where X is the predicted value of the current indicator, and Xmin X is the minimum of all predicted values ​​for the current indicator. max This is the maximum value of all predicted values ​​for the current indicator. According to the normalization process, N2O norm CH 4norm Carbon_Sequestration norm Net_GHG norm The normalized values ​​of these four indicators are obtained. Through this standardization mechanism, the system achieves comparability and weighting in a unified evaluation system for multiple target indicators, laying the foundation for subsequent individualized optimization.

[0078] S703 constructs a comprehensive scoring model based on four indicators and their weights. After completing all scenario simulations and indicator normalization, a user-driven target preference function is further introduced to enable personalized optimization recommendations for multiple targets. Users can set the weights for each indicator, including N2O emission weight w1, CH4 emission weight w2, Carbon_Sequestration weight w3, and Net_GHG weight w4, according to their actual management needs, with the weight values ​​ranging from [0,1] and summing to 1. Therefore, the expression for the comprehensive scoring model is:

number

[0079] S704 generates recommended strategies for the corresponding weight combinations based on the maximum overall score. Each management scenario is scored based on the above scoring function, with higher scores indicating better representation under the current preferred settings. Finally, the management scenario with the highest score from all 100,800 management scenarios is selected as the optimal management recommendation for the current parcel and target, achieving an organic integration of regional adaptability and user orientation.

[0080] Note that while the relevant data differs for different cultivation regions, and therefore the relevant models and their parameters may also differ, the prediction process is the same, so we will omit the explanation.

[0081] Based on a similar inventive concept, the present invention further provides a net carbon emission prediction system for agricultural land. An acquisition module used to acquire crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas, and to divide them into training sets and validation sets, wherein the crop-related data includes crop type data, management measures data, soil data, and climate data, and the acquisition module... A first model building module used to construct a nitrous oxide emission prediction model based on crop-related data and actual nitrous oxide emission measurement data, A second model building module is used to construct a machine learning-based methane emission prediction model based on agricultural crop-related data and actual methane emission measurement data. A third model building module used to construct a machine learning-based carbon sequestration efficiency prediction model based on agricultural crop-related data and measured carbon sequestration efficiency data, A prediction module used to calculate predicted methane emissions, predicted nitrous oxide emissions, and predicted carbon sequestration efficiency, respectively, based on crop-related data, a nitrous oxide emission prediction model, a methane emission prediction model, and a carbon sequestration efficiency prediction model, A calculation module used to calculate net carbon emission predictions based on predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration efficiency, and a predetermined net carbon emission model, It includes a recommendation module used to generate recommended strategies based on predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, predicted net carbon emissions, and their corresponding weights.

[0082] Based on a similar inventive concept, the present invention further provides a farmland net carbon emission forecasting device comprising a processor and memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to carry out the steps of the above method.

[0083] The agricultural land net carbon emission prediction method, agricultural land net carbon emission prediction system, and agricultural land net carbon emission prediction device of the present invention can be applied to all regions of all countries, and are not limited to the above-mentioned countries and regions.

[0084] The above are merely preferred embodiments of the present invention, and it should be noted that those skilled in the art can make several 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. Step S1 involves acquiring crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas, and dividing them into a training set and a validation set, wherein the crop-related data includes crop type data, management measures data, soil data, and climate data. Step S2 involves constructing a machine learning-based nitrous oxide emission prediction model based on the aforementioned crop-related data and actual nitrous oxide emission data, wherein S2 is performed as follows: Nitrogen input, N ２ O emissions, natural nitrogen under conditions without nitrogen input ２ O emission factor and N ２ Step S201 involves constructing a nonlinear emission response function based on a sensitivity factor for O emissions to nitrogen application, A twin neural network is constructed, and the crop-related data and nitrous oxide emission measurement data from the training set and validation set are used to analyze the natural N ２ Step S202 involves determining the values ​​of the oxygen emission coefficient and the susceptibility factor, The aforementioned natural N ２ Step S2 includes step S203, which involves substituting the values ​​of the O emission coefficient and sensitivity factor into the nonlinear emission response function to obtain the nitrous oxide emission prediction model, Step S3 involves constructing a machine learning-based methane emission prediction model based on the aforementioned crop-related data and actual methane emission measurement data. Step S4 involves constructing a machine learning-based carbon sequestration efficiency prediction model based on the aforementioned crop-related data and measured carbon sequestration efficiency data. Step S5 involves calculating the predicted methane emission value, predicted nitrous oxide emission value, and predicted carbon sequestration efficiency value, respectively, based on the aforementioned crop-related data, nitrous oxide emission prediction model, methane emission prediction model, and carbon sequestration efficiency prediction model. A method for predicting net carbon emissions from farmland, characterized in that a computer performs step S6 of calculating a net carbon emission prediction value based on the nitrous oxide emission prediction value, methane emission prediction value, carbon sequestration efficiency prediction value, and a predetermined net carbon emission model.

2. The method for predicting net carbon emissions from farmland according to claim 1, wherein step S1 further includes the steps of preprocessing the crop-related data, performing integer coding on the classification variables, performing normalization on all variables, and using the processed data as basic data to divide into a training set and a validation set.

3. The expression for the nonlinear emission response function is: [Math 12] And, Here, G Ｎ２Ｏ represents the NO emission value, N ２ represents the nitrogen input value, a represents the natural NO ｒａｔｅ emission coefficient under the condition of no nitrogen input, b represents the sensitivity factor of NO ２ emission to the nitrogen application rate, The agricultural land net carbon emission prediction method according to claim 1, characterized in that. ２ ​

4. Step S3 is, Step S301 involves setting a methane emission candidate algorithm and its important parameters, Step S302 involves performing cross-validation training on the methane emission candidate algorithms based on the crop-related data and actual methane emission measurement data in the training set, respectively. The method for predicting net carbon emissions from agricultural land according to claim 1, further comprising step S303 of determining an optimal methane emission candidate algorithm as a methane emission prediction model using predetermined evaluation indicators.

5. The aforementioned step S4 is, Step S401 involves setting a candidate algorithm for carbon sequestration efficiency and its important parameters, Step S402 involves performing cross-validation training on the carbon sequestration efficiency candidate algorithms based on the crop-related data and measured carbon sequestration efficiency data in the training set, respectively. The method for predicting net carbon emissions from agricultural land according to claim 1, comprising step S403 of determining an optimal carbon sequestration efficiency candidate algorithm as a carbon sequestration efficiency prediction model using predetermined evaluation indicators.

6. The aforementioned step S6 is, Step S601: Calculate the predicted greenhouse gas carbon emission value based on the predicted nitrous oxide emission value and the predicted methane emission value. Step S602: Calculates a predicted carbon fixation value based on the predicted carbon fixation efficiency value and the carbon input value. The method for predicting net carbon emissions from farmland according to claim 1, comprising step S603, which calculates a net carbon emission forecast value based on the greenhouse gas carbon emission forecast value and the carbon sequestration forecast value.

7. The formula for calculating the predicted net carbon emissions is: [Number 13] And, The formula for calculating predicted greenhouse gas carbon emissions is: [Number 14] And, The formula for calculating carbon sequestration predictions is: [Number 15] And, Here, G Ｎ２Ｏ This represents the predicted value of nitrous oxide emissions, G ＣＨ４ represents the predicted methane emission value, NCE represents the predicted carbon sequestration efficiency value, C ｉｎｐｕｔ The method for predicting net carbon emissions from agricultural land according to claim 6, characterized in that represents the carbon input value.

8. Step S7 is to generate a recommended plan based on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, predicted net carbon emissions, and corresponding weights, Step S701 involves constructing a scenario matrix for any variety in any cultivation area based on continuous variables, classification variables, predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, and predicted net carbon emissions, wherein the continuous variables include carbon input and nitrogen input, and the classification variables include cultivation method, number of fertilization applications, fertilization method, fertilizer type, and planting density. Step S702 involves performing a positive normalization process on the predicted nitrous oxide emissions, predicted methane emissions, predicted carbon sequestration, and predicted net carbon emissions for any scenario in the scenario matrix to obtain four indicators. Step S703 involves constructing a comprehensive scoring model based on the four indicators and their weights, The method for predicting net carbon emissions from agricultural land according to claim 1, further comprising step S7, which includes step S704, where step S704 generates a recommended combination of weights corresponding to the maximum value of the overall score.

9. An acquisition module used to acquire crop-related data, nitrous oxide emission measurement data, methane emission measurement data, and carbon sequestration efficiency measurement data for different crops in different cultivation areas, and to divide them into training sets and validation sets, wherein the crop-related data includes crop type data, management measures data, soil data, and climate data. A first model building module used to construct a machine learning-based nitrous oxide emission prediction model based on the aforementioned crop-related data and actual nitrous oxide emission measurement data, wherein the module is: Nitrogen input, N ２ O emissions, natural nitrogen under conditions without nitrogen input ２ O emission factor and N ２ The steps include constructing a nonlinear emission response function based on a sensitivity factor for O emissions to nitrogen application, A twin neural network is constructed, and the crop-related data and nitrous oxide emission measurement data from the training set and validation set are used to analyze the natural N ２ A step to determine the values ​​of the O emission coefficient and the susceptibility factor, The aforementioned natural N ２ A first model building module performs the steps of: substituting the values ​​of the emission coefficient and sensitivity factor into the nonlinear emission response function to obtain the nitrous oxide emission prediction model; A second model building module used to construct a machine learning-based methane emission prediction model based on the aforementioned crop-related data and actual methane emission measurement data, A third model building module used to construct a machine learning-based carbon sequestration efficiency prediction model based on the aforementioned crop-related data and measured carbon sequestration efficiency data, A prediction module used to calculate predicted methane emissions, predicted nitrous oxide emissions, and predicted carbon sequestration efficiency, respectively, based on the aforementioned crop-related data, nitrous oxide emission prediction model, methane emission prediction model, and carbon sequestration efficiency prediction model, A farmland net carbon emission prediction system, characterized by including a calculation module used to calculate a net carbon emission prediction value based on the nitrous oxide emission prediction value, methane emission prediction value, carbon sequestration efficiency prediction value, and a predetermined net carbon emission model.

10. The farmland net carbon emission forecasting system according to claim 9, further comprising a recommendation module used to generate a recommended plan based on the nitrous oxide emission forecast, methane emission forecast, carbon sequestration forecast, net carbon emission forecast and corresponding weights.

11. A farmland net carbon emission prediction device comprising a processor and memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program and realize the steps of the method according to claim 1.

Images