A soil organic carbon estimation method, device and electronic equipment

Abstract

The application provides a soil organic carbon estimation method, device and electronic equipment, wherein the soil organic carbon estimation method comprises: obtaining site information of a target site; the site information comprises soil properties, environmental data, vegetation data, location information and land use data of the target site; an initial physical model is called, and an estimated value of a key parameter of the initial physical model is obtained by inversion based on the site information of the target site in an observation year; a key parameter estimator is constructed, and the key parameter estimator is trained and fine-tuned; the trained key parameter estimator is embedded into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid. Through the application, the unity of model precision and interpretability is realized, and the problem of poor soil organic carbon estimation effect in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental science, and in particular, to a soil organic carbon estimation method, device and electronic equipment. BACKGROUND

[0002] The soil carbon pool is the largest carbon pool in the terrestrial ecosystem, and its total organic carbon storage is about 2-3 times that of the terrestrial vegetation carbon pool and more than 2 times that of the atmospheric carbon pool. Small changes in the soil carbon pool can significantly affect the concentration of greenhouse gases in the atmosphere, thereby having a profound impact on global climate change. Therefore, accurately simulating and predicting the dynamic changes of soil organic carbon has important scientific value for understanding global carbon cycles.

[0003] Currently, the research on soil organic carbon cycle mainly relies on long-term positioning experiments and physical models. Long-term positioning experiments can monitor the dynamic evolution of regional soil organic carbon, but are limited by observation costs and time, and the obtained data is discrete in time and space, making it difficult to fully reflect the long-term change trend and spatial distribution characteristics of soil organic carbon. Traditional process-based physical models are based on the biogeochemical mechanism of soil carbon cycle and can describe the input, transformation and decomposition processes of soil organic carbon. However, these models have high requirements for input parameters, and the inherent uncertainty of the model structure may lead to simulation errors.

[0004] In recent years, data-driven methods have shown significant advantages in environmental science and ecological modeling. Using large-scale observation data, machine learning methods can mine complex relationships and improve the predictive ability of models. However, traditional data-driven methods lack physical constraints and only fit data, which may lead to simulation results deviating from actual biogeochemical processes, limiting their application in soil organic carbon cycle simulation.

[0005] There is currently no effective solution to the problem of poor soil organic carbon estimation effect of existing related technologies. SUMMARY

[0006] The present application provides a soil organic carbon estimation method, device and electronic equipment to solve the problem of poor soil organic carbon estimation effect of existing related technologies.

[0007] In a first aspect, the present application provides a soil organic carbon estimation method, comprising:

[0008] Obtaining site information of a target site; the site information includes soil properties, environmental data, vegetation data, location information and land use data of the target site;

[0009] Calling an initial physical model, and based on the site information of the target site in the observation year, obtaining the estimated value of the key parameters of the initial physical model by inversion.

[0010] constructing a key parameter estimator and training the key parameter estimator;

[0011] embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid;

[0012] predicting the organic carbon information of the to-be-tested soil through the soil organic carbon simulation model to obtain the soil organic carbon turnover of the to-be-tested soil.

[0013] According to the soil organic carbon estimation method provided by the application, after obtaining the site information of the target site, the site information is pretreated, including:

[0014] For the remote sensing data and reanalysis data in the site information, the site information is processed to form structured table data;

[0015] The organic carbon content change value of the target site starting from the observation year is subjected to quality control, and abnormal values are removed.

[0016] According to the soil organic carbon estimation method provided by the application, the site information is processed to form structured table data, including:

[0017] The site information is subjected to re-projection and Kriging function interpolation processing to extract site information with the same spatial and temporal resolution under a unified spatial coordinate system;

[0018] The site information is matched and integrated through the time information and the latitude and longitude coordinates of the target site to obtain structured table data.

[0019] According to the soil organic carbon estimation method provided by the application, an initial physical model is called, and based on the site information of the target site in the observation year, the estimated value of the key parameter of the initial physical model is obtained by inversion, including:

[0020] Introducing a carbon input correction factor, and taking the carbon input correction factor as the key parameter of the initial physical model;

[0021] The carbon input correction factor is inverted by a Markov chain Monte Carlo method to obtain the estimated value of the carbon input correction factor.

[0022] According to the soil organic carbon estimation method provided by the application, the carbon input correction factor is inverted by a Markov chain Monte Carlo method to obtain the estimated value of the carbon input correction factor, including:

[0023] a Bayesian optimization problem is constructed with the sum of squared errors between the output of the initial physical model and the measured soil organic carbon content as the objective function;

[0024] The posterior probability distribution is constructed to set the likelihood function;

[0025] The Metropolis-Hastings algorithm is used to perform the sampling process of the Markov chain Monte Carlo method, and a sequence of corrected factor samples conforming to the posterior probability distribution is generated through multiple iterations, and the estimated value of the carbon input correction factor is determined therefrom.

[0026] According to the soil organic carbon estimation method provided by the application, the key parameter estimator is constructed, and the key parameter estimator is trained and fine-tuned, including:

[0027] The random forest algorithm is used to construct the key parameter estimator;

[0028] Based on the site information of the target site, an input feature vector is generated;

[0029] The input feature vector is used as input, and the carbon input correction factor is used as label, and the key parameter estimator is trained and fine-tuned; the key parameter is used for dynamic parameter prediction according to input.

[0030] According to the soil organic carbon estimation method provided by the application, the key parameter estimator is trained and fine-tuned, including:

[0031] The input feature vector is input into the key parameter estimator to obtain the mean value of the predicted value corresponding to each decision tree;

[0032] Based on the mean value of the predicted value corresponding to each decision tree and the estimated value of the key parameter, the hyperparameters of the parameter estimator are optimized with the least square error as the objective function.

[0033] According to the soil organic carbon estimation method provided by the application, the trained key parameter estimator is embedded into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid, including:

[0034] Based on the site information of the target site, the initial physical model is run, and the trained key parameter estimator is called in real time to dynamically generate a carbon input correction factor with spatial difference;

[0035] The carbon input correction factor and the primary productivity of vegetation of the target site are combined to determine the carbon input amount of the target site;

[0036] A bias monitoring mechanism is introduced to update and correct the key parameter estimator to obtain a soil organic carbon simulation model driven by physical and data hybrid.

[0037] In a second aspect, the present application also provides a soil organic carbon estimation device, comprising:

[0038] an acquisition module configured to acquire site information of a target site, wherein the site information comprises soil properties, environmental data, vegetation data, location information and land use data of the target site;

[0039] a processing module configured to call an initial physical model and inversely derive estimated values of key parameters of the initial physical model based on site information of the target site in an observation year;

[0040] a training module configured to construct a key parameter estimator and fine-tune the key parameter estimator;

[0041] an embedding module configured to embed the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data mixing;

[0042] an estimation module configured to predict organic carbon information of a to-be-tested soil through the soil organic carbon simulation model to obtain a soil organic carbon turnover of the to-be-tested soil.

[0043] In a third aspect, the present application also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the soil organic carbon estimation method according to the first aspect.

[0044] In a fourth aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the soil organic carbon estimation method according to the first aspect.

[0045] In a fifth aspect, the present application also provides a computer program product comprising a computer program, wherein the computer program is executable by a processor to implement the soil organic carbon estimation method according to the first aspect.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] The soil organic carbon estimation method provided by the present application effectively overcomes the problem of large parameter uncertainty in traditional physical models and overcomes the defect of pure data-driven methods lacking process mechanism explanation, realizes the unification of model precision and interpretability, and solves the problem of poor soil organic carbon estimation effect in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0049] Figure 1 is a flowchart of the soil organic carbon estimation method provided by the present application;

[0050] Figure 2 is a schematic diagram of the soil organic carbon estimation process in the embodiment of the present application;

[0051] Figure 3 is a schematic diagram of the key parameter label obtained by inversion in the embodiment of the present application;

[0052] Figure 4 is a construction schematic diagram of the key parameter estimator in the embodiment of the present application;

[0053] Figure 5 is a traditional physical model simulation scatter plot from 2009 to 2015 in the embodiment of the present application;

[0054] Figure 6 is a hybrid driving model simulation scatter plot from 2009 to 2015 in the embodiment of the present application;

[0055] Figure 7 is a traditional physical model simulation scatter plot from 2015 to 2018 in the embodiment of the present application;

[0056] Figure 8 is a hybrid driving model simulation scatter plot from 2015 to 2018 in the embodiment of the present application;

[0057] Figure 9 is a structural block diagram of the soil organic carbon estimation device provided by the present application;

[0058] Figure 10 is a structural schematic diagram of the electronic device provided by the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0060] The present application provides a soil organic carbon estimation method,Figure 1 is a flowchart of the soil organic carbon estimation method provided by the present application, as shown in Figure 1 , the method comprises the following steps:

[0061] Step S101, obtaining site information of a target site; the site information comprises soil properties, environmental data, vegetation data, location information and land use data of the target site;

[0062] Step S102, calling an initial physical model, and based on the site information of the target site in the observation years, inversely obtaining an estimated value of a key parameter of the initial physical model;

[0063] Step S103, constructing a key parameter estimator, and training and fine-tuning the key parameter estimator;

[0064] Step S104, embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid;

[0065] Step S105, predicting the organic carbon information of the soil to be measured by the soil organic carbon simulation model to obtain the soil organic carbon turnover of the soil to be measured.

[0066] In the method, first, the site information of the target site is obtained, which can effectively represent the soil condition and other environmental conditions of the target site. Then, the initial physical model is called, and the key parameters of the initial physical model are determined according to the site information of the target site in the observation years, and the estimated value of the key parameters is inversely obtained. Exemplarily, the physical model adopted is a soil carbon cycle first-order dynamics model, such as a soil organic matter turnover model (Rothamsted Carbon Model, RothC), a process-based land ecosystem biogeochemical cycle model (CENTURY), etc., which belongs to a typical carbon pool turnover process model. Then, the key parameter estimator is constructed, and the machine learning method is used to train and fine-tune the key parameter estimator. Then, the trained key parameter estimator is embedded into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid. Finally, the organic carbon information of the soil to be measured is predicted by the soil organic carbon simulation model to obtain the soil organic carbon turnover of the soil to be measured. In the above process, the key parameter estimator constructed by coupling the physical mechanism model and the machine learning method not only effectively overcomes the problem of large parameter uncertainty in the traditional physical model, but also overcomes the defect of lacking process mechanism explanation in the pure data-driven method, realizes the unification of model precision and interpretability, and solves the problem of poor soil organic carbon estimation effect in the existing related technology.

[0067] Figure 2 is a schematic diagram of the soil organic carbon estimation process in the embodiment of the present application, as shown inFigure 2 In some of the embodiments, after obtaining the site information of the target site, the site information is preprocessed, including: for remote sensing data and reanalysis data in the site information, the site information is processed to form structured table data; and for the organic carbon content change value of the target site starting from the observation year, quality control is performed to eliminate abnormal values.

[0068] Specifically, the site information is processed to form structured table data, including: the site information is reprojected and processed by Kriging function interpolation to extract site information with the same spatial and temporal resolution under a unified spatial coordinate system; and the site information is matched and integrated through the time information and the latitude and longitude coordinates of the target site to obtain the structured table data.

[0069] For example, the soil properties, environmental data, vegetation data, location information, and land use data of the target site are obtained, wherein the soil property data includes the surface soil organic carbon content, soil clay content, sand content, soil pH value, soil nitrogen content, and soil bulk density data of the target site measured in the starting year and the ending year; the environmental data includes the monthly average temperature, monthly precipitation, monthly evapotranspiration, and monthly irrigation of the target site; the vegetation data includes the normalized vegetation index of the target site; the land use data includes the land cover type of the target site; and the location information data includes the latitude and longitude location information of the target site.

[0070] The initial value of the soil organic carbon observation period is used for the initialization of the initial physical model and the input of the key parameter estimator during training, and the end value of the observation period is only used for key parameter inversion calculation and accuracy evaluation, and is not used as an input of the initial physical model to participate in the simulation process; the remaining environmental, soil, and vegetation data are inputs of the initial physical model and the key parameter estimator, which are used for model construction and actual prediction stage.

[0071] The obtained remote sensing data (such as normalized vegetation index, net primary productivity, monthly irrigation, etc.) and reanalysis data (such as monthly average temperature, monthly precipitation, monthly potential evapotranspiration, etc.) need to be reprojected and processed by Kriging function interpolation to extract data with the same spatial and temporal resolution under a unified spatial coordinate system. Through the matching and integration of the time information and the latitude and longitude coordinates of the target site, structured table data is finally formed, which is convenient for subsequent modeling analysis.

[0072] The initial surface soil organic carbon content change value of the target research site is subjected to quality control to eliminate abnormal values. The abnormal value is defined as:

[0073]

[0074] wherein, represents the observation year interval, represents the sample point In the year Soil organic carbon content, Indicates the year and The minimum value of soil organic carbon content. The above expression means that if the annual average change in soil organic carbon content exceeds 25% between any two observation years, the sample point is considered to have a change that exceeds the range that the model can simulate, and may be affected by human interference or measurement errors, and is therefore excluded.

[0075] In some embodiments, step S102 involves calling the initial physical model and, based on the site information of the target site during the observation year, inverting to obtain estimated values ​​of the key parameters of the initial physical model, including: introducing a carbon input correction factor and using the carbon input correction factor as a key parameter of the initial physical model; and inverting the carbon input correction factor using the Markov Chain Monte Carlo (MCMC) method to obtain an estimated value of the carbon input correction factor.

[0076] Specifically, the carbon input correction factor is inverted using the Markov chain Monte Carlo method to obtain an estimated value of the carbon input correction factor. This includes: constructing a Bayesian optimization problem with the sum of squared errors between the output of the initial physical model and the measured soil organic carbon content as the objective function; constructing a posterior probability distribution and defining a likelihood function; and using the Metropolis-Hastington algorithm to execute the sampling process of the Markov chain Monte Carlo method, generating a correction factor sample sequence that follows the posterior probability distribution through multiple iterations, and determining the estimated value of the carbon input correction factor from it.

[0077] For example, the initial physical model uses a monthly time step and considers the input, decomposition, transformation, and loss processes of soil carbon, simulating the dynamic evolution of topsoil organic carbon content over time. The initial physical model typically consists of one or more active carbon pools and one inert carbon pool. Carbon transformation between the active carbon pools occurs through first-order kinetic equations, and the decomposition rate is regulated by environmental factors (such as temperature, soil moisture, and land cover), thus accurately reflecting the interactions between carbon pools and regional environmental differences.

[0078] The conversion process between carbon pools follows a first-order kinetic equation, and the core governing equation can be expressed as:

[0079]

[0080] in, Indicates the activated carbon library in Carbon reserves at any given time. Indicates time The distribution of exogenous carbon inputs (such as crop residues) among carbon pools. Represents the vector of environmental regulatory factors. f T This is a temperature correction factor. f W Soil moisture correction factor, f C For land cover correction factors, This represents the first-order decomposition rate constant matrix for each carbon pool. , This represents the Hadman product, used to multiply the decomposition rate by the environmental factor influence function.

[0081] In this embodiment, the amount of carbon input used to drive the soil organic carbon cycle process in the initial physical model is... Estimated based on net primary production (NPP) of plants. Model carbon inputs in natural ecosystems. The relationship with plant net primary productivity can be expressed as follows:

[0082]

[0083] in, Indicates the amount of carbon input. Indicates net primary productivity. This refers to the carbon used to increase plant biomass. This portion of carbon is used for plant growth and development, including the increase in biomass of the plant's roots and aboveground parts. This refers to carbon emissions resulting from human activities such as harvesting. This indicates the carbon returned through feces and grazing excrement; This represents the portion flowing into the topsoil, characterizing the proportion of net primary productivity flowing into the topsoil.

[0084] In large-scale simulations, it is often difficult to obtain... , , and The precise values ​​of these coefficients are important because they can vary significantly across different regions, climates, and land-use conditions. To improve the accuracy of the estimation, this embodiment introduces a carbon input correction factor. This is used to dynamically adjust the original carbon input estimate. The adjusted carbon input can be expressed as:

[0085]

[0086] in, Indicates the amount of carbon input. This represents the carbon input correction factor. express Net primary productivity of plants at any given time.

[0087] The initial physical model used in this method is a typical first-order reaction kinetic soil carbon cycle model. This type of model is widely used to simulate organic carbon conversion processes. Its core structure is based on the first-order conversion relationship between carbon pools and is driven by environmental factors. However, current models of this type generally face two common problems in practical applications: First, they are sensitive to some input parameters (such as the amount of carbon input, decomposition rate constant, carbon input allocation ratio, etc.), and the parameters have high uncertainty, making it difficult to adapt to the heterogeneous environmental conditions of different regions through uniform settings; second, at the regional scale, key process parameters such as carbon input allocation and carbon conversion rate constant are difficult to observe directly and need to rely on indirect estimation, which is prone to introducing systematic bias and affecting the accuracy and generalization ability of the estimation results.

[0088] In this method, due to the significant uncertainty in estimating plant carbon input at the regional scale, and the high sensitivity of the model's carbon input to soil organic carbon output, a carbon input correction factor α is selected as a key parameter to be evaluated. This factor can be used to correct the systematic bias introduced by the original estimated net primary productivity value, thereby improving the regional adaptability of the carbon input and the accuracy of the model output.

[0089] In other research applications, if the carbon input is relatively reliable but the processes within the carbon pool exhibit significant spatial variability, the decomposition rate constants of each active carbon pool can also be selected as parameters to be evaluated. Parameter selection should be based on a comprehensive assessment of the model structure, the availability of observational data, and the ecological characteristics of the target area, using methods such as sensitivity analysis.

[0090] To determine the carbon input correction factor at the regional or sample level This method uses measured soil organic carbon content from observed years to calculate key parameter values. In this embodiment, the Markov chain Monte Carlo method is used to correct the carbon input factor. The inversion estimation is performed, and the specific calculation process is as follows: Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the inversion and acquisition of key parameter labels in an embodiment of the present invention. This is to invert the key parameters in the initial physical model. A Bayesian optimization problem is constructed with the sum of squared errors between the initial physical model output and the measured soil organic carbon as the objective function, and its form is as follows:

[0091]

[0092] in, Describe the objective function. Indicates the initial physical model under given conditions. The value obtained from the simulation is the first Simulated soil organic carbon values ​​for each sample point in the observed year. This indicates the measured value of soil organic carbon at that sample point in the year of observation.

[0093] Markov chain Monte Carlo sampling is achieved by constructing a posterior probability distribution:

[0094]

[0095] in, Let be the likelihood function, representing the correction factor given the key parameters. Under these conditions, the probability of observed SOC occurring is determined by the error between the observed value and the model simulation value, and it is assumed that the error follows a normal distribution. Let α represent the prior distribution, which can be set as a uniform distribution α ~ U(a,b), defined as [-10, 10]. Then, the Metropolis-Hastings algorithm is used to perform the Markov chain Monte Carlo sampling process. Through multiple iterations, a posterior distribution is generated. The sample sequences are used to extract the posterior mean or median as an estimate of the carbon input correction factor, which is then used to construct the training dataset as labels.

[0096] In some embodiments, step S103, constructing a key parameter estimator and training and fine-tuning the key parameter estimator, includes: constructing the key parameter estimator using a random forest algorithm; generating an input feature vector based on the site information of the target site; using the input feature vector as input and the carbon input correction factor as a label to train and fine-tune the key parameter estimator; the key parameters are used to perform dynamic parameter prediction based on the input.

[0097] Specifically, the key parameter estimator is trained and fine-tuned, including: inputting the input feature vector into the key parameter estimator to obtain the mean of the predicted values ​​corresponding to each decision tree; and optimizing the hyperparameters of the parameter estimator based on the mean of the predicted values ​​corresponding to each decision tree and the estimated values ​​of the key parameters, with minimizing the mean squared error as the objective function.

[0098] For example, Figure 4 This is a schematic diagram illustrating the construction of the key parameter estimator in an embodiment of the present invention, as shown below. Figure 4 As shown, the key parameter correction factor The estimated values of the key parameters are used as label data for the training of the key parameter estimator. The input data are the soil properties (initial soil organic carbon content, clay content, sand content, pH value, nitrogen content, bulk density, etc.), land cover type, environmental data (average air temperature, average precipitation, average potential evapotranspiration, and latitude and longitude position information obtained in the above steps, which constitute the input feature vector. Before inputting the key parameter estimator, all input feature vectors are standardized to ensure consistency of data with different dimensions. For discrete classification variables such as land use types (e.g., forest land, farmland, grassland, etc.), One-Hot Encoding is used for numerical processing. The conversion process can be represented by the following formula:

[0099]

[0100] wherein, represents the One-Hot Encoding function, represents the value of the classification variable of the th sample, represents the possible values of all categories of the classification variable, represents the indicator function, which takes the value of 1 when the condition in the parentheses is true, and 0 otherwise.

[0101] The dataset is divided into training set and validation set in the ratio of 7:3, and the random forest algorithm is used to construct the key parameter estimator. The key parameter estimator realizes dynamic parameter prediction through the following formula:

[0102]

[0103] wherein, represents the dynamic parameter to be predicted; represents the mapping function learned by the random forest, which is used to give the predicted value according to the input feature vector . The input feature vector includes soil properties, land cover, environmental data, and latitude position information, represents the prediction function of the th tree. The output of the key parameter estimator constructed by the random forest algorithm is the mean value of the prediction values of each decision tree, i.e. , represents the number of decision trees in the random forest. The physical constraint loss function used in its training process is as follows:

[0104]

[0105] wherein, in the loss function represents the The true carbon input correction factor of a sample (obtained through MCMC inversion), represents a predicted value, represents the number of training samples, is a regularization coefficient, represents a regularization term used to control the complexity of the model and prevent overfitting.

[0106] In the training process of the key parameter estimator, a Bayesian optimization method is used to optimize the hyperparameters of the key parameter estimator, with the mean square error (MSE) as the objective function, so as to improve the prediction performance and generalization ability of the model, and the expression is as follows:

[0107]

[0108] wherein, represents the optimal hyperparameters, represents a set of hyperparameters, including the number of trees, the maximum depth, the minimum number of leaf nodes, etc., represents the mean square error of the key parameter estimator obtained by training on the validation set with the hyperparameters

[0109] In some embodiments, step S104, embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid, includes: based on the site information of the target site, running the initial physical model, and calling the trained key parameter estimator in real time, dynamically generating a carbon input correction factor with spatial difference; combining the carbon input correction factor and the primary productivity of the vegetation of the target site to determine the carbon input amount of the target site; introducing a bias monitoring mechanism to update and correct the key parameter estimator, and obtaining a soil organic carbon simulation model driven by physical and data hybrid.

[0110] For example, during the running of the initial physical model, the trained key parameter estimator is called in real time, and a correction factor with spatial difference is dynamically generated according to the input environmental driving data (such as climate, soil properties, vegetation index, etc.) , and the carbon input amount is calculated in combination with the net primary productivity of vegetation, and the specific formula is as follows:

[0111]

[0112] wherein, represents the carbon input vector of each carbon pool, represents the spatial position, represents the time, represents the net primary productivity, ​This represents the proportion vector of carbon inputs allocated to each carbon pool, which is determined based on different land use types.

[0113] To further improve the model's adaptability across different regions and time scales, this embodiment introduces a deviation monitoring mechanism during the soil organic carbon simulation process. This mechanism continuously monitors the residuals between the simulated and measured values. If the residuals systematically exceed a preset threshold in a specific region or time period, the error characteristics are recorded for subsequent updates or corrections to the key parameter estimator.

[0114] The constructed soil organic carbon simulation model is run, and the model output includes the storage capacity and turnover rate of each soil carbon pool. During the operation, the total soil organic carbon storage is... The dynamic changes can be represented by the following formula:

[0115]

[0116] in, This indicates that the site is in Total soil organic carbon reserves at the current time. , , and The activated carbon libraries at this site represent different decomposition rates. Carbon content at a given time This indicates the amount of inert organic carbon that does not change over time.

[0117] Based on the above embodiments, a hybrid physical and data-driven soil organic carbon simulation model was constructed, and simulation experiments were conducted in typical ecosystems such as grassland, farmland, and forest at a certain regional site. The estimation results of the hybrid model were compared with those of the traditional physical model, and the main evaluation index included the coefficient of determination R. 2 The root mean square error (RMSE) and mean absolute error (MAE) are expressed as follows:

[0118]

[0119]

[0120]

[0121] in, The coefficient of determination is represented by the coefficient of determination. Indicates the first i Measured soil organic carbon values ​​for each sample. Indicates the first i Simulated soil organic carbon values ​​for each sample. denotes the total number of samples, denotes the root mean square error, denotes the mean absolute error.

[0122] The measured value of soil organic carbon in 2009 is used as the initial input to simulate the soil organic carbon in 2015 and compared with the measured value; and the measured value in 2015 is used as the initial input to simulate the soil organic carbon in 2018 and verified. The model accuracy evaluation results of the above two verification years are shown in Figures 5-8 Figure 5 is the simulation scatter plot of the traditional physical model in the embodiment of the application from 2009 to 2015, Figure 6 is the simulation scatter plot of the mixed driving model in the embodiment of the application from 2009 to 2015; Figure 7 is the simulation scatter plot of the traditional physical model in the embodiment of the application from 2015 to 2018, Figure 8 is the simulation scatter plot of the mixed driving model in the embodiment of the application from 2015 to 2018.

[0123] From the above comparative simulation experiment, the application proposes a mixed driving modeling method fusing a physical process model and a machine learning estimator to solve the problems of large parameter uncertainty and weak spatial generalization ability in the traditional soil organic carbon model. The method constructs a nonlinear estimation model of carbon input correction factor , and couples it with the simulation framework of the soil carbon cycle physical model to realize intelligent regulation and control of carbon input according to local conditions and timely response, significantly improving the estimation accuracy and regional adaptability of soil organic carbon. Secondly, the application makes full use of multi-source environmental data such as soil properties, environmental factors, vegetation index, land cover type and geographic location to construct a mixed driving model with physical constraints and interpretability, enhancing the rationality and scientificity of the model structure. At the same time, the application constructs an integrated modeling process from parameter inversion, model training to soil organic carbon estimation, with the advantages of high efficiency, convenience and strong scalability, which can efficiently realize the simulation of dynamic changes of soil organic carbon in large-scale and multi-ecosystem. In the verification of typical ecosystems, the model of the application shows higher determination coefficient R 2 , lower root mean square error RMSE and lower mean absolute error MAE compared with the traditional pure physical model, showing good versatility and precision advantage. The method can be widely applied to carbon sink monitoring, carbon neutralization path design, soil health assessment and other fields, and has important practical value and broad application prospect.

[0124] ​In summary, the method effectively solves the problem of large parameter uncertainty in traditional physical models by coupling a parameter estimator constructed by a physical mechanism model and a machine learning method, overcomes the lack of process mechanism explanation in pure data-driven methods, and realizes the unification of model precision and interpretability. Moreover, the method is simple to operate, low in data acquisition cost, and only relies on conventional environmental data for model running, without the need for complex parameter tuning process, making it suitable for scenarios lacking continuous high-quality observation data, and having good practicality and promotional value. In addition, the method has good generalization ability and applicability, and can be compatible with different land use types and different climate conditions for soil organic carbon simulation tasks, and is suitable for large-scale regional simulation and prediction. Furthermore, the method improves the calculation efficiency of the model, and does not need to rely on the initialization step or empirical formula estimation of external carbon input in traditional physical models, making it suitable for large-scale soil organic carbon dynamic estimation and scenario prediction.

[0125] The application further provides a soil organic carbon estimation device, which is described as follows. Figure 9 is a structural block diagram of the soil organic carbon estimation device provided by the application, as shown in Figure 9 The device comprises:

[0126] The acquisition module 901 is configured to acquire site information of a target site, wherein the site information comprises soil properties, environmental data, vegetation data, location information and land use data of the target site.

[0127] The processing module 902 is configured to call an initial physical model and inversely derive an estimated value of a key parameter of the initial physical model based on site information of the target site in an observation year.

[0128] The training module 903 is configured to construct a key parameter estimator and fine-tune the key parameter estimator.

[0129] The embedding module 904 is configured to embed the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data mixing.

[0130] The estimation module 905 is configured to predict the organic carbon information of the to-be-tested soil by using the soil organic carbon simulation model to obtain the soil organic carbon turnover of the to-be-tested soil.

[0131] In use, first, the acquisition module 901 acquires site information of the target site, which can effectively represent the soil conditions and other environmental conditions of the target site. Then, the processing module 902 calls an initial physical model, and determines the key parameters of the initial physical model according to the site information of the target site in the observation years, and inverses the estimated value of the key parameters. Exemplarily, the physical model adopted is a first-order dynamics model of soil carbon cycle, such as a soil organic matter turnover model (Rothamsted Carbon Model, RothC), a process-based land ecosystem biogeochemical cycle model (CENTURY), etc., which belongs to a typical carbon pool turnover process model. The training module 903 then constructs a key parameter estimator and trains and fine-tunes the key parameter estimator by using a machine learning method. Then, the embedding module 904 embeds the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid. Finally, the estimation module 905 predicts the organic carbon information of the to-be-measured soil by using the soil organic carbon simulation model to obtain the soil organic carbon turnover amount of the to-be-measured soil. In the above process, the key parameter estimator constructed by coupling the physical mechanism model and the machine learning method not only effectively overcomes the problem of large parameter uncertainty in the traditional physical model, but also overcomes the defect of lacking process mechanism explanation in the pure data-driven method, realizes the unification of model precision and interpretability, and solves the problem of poor soil organic carbon estimation effect in the existing related technologies.

[0132] Figure 10 An example of a schematic diagram of the physical structure of an electronic device is shown in Figure 10 The electronic device can include a processor 1001, a communications interface 1002, a memory 1003, and a communications bus 1004, wherein the processor 1001, the communications interface 1002, and the memory 1003 communicate with each other through the communications bus 1004. The processor 1001 can invoke a logical instruction in the memory 1003 to execute a soil organic carbon estimation method, which includes:

[0133] Acquiring site information of a target site; the site information includes soil properties, environmental data, vegetation data, location information, and land use data of the target site;

[0134] Calling an initial physical model, and inversing an estimated value of a key parameter of the initial physical model based on site information of the target site in observation years;

[0135] Constructing a key parameter estimator and training and fine-tuning the key parameter estimator;

[0136] Embed the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid.

[0137] The soil organic carbon simulation model is used to predict the organic carbon information of the to-be-tested soil to obtain the soil organic carbon turnover of the to-be-tested soil.

[0138] In addition, the logical instructions in the memory 1003 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0139] On the other hand, the present application also provides a computer program product, the computer program product comprising a computer program, the computer program being storable on a non-transitory computer readable storage medium, and the computer program being executable by a processor, the computer being capable of executing the soil organic carbon estimation method provided by the above-mentioned methods, the method comprising:

[0140] Obtaining site information of the target site; the site information includes soil properties, environmental data, vegetation data, location information and land use data of the target site;

[0141] Calling an initial physical model, and based on the site information of the target site in the observation year, obtaining the estimated value of the key parameters of the initial physical model by inversion;

[0142] Constructing a key parameter estimator and training and fine-tuning the key parameter estimator;

[0143] Embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid;

[0144] The soil organic carbon simulation model is used to predict the organic carbon information of the to-be-tested soil to obtain the soil organic carbon turnover of the to-be-tested soil.

[0145] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the soil organic carbon estimation method provided by each of the above methods, the method comprising:

[0146] obtaining site information of the target site; the site information comprising soil properties, environmental data, vegetation data, location information and land use data of the target site;

[0147] calling an initial physical model, and inversely deriving an estimated value of a key parameter of the initial physical model based on site information of the target site in the observation years;

[0148] constructing a key parameter estimator and training and fine-tuning the key parameter estimator;

[0149] embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid;

[0150] predicting the organic carbon information of the soil to be measured through the soil organic carbon simulation model to obtain the soil organic carbon turnover of the soil to be measured.

[0151] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0152] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in terms of the contribution to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0153] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for estimating soil organic carbon, characterized by, The method comprises the following steps: obtaining site information of a target site; the site information comprises soil properties, environmental data, vegetation data, location information and land use data of the target site; calling an initial physical model, and inversely deriving an estimated value of a key parameter of the initial physical model based on site information of the target site in observation years; constructing a key parameter estimator and fine-tuning the key parameter estimator; embedding the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by physical and data hybrid; predicting organic carbon information of a to-be-tested soil through the soil organic carbon simulation model to obtain a soil organic carbon turnover of the to-be-tested soil; calling an initial physical model, and inversely deriving an estimated value of a key parameter of the initial physical model based on site information of the target site in observation years, comprising: introducing a carbon input correction factor, and taking the carbon input correction factor as the key parameter of the initial physical model; inversely deriving the carbon input correction factor through a Markov chain Monte Carlo method to obtain an estimated value of the carbon input correction factor; constructing a key parameter estimator and fine-tuning the key parameter estimator, comprising: constructing a key parameter estimator by using a random forest algorithm; generating an input feature vector based on the site information of the target site; fine-tuning the key parameter estimator by taking the input feature vector as input and taking the carbon input correction factor as label; the key parameter is used for dynamic parameter prediction according to input; fine-tuning the key parameter estimator, comprising: inputting the input feature vector into the key parameter estimator to obtain a mean value of prediction values corresponding to each decision tree; optimizing hyperparameters of the parameter estimator based on the mean value of prediction values corresponding to each decision tree and the estimated value of the key parameter, taking minimizing mean square error as an objective function.

2. The method of estimating soil organic carbon according to claim 1, characterized by, After obtaining the site information of the target site, pre-processing the site information, comprising: processing the site information to form structured table data, for remote sensing data and reanalysis data in the site information; performing quality control on the organic carbon content change value of the target site in observation years to eliminate abnormal values.

3. The method of estimating soil organic carbon according to claim 2, characterized by, processing the site information to form structured table data, comprising: performing re-projection and Kriging function interpolation processing on the site information to extract site information with the same spatio-temporal resolution under a unified spatial coordinate system; matching and integrating the site information through time information and latitude and longitude coordinates of the target site to obtain structured table data.

4. The method of estimating soil organic carbon according to claim 1, characterized by, inversely deriving the carbon input correction factor through a Markov chain Monte Carlo method to obtain an estimated value of the carbon input correction factor, comprising: constructing a Bayesian optimization problem with the sum of squared errors between the output of the initial physical model and the measured soil organic carbon content as an objective function; constructing a posterior probability distribution to set a likelihood function; The sampling process of the Markov Chain Monte Carlo method is performed by using the Metropolis-Hastings algorithm to generate a sequence of samples of the correction factor subject to the posterior probability distribution through multiple iterations, and the estimated value of the carbon input correction factor is determined therefrom.

5. The method of estimating soil organic carbon according to claim 1, wherein, The trained key parameter estimator is embedded into the initial physical model to obtain a soil organic carbon simulation model driven by hybrid physical and data, including: Based on the site information of the target site, the initial physical model is run, and the trained key parameter estimator is called in real time to dynamically generate carbon input correction factors with spatial differences. The carbon input amount of the target site is determined by combining the carbon input correction factor and the primary productivity of vegetation of the target site. A bias monitoring mechanism is introduced to update and correct the key parameter estimator to obtain a soil organic carbon simulation model driven by hybrid physical and data.

6. A soil organic carbon estimation device for implementing the soil organic carbon estimation method according to any one of claims 1 to 5, characterized by, It includes: An acquisition module is configured to acquire site information of a target site. The site information includes soil properties, environmental data, vegetation data, location information, and land use data of the target site. A processing module is configured to call an initial physical model and obtain estimated values of key parameters of the initial physical model based on site information of the target site in the observation period. A training module is configured to construct a key parameter estimator and train and fine-tune the key parameter estimator. An embedding module is configured to embed the trained key parameter estimator into the initial physical model to obtain a soil organic carbon simulation model driven by hybrid physical and data. An estimation module is configured to predict the organic carbon information of a to-be-tested soil by using the soil organic carbon simulation model to obtain the soil organic carbon turnover of the to-be-tested soil.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the soil organic carbon estimation method of any one of claims 1-5.

Images