Land ecosystem carbon source sink assimilation prediction method, device and equipment based on adaptive moment estimation and medium

Abstract

The invention discloses a land ecosystem carbon source sink assimilation prediction method and device based on adaptive moment estimation, equipment and a medium. The method comprises the following steps: collecting a meteorological data set; the meteorological data set is input into the optimized carbon cycle model to predict the carbon flux, the carbon cycle model comprises a photosynthesis model and a respiratory action model, the photosynthesis model is used for obtaining photosynthesis flux, and the respiratory action model is used for obtaining respiratory action flux; wherein the optimization process of the carbon cycle model comprises the following steps: carrying out data assimilation on model parameters of the carbon cycle model by adopting an adaptive moment estimation algorithm, and constraining carbon flux to obtain an initially optimized carbon cycle model; fitting photosynthesis flux and respiratory flux based on the carbon cycle model of the initial optimization model parameters, and automatically constraining flux sum with poor fitting effect according to the fitting effect to obtain a final optimized carbon cycle model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon cycle, in particular to a land ecosystem carbon source and sink assimilation prediction method and device based on adaptive matrix estimation, equipment and medium. BACKGROUND

[0002] Global climate change and rapid changes in the ecological environment have had a profound impact on the Earth's ecosystem. The continuous increase in greenhouse gas emissions has led to a continuous rise in global temperature, an increase in extreme weather events, a rise in sea level, and a sharp decline in biodiversity. These problems not only threaten the sustainable development of human society, but also seriously disrupt the natural balance of the Earth. Carbon cycle is a key process in the Earth's ecosystem, involving atmospheric carbon dioxide exchange, plant photosynthesis, soil and ocean carbon storage, and fossil fuel combustion. The balance of carbon cycle is crucial for maintaining global climate stability and biodiversity. However, human activities have disrupted the natural balance of carbon cycle, leading to a continuous rise in atmospheric carbon dioxide concentration.

[0003] Therefore, how to conveniently and effectively predict the ecosystem (especially the carbon cycle) has become a technical problem to be solved. At present, there are many models that model net ecosystem exchange (NEE) through meteorological data. The key goal of modeling is to optimize the parameters of the model through existing meteorological data and flux data, so as to predict future carbon cycle. Traditional data assimilation algorithms mainly use Monte Carlo simulation and traditional Newton method based on Hessian matrix for optimization. For example, in the process of data assimilation, a quasi-Newton method (BFGS) is used. This method obtains gradient information by numerical approximation, but as the model becomes more complex, this optimization method is prone to local optimal solution and is sensitive to initial values. The optimization in high parameter space is prone to local minimum rather than global minimum. The above optimization methods still have deficiencies in prediction accuracy, and the stability of some assimilation algorithms needs to be improved. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, in view of the above problems, the purpose of the present application is to provide a land ecosystem carbon source and sink assimilation prediction method and device based on adaptive matrix estimation, equipment and medium, which can better predict future carbon flux and effectively improve the accuracy of forest ecosystem carbon flux estimation by modeling the carbon cycle of the land ecosystem and obtaining its optimal model parameters.

[0005] In order to achieve the above application purpose, the technical scheme adopted by the present application is: In a first aspect, the present application provides a land ecosystem carbon source and sink assimilation prediction method based on adaptive matrix estimation, comprising: collecting a meteorological dataset; inputting the meteorological dataset into an optimized carbon cycle model to predict carbon fluxes, the carbon cycle model comprising a photosynthesis model and a respiration model, the photosynthesis model being used to obtain a photosynthesis flux , the respiration model being used to obtain a respiration flux ; wherein the optimization process of the carbon cycle model is: using an adaptive moment estimation algorithm to data assimilate model parameters of the carbon cycle model and to constrain carbon fluxes to obtain an initially optimized carbon cycle model; fitting the photosynthesis flux and the respiration flux based on the initially optimized model parameters of the carbon cycle model, and automatically constraining fluxes and with poor fitting effects according to the fitting effects, to obtain a finally optimized carbon cycle model.

[0006] In some possible implementations, the photosynthesis model is expressed using a Michaelis-Menten model:

[0007] wherein the photosynthesis flux is an ecosystem primary productivity; A max is a maximum photosynthetic rate; I is an irradiance; is a canopy light use efficiency.

[0008] In some possible implementations, the respiration model is expressed using a Lloyd & Tayor model or a Q10-SW model:

[0009] wherein the respiration flux RE is an ecosystem respiration; B R is an ecosystem basal respiration at a reference temperature; Q 10 is a temperature response factor of the ecosystem respiration, T is a soil temperature, T ref is the reference temperature.

[0010] In some possible implementations, the carbon fluxes are determined using the following formula:

[0011] In some possible implementations, an adaptive moment estimation algorithm is used to data assimilate model parameters of the carbon cycle model and to constrain carbon fluxes to obtain an initially optimized carbon cycle model, and the process is as follows: For each carbon cycle model, set initial values, value ranges, time steps, maximum number of iterations, and convergence thresholds for the model parameters. The model parameters for the carbon cycle model include photosynthetic parameters and respiration parameters. Meteorological data and flux data are acquired. Meteorological data is used as driving data for forward simulation of the carbon cycle model, while flux data is used as constraint data to evaluate the simulation error of the carbon cycle model and construct the loss function. In each parameter candidate value P (k) The carbon cycle model is run at time steps to obtain the model-predicted flux sequence at each time step. ( P (k)), where, j The different constraint fluxes are NEE, GEP, and RE, respectively. The parameters are updated using the following loss function:

[0012] in, L Represents the loss function. m The number of flux data representing the constraints. n Characterization in the j The data types of data points in a group of observations Indicates the first j The standard deviation of each data point in the group of observations x Indicates the measured value. P Indicates the parameters to be trained; For each respiratory and photosynthetic parameter, the carbon cycle model is run with initial values ​​to obtain a set of simulation results. The loss function is then calculated according to the above method. After obtaining the loss value, it is updated using an adaptive moment estimation algorithm. The process is repeated until the loss value remains unchanged each time, thus obtaining the optimal set of parameters.

[0013] In some possible implementations, meteorological data include temperature, photosynthetically active radiation, relative humidity, and soil moisture content; flux data include primary productivity, net ecosystem exchange, and ecosystem respiration.

[0014] In some possible implementations, the carbon cycle model based on the initial optimized model parameters affects the photosynthetic flux. and respiratory flux Perform a fit and automatically adjust the flux for poor fits based on the fit results. and By applying constraints, the final optimized carbon cycle model is obtained. The process is as follows: When the loss function is constructed using NEE observation data as constraint variables, and the parameters are updated using the adaptive moment estimation algorithm, a preliminary optimized parameter set is obtained. After the preliminary optimization of NEE is completed, the photosynthesis flux GEP and the respiration flux RE in the carbon cycle model are fitted based on the initial optimization result, and in the fitting process, the carbon cycle model compares the fitting degree R between the simulation value GEP, RE calculated by the model and the GEP, RE data obtained by observation 2 , automatically identifies the carbon flux component with lower fitting accuracy, dynamically adjusts the assimilation strategy based on the error feedback mechanism, and integrates the flux with poor fitting effect into the optimization objective function, and forms a joint loss function with NEE, automatically monitors the residual change of each flux during the optimization process, and dynamically adjusts the weight of each flux according to the error feedback, and realizes adaptive constraint fusion.

[0015] In a second aspect, the present application also provides a land ecosystem carbon source and sink assimilation prediction device based on adaptive matrix estimation, comprising: A data collection unit configured to collect a meteorological data set; A carbon flux prediction unit configured to input the meteorological data set into the optimized carbon cycle model to predict the carbon flux, the carbon cycle model comprising a photosynthesis model and a respiration model, the photosynthesis model being used to obtain the photosynthesis flux , and the respiration model being used to obtain the respiration flux ; wherein the optimization process of the carbon cycle model is as follows: An adaptive matrix estimation algorithm is used to assimilate the model parameters of the carbon cycle model, and the carbon flux is constrained to obtain an initially optimized carbon cycle model; The carbon cycle model based on the initially optimized model parameters is used to fit the photosynthesis flux And the respiration flux , and the fluxes with poor fitting effect are automatically constrained according to the fitting effect And , to obtain the finally optimized carbon cycle model.

[0016] In a third aspect, the present application also provides an electronic device, comprising: at least one processor; and a memory connected with the processor in communication; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to execute the method.

[0017] In a fourth aspect, the present application also provides a computer readable storage medium storing one or more programs, the one or more programs comprising computer instructions for causing a computer to execute the method.

[0018] The present application has the following characteristics due to the above technical solutions: 1. The assimilation technique used in this invention is the Adam algorithm, which was first applied to ecosystem modeling. The assimilation model is implemented using R... 2 Compared with the ecological model MCMC algorithm and L-BFGS-L algorithm, it has better simulation effect; the assimilation technology used in this invention is not sensitive to the initial value of the model parameters. Compared with the L-BFGS-L algorithm, the optimization results of this model converge to the same set of values ​​for different initial values, rather than changing with the change of the initial value.

[0019] 2. When only NEE optimization is used, NEE has a good fitting effect, but the fitting effect for GEP and RE is uncertain. This invention automatically identifies the fitting effect of the three fluxes and aligns the fluxes with poor fitting effects, adjusting the fitting effect R of the poorly fitting fluxes. 2 Specifically, the improvement was achieved by using a single flux fitting residual to evaluate the dynamic addition of new constraints to enhance the overall model performance.

[0020] 3. The data assimilation method for estimating carbon flux in forest ecosystems proposed in this invention improves the fitting effect on carbon flux observation data while taking into account the complexity of the model. This method can effectively reduce the sensitivity of traditional models to initial parameters and achieve robust prediction of carbon source and sink status.

[0021] In summary, this invention can be widely applied to the prediction of carbon source and sink states. Attached Figure Description

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting carbon source assimilation in terrestrial ecosystems according to an embodiment of the present invention.

[0023] Figure 2 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0025] Although terms such as first, second, third, etc., may be used in this document to describe multiple elements, components, regions, layers, and / or segments, these elements, components, regions, layers, and / or segments should not be limited by these terms. These terms may be used only to distinguish one element, component, region, layer, or segment from another. Unless the context clearly indicates otherwise, terms such as "first," "second," and other numerical terms used herein do not imply order or sequence. Therefore, the first element, component, region, layer, or segment discussed below may be referred to as the second element, component, region, layer, or segment without departing from the teachings of the exemplary embodiments.

[0026] For ease of description, spatial relative terms may be used in the text to describe the relationship of one element or feature relative to another element or feature as shown in the figure. These relative terms include, for example, "inside," "outside," "middle," "outer," "below," "above," etc. Such spatial relative terms are intended to include different orientations of the device in use or operation, other than those depicted in the figure.

[0027] Existing optimization methods still have shortcomings in prediction accuracy, and the stability of some assimilation algorithms needs improvement. This invention provides a method, apparatus, equipment, and medium for predicting carbon source-sink assimilation in terrestrial ecosystems based on adaptive moment estimation. The method includes: collecting meteorological datasets; inputting the meteorological datasets into an optimized carbon cycle model to predict carbon flux; the carbon cycle model includes a photosynthesis model and a respiration model, with the photosynthesis model used to obtain photosynthetic flux. Respiratory models are used to obtain respiratory flux. The optimization process of the carbon cycle model is as follows: The model parameters of the carbon cycle model are assimilated using an adaptive moment estimation algorithm, and the carbon flux is constrained to obtain an initial optimized carbon cycle model; based on the initial optimized model parameters, the photosynthetic flux is then evaluated using the carbon cycle model. and respiratory flux Perform a fit and automatically adjust the flux for poor fits based on the fit results. and By applying constraints, the final optimized carbon cycle model is obtained. Therefore, this invention improves the fitting effect on carbon flux observation data while taking into account model complexity, effectively reduces the sensitivity of traditional models to initial parameters, and achieves robust prediction of carbon source and sink states.

[0028] Exemplary embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey its scope to those skilled in the art. To make the invention clearer, the technical terms appearing in the invention are explained.

[0029] Carbon sources: Carbon sources refer to processes, activities, or systems that release carbon dioxide (CO2) or other greenhouse gases into the atmosphere. Natural carbon sources include soil respiration, plant and animal decomposition, forest fires, and volcanic eruptions. Anthropogenic carbon sources mainly include the combustion of fossil fuels, industrial production, transportation, and land-use changes (such as deforestation). Carbon sources increase atmospheric CO2 concentration by releasing greenhouse gases, making them a significant driver of global warming.

[0030] Carbon sinks: Carbon sinks refer to the processes or systems that absorb carbon dioxide from the atmosphere and fix it in media such as plants, soil, and oceans. Typical carbon sinks include ecosystems such as forests, grasslands, wetlands, and marine phytoplankton, which convert CO2 into organic carbon through photosynthesis and store it, thereby reducing the carbon concentration in the atmosphere over a certain period. Effective carbon sinks are a key means to mitigate climate change and achieve carbon neutrality goals.

[0031] Identifying carbon sources and sinks is a key issue in climate change research. Carbon sources and sinks are represented by the positive or negative value of carbon flux; a positive value indicates a carbon source, and a negative value indicates a carbon sink. To more accurately estimate and optimize net ecosystem exchange carbon flux models, traditional empirical or single-algorithm-based models may face limitations, while combining advanced data assimilation and parameter optimization methods can significantly improve model performance. Addressing common problems in existing data assimilation processes, such as the inability to find the optimal solution and the sensitivity of algorithms to initial values, this invention assimilates model parameters based on existing models and data.

[0032] Example 1: As Figure 1 As shown in the figure, the specific implementation process of the terrestrial ecosystem carbon source-sink assimilation prediction method based on adaptive moment estimation provided in this embodiment is as follows: S1. Collect meteorological datasets and flux datasets.

[0033] In this embodiment, meteorological and carbon data of the target area are collected to determine the dataset. Meteorological data, as fundamental information describing regional climate characteristics, is of great significance in improving the accuracy of local weather forecasts, enhancing disaster prevention and mitigation capabilities, and deepening our understanding and response to climate change. Long-term, continuous meteorological observations can reveal important driving factors reflecting ecosystem processes.

[0034] Furthermore, meteorological data includes key variables such as temperature, photosynthetically active radiation (PAR), relative humidity, and soil moisture content. Surface ecosystem flux data includes gross primary productivity, net ecosystem exchange, and ecosystem respiration. Meteorological data can be referred to as driving data, and flux data as constraining data. Driving data can be acquired, for example, every 30 minutes, and constraining data can also be acquired, for example, every 30 minutes; this is just one example, and not a limitation.

[0035] Furthermore, various sensors can be used to collect driving data, such as temperature sensors and humidity sensors, which will not be elaborated here. Since sensor malfunctions may occur during data collection, resulting in abnormal data and invalid data, it is necessary to process the acquired data after it is obtained. Data processing includes cleaning up abnormal data, identifying and correcting or deleting errors, duplicates, or incomplete records, etc. Data processing enables data quality control. Specifically, the data processing method in this embodiment is to delete missing or incomplete records, and so on, but is not limited to this.

[0036] S2, Training the carbon cycle model.

[0037] In this embodiment, a carbon cycle model is trained based on data from the dataset. The carbon cycle model uses meteorological data to predict fluxes, and then uses flux data to constrain the carbon cycle model parameters. The carbon cycle model describes the respiration and photosynthesis of a forest ecosystem with a relatively simple structure. This relatively simple structure avoids overfitting during model data assimilation and allows for efficient use of computational resources. It has been widely used in carbon cycle simulations in various regions around the world.

[0038] Furthermore, the photosynthesis model in the carbon cycle model can be selected from the Michaelis-Menten series, where the photosynthetic intensity of the ecosystem (usually measured by GEP) is related to photosynthetically active radiation and conforms to a rectangular hyperbolic equation. The photosynthesis model is expressed using the Michaelis-Menten model.

[0039] In the formula, GEP represents primary productivity of the ecosystem; Amax is the maximum photosynthetic rate, which refers to the CO2 assimilation rate of an ecosystem under light saturation. It represents the potential of the ecosystem's photosynthetic capacity and is an important functional characteristic of the ecosystem, determining the ecosystem's carbon assimilation. It is equivalent to the upper limit of GEP and defines the maximum carbon assimilation efficiency. I represents radiance, which is the light efficiency available to plants. It refers to the canopy light utilization efficiency, which controls the efficiency of photosynthetic carbon fixation.

[0040] Considering that plant photosynthesis is affected by stomatal opening and closing, VPD influences photosynthetic efficiency by affecting stomatal opening and closing. When VPD is too high, plants will close their stomata to avoid excessive water evaporation, leading to a decrease in photosynthetic efficiency. Therefore, Amax can be adjusted as follows:

[0041] in, is the maximum photosynthetic rate under no vapor pressure stress, k is the maximum coefficient of ecosystem respiration to VPD, VPD is the air saturation vapor pressure at canopy height, and VPD0 represents the minimum vapor pressure at which plants are subjected to vapor pressure stress, set to 10 hPa.

[0042] Furthermore, respiration models can be selected from Lloyd & Taylor models, Q10-SW models, etc. Among them, the Lloyd & Taylor model better describes the variability of ecosystem respiration and has better performance in describing plant responses under low and high temperatures.

[0043] Among them, RE stands for ecosystem respiration, and B... R It is the basic respiration of the ecosystem at the reference temperature; Q 10 It is the response factor of ecosystem respiration to temperature, representing the increase in ecosystem respiration for every 10 degrees Celsius increase in temperature. T is the soil temperature at 5 cm depth. ref This is the reference temperature, taken as T. ref It is 10.

[0044] Considering the close relationship between ecosystem respiration and soil moisture, under drought conditions, plant dehydration and slow metabolism hinder the diffusion of organic carbon at the bottom. With adequate moisture, enzymatic reactions proceed smoothly, and gas exchange is unimpeded. Under excessive moisture conditions, pores are filled, limiting oxygen diffusion. The Q10-SW model performs better in terms of temperature and moisture under arid soil conditions.

[0045] In the formula, a and b are the function coefficients corresponding to the model, and SW This refers to the soil moisture content factor.

[0046] Furthermore, once the equations for photosynthesis and respiration are determined, carbon flux NEE (a key indicator measuring whether an ecosystem is "net absorber" or "net emitter" over a specific period) can be determined by the following formula:

[0047] S3, Model parameter data assimilation.

[0048] In this embodiment, Adaptive Moment Estimation (AMO) is a gradient-based first-order optimization method widely used in deep learning, nonlinear system modeling, and parameter optimization problems of high-dimensional complex models. This algorithm dynamically adjusts the learning rate of each parameter by simultaneously considering the first-order momentum (mean) and second-order momentum (variance estimation) of the gradient, thereby achieving an adaptive and efficient optimization process. Compared with traditional gradient descent, Adam has significant advantages: Adaptive Learning Rate: Adam assigns an independent learning rate to each parameter, and the update of the learning rate is influenced by both the historical gradient mean and variance of that parameter. This mechanism makes the parameter changes more smoothly across different dimensions, avoiding the oscillations or slow convergence problems caused by uniformly setting the learning rate. Due to the introduction of momentum and variance information, Adam has a faster convergence speed than ordinary gradient descent in most practical problems, and is particularly suitable for optimization tasks with nonlinear objective functions and complex surfaces. Strong Robustness and Insensitivity to Initial Values: Even when faced with poorly selected initial parameters, the Adam algorithm can still gradually approach the optimal solution, exhibiting good stability and fault tolerance. This makes it particularly suitable for scenarios with multiple local minima, such as ecosystem modeling.

[0049] In this embodiment, an adaptive moment estimation algorithm is used to assimilate the model parameter data (including but not limited to key parameters such as maximum photosynthetic rate Amax, canopy light utilization efficiency LUE, basal respiratory rate BR, and respiratory temperature response coefficient Q10) of the trained carbon cycle model. In optimizing the trained carbon cycle model, meteorological data is used as the driving force. Carbon flux is measured every 30 minutes, and the predicted carbon flux is compared with the actual carbon flux value using a loss function to achieve data assimilation. The specific process is as follows: The carbon cycle model consists of a respiration model and a photosynthesis model, including the Michaelis-Menten series of photosynthesis models and the Lloyd & Taylor, Q10-SW respiration model. Parameters to be assimilated include: Photosynthetic parameters: maximum photosynthetic rate Amax, canopy light use efficiency LUE, and the response coefficient k related to water vapor stress. Respiratory parameters: baseline respiration rate BR, temperature response coefficient Q10, and correction coefficients a and b for respiration on soil moisture content. Initial values ​​P(0), reasonable value ranges (upper and lower bounds), time steps Δt (e.g., 30 minutes), maximum number of iterations Nmax, and convergence threshold ε are set for each parameter.

[0050] Read the driving data (meteorological data) and constraint data (flux observations). The driving data is used for forward model simulation; the constraint data is used to evaluate the model simulation error and construct the loss function.

[0051] The observation data is cleaned according to the time steps of the model run, and missing or outlier values ​​are preprocessed (imputed or deleted) to ensure that the observation sequence used for assimilation is available.

[0052] For each candidate parameter value P(k), the carbon cycle model (Equations 1-5) is run at a time step of 30 minutes to obtain the model-predicted flux sequence at each time point. (P(k)), where, j This represents different constraint fluxes (such as NEE, GEP, RE), where k is the node at each runtime.

[0053] The parameters are updated using the following loss function:

[0054] in, L Represents the loss function. m The number of flux data representing the constraints. n Characterization in the j The data types of data points in a group of observations Indicates the first j The standard deviation of each data point in the group of observations x Indicates the measured value. Indicates parameters P The modeling results, parameters P This represents the parameters to be trained.

[0055] Specifically, during the data assimilation process, respiratory and photosynthetic parameters are also given corresponding constraints. Giving corresponding constraints means incorporating the observation error of a certain flux into the loss function (as a weight), or incorporating the flux as an independent term into the joint loss function, thereby applying soft constraints to it during the parameter estimation process.

[0056] Specifically, for each respiratory and photosynthetic parameter, there is a given set of initial values. The model is run with these initial values ​​to obtain a set of simulation results. The loss function described above is then used to calculate the loss value, which is updated using the Adam method. This process is repeated until the loss value remains unchanged, indicating the optimal set of parameters. For example, when optimizing only NEE, the loss value is calculated by comparing the fitted NEE result with the actually observed NEE. This invention also explores the robustness of the algorithm, as described below: In existing technologies, traditional gradient descent-based parameter optimization methods generally suffer from sensitivity to initial parameter settings, easily leading to unstable optimization results or getting trapped in local optima.

[0057] S4, Multi-source data constraints.

[0058] In this embodiment, to further improve the fitting accuracy of the carbon cycle model at the levels of photosynthetic GEP and respiration RE, after completing the initial optimization constrained only by net ecosystem exchange (NEE), a multi-source flux constraint mechanism is introduced. Adaptive fusion optimization of the model is achieved through dynamic feedback. This process includes the following steps: In the initial stage, the loss function is constructed using only NEE observation data as constraint variables:

[0059] Where, x NEE For the observed NEE data, NEE (P) represents the carbon cycle model in terms of parameters. The simulation results below, Given the standard deviation of NEE, the adaptive moment estimation (Adam) algorithm is used to update the parameters, resulting in a preliminarily optimized parameter set. 1. The goal of this optimization stage is to ensure that the model has a good fit on the scale of the overall carbon balance (NEE).

[0060] The second step is to identify fluxes with insufficient constraints. Based on the preliminary optimization parameters... 1. The driving model calculates the corresponding simulated sequences of photosynthetic flux (GEP) and respiratory flux (RE). By calculating the goodness-of-fit R² of different fitted fluxes (such as NEE, GEP, and RE), the system automatically evaluates which fluxes have poor fitting effects and incorporates the errors of these fluxes into the joint loss function for optimization, thereby achieving adaptive constraints on multi-source data. Specifically, the steps are as follows: The simulation results are compared with the observed data, and the goodness of fit R for each flux is calculated. 2 ;

[0061] Where, j∈{NEE,GEP,RE}, , , The three parameters in the corresponding parentheses are i The predicted value, observed value, and observed mean at any given time.

[0062] The system automatically identifies flux components with low fitting accuracy. For example, when the goodness of fit R of NEE is low... 2 NEE >0.7, while the goodness of fit of RE is RRE 2 If the value is less than 0.4 and the difference between the two exceeds 0.3, the system determines that the RE flux constraint is insufficient.

[0063] The multi-source constraint stage involves incorporating the poorly fitting flux into the joint loss function and performing joint optimization with NEE: *

[0064] If GEP performs poorly, the loss function described above is modified as follows: *

[0065] The Adam algorithm is used to update the parameters under this joint loss function, resulting in a new parameter set P2*.

[0066] During optimization, the system automatically monitors the residual changes for each flux and dynamically adjusts the weights of each flux based on error feedback, achieving adaptive constraint fusion. Convergence and output phase: Optimization stops when the change in the joint loss function falls below a preset threshold or the maximum number of iterations is reached. The final optimized parameters P2* are output, and the corresponding NEE, GEP, and RE simulation results are generated. After multi-source constraint optimization, the model's goodness of fit for each carbon flux is significantly improved, especially the simulation accuracy for respiratory flux (RE), thereby enhancing the model's overall predictive ability for the carbon source and sink status of the ecosystem.

[0067] Specifically, multi-source data constraints refer to automatically determining which fluxes need enhanced constraints based on their fitting performance during the assimilation process and dynamically adding them to the loss function. When fluxes with high-quality observation data are introduced into the constraints, the model can reduce the risk of overfitting to a single flux (such as NEE) and improve generalization ability. Adaptive fusion is achieved through a weight adjustment mechanism, which can dynamically balance the contributions of different fluxes without manual intervention. Since the previous step only optimized NEE (optimization only for NEE means that the loss function only considers the error term of NEE), in addition to NEE, the model can also fit GEP and RE. These three fluxes are three important fluxes. After optimization using only NEE, the fitting effect of GEP and RE may not be good. Therefore, it is necessary to add the data with poor constraint effect to the loss function.

[0068] After constraining the carbon flux (constraint means obtaining the optimal set of model parameters by only considering the loss function of NEE), the photosynthetic flux and respiration flux are fitted based on the carbon cycle model (corresponding to the optimal parameters obtained only for NEE optimization). The corresponding parameters can also be used to calculate GEP and RE according to the formulas mentioned above. Fluxes with poor fitting results are automatically constrained based on the fitting effect (the fitting process uses assimilated parameters, based on the model structure, and meteorological data to obtain the fitted flux results. Assimilation refers to using the Adam method and obtaining the parameters using the steps described above). After completing the initial fitting of the total carbon flux (NEE) (meaning the loss function only includes the NEE values) and parameter optimization, further fitting analysis can be performed on the two core biological processes in the carbon cycle model, photosynthetic flux (GEP) and respiration flux (RE), based on the optimization results. Since NEE is essentially a synthesis of GEP and RE, its accurate simulation depends on the model's detailed characterization of these two processes; therefore, in-depth fitting of each sub-process is of great significance.

[0069] During the fitting process, the carbon cycle model is compared with the goodness of fit R between simulated values ​​(GEP and RE calculated by the model) and observed data (such as GEP and RE data obtained from site observations). 2 It automatically identifies carbon flux components with low fitting accuracy. For example, when the carbon cycle model fits NEE well (please provide an illustrative definition of "well") but has significant RE fitting error (photosynthesis goodness of fit R), it may fail to identify these components. 2 If the goodness of fit is 0.3 higher than that of the respiratory flux RE, the system will determine that there is a modeling bias or insufficient parameter estimation in the RE process. To improve the overall simulation accuracy, the system can dynamically adjust the assimilation strategy based on the error feedback mechanism (calculating the GEP and RE based on the optimal parameters obtained only for NEE as the loss function, and then calculating the goodness of fit R between the simulated GEP and RE). 2If a certain flux, such as RE, has a poor fit, then that flux is included as part of the loss constraint. The poorly fitted RE flux is incorporated into the optimization objective function and, together with NEE, forms a joint loss function. Loss function considering only NEE:

[0070] in, n j Let be the number of days observed, σ be the variance of the observed data, and x be the measured value of NEE. NEE data simulated for model parameters; The joint loss function is as follows: Considering the loss function of GEP, it is: *

[0071] The loss function after considering RE constraints is shown above; specifically, it involves multiplying the loss values ​​of RE. This process no longer requires human intervention but relies on the carbon cycle model's automatic evaluation of the residual structure, achieving adaptive constraint fusion of multi-flux data. Adaptive constraint fusion refers to the system's ability to automatically determine which fluxes have insufficient constraints based on the fitting effects of different carbon flux components (NEE, GEP, RE) during model parameter optimization, and dynamically incorporate them into the loss function. This allows for simultaneous consideration of the constraint effects of multiple fluxes on parameters in subsequent optimizations, and also improves the fitting effect of constrained fluxes. This method enhances the model's predictive generalization ability: under different climatic backgrounds or time scales, it reduces overfitting caused by single flux control, enhancing the model's adaptability and generalization. In summary, this invention constructs a data-driven feedback optimization closed loop: by continuously monitoring the carbon cycle model's fitting effect and dynamically selecting the key constraints, it establishes an optimization process from identification and model response to strategy adjustment, possessing good versatility and automation characteristics.

[0072] This embodiment presents the fitting results of the carbon cycle model for primary productivity (GEP) and ecosystem respiration (RE) when net ecosystem carbon flux (NEE) is used as the constraint only. Although the carbon cycle model can reproduce the observed trend of NEE well, it has a significant bias in the simulation of RE, with a goodness of fit (R²) of only about 0.3, indicating that a single flux constraint is insufficient to fully reflect the carbon process components within the ecosystem. In contrast, the fitting effect is significantly improved when ecosystem respiration (RE) and carbon flux (NEE) are introduced as joint constraint variables during the assimilation process. It can be clearly seen that the model's ability to simulate RE is significantly improved, with the goodness of fit R² increasing to about 0.6, an improvement of more than 50%. This result shows that the introduction of multi-source observational data can effectively enhance the model's ability to express key ecological processes, especially demonstrating higher accuracy and stability in modeling respiration processes.

[0073] Example 2: Following the method for improving the accuracy of NEE estimation in forest ecosystems based on adaptive moment estimation provided in Example 1, this example provides a device for improving the accuracy of NEE estimation in forest ecosystems based on adaptive moment estimation. The device provided in this example can implement the method for improving the accuracy of NEE estimation in forest ecosystems based on adaptive moment estimation in Example 1. This device can be implemented through software, hardware, or a combination of both. For ease of description, this example is described by dividing the functionality into various units. Of course, in implementation, the functions of each unit can be implemented in one or more software and / or hardware components. For example, the device may include integrated or separate functional modules or units to execute the corresponding steps in the methods of Example 1. Since the device in this example is basically similar to the method example, the description process of this example is relatively simple. For relevant details, please refer to the description in Example 1. The example of the device for improving the accuracy of NEE estimation in forest ecosystems based on adaptive moment estimation provided by this invention is merely illustrative.

[0074] Specifically, this embodiment also provides a terrestrial ecosystem carbon source-sink assimilation prediction device based on adaptive moment estimation, comprising: The data collection unit is configured to collect meteorological datasets; The carbon flux prediction unit is configured to input meteorological datasets into an optimized carbon cycle model to predict carbon flux. The carbon cycle model includes a photosynthesis model and a respiration model; the photosynthesis model is used to obtain the photosynthetic flux. Respiratory models are used to obtain respiratory flux. The optimization process of the carbon cycle model is as follows: An adaptive moment estimation algorithm is used to assimilate the model parameters of the carbon cycle model and to constrain the carbon flux to obtain an initial optimized carbon cycle model. A carbon cycle model based on initial optimized model parameters for photosynthetic flux and respiratory flux Perform a fit and automatically adjust the flux for poor fits based on the fit results. and By applying constraints, the final optimized carbon cycle model is obtained.

[0075] Example 3: This example provides an electronic device corresponding to the method for improving the accuracy of forest ecosystem NEE estimation based on adaptive moment estimation provided in Example 1. The electronic device can be an electronic device for the client, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the method of Example 1.

[0076] like Figure 2 As shown, the electronic device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the method of Embodiment 1. The implementation principle and technical effects are similar to those of Embodiment 1, and will not be repeated here. Those skilled in the art will understand that... Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computing device on which the present application is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0077] In a preferred embodiment, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), and optical discs.

[0078] In a preferred embodiment, the processor can be any type of general-purpose processor such as a central processing unit (CPU) or a digital signal processor (DSP), and is not limited thereto.

[0079] Example 4: This example provides a computer-readable storage medium for storing one or more programs, the one or more programs including computer instructions, which, when executed by a computer, cause the computer to perform the method provided in Example 1 above.

[0080] In a preferred embodiment, the computer-readable storage medium may be a tangible device for holding and storing instructions executable, such as, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. The computer-readable storage medium stores computer program instructions that cause a computer to perform the method provided in Embodiment 1 above.

[0081] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In the description of this specification, the terms "a preferred embodiment," "furthermore," "specifically," "in this embodiment," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments in this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting carbon source-sink assimilation in terrestrial ecosystems based on adaptive moment estimation, characterized in that, The method comprises: collecting a meteorological data set; The meteorological data set is input into an optimized carbon cycle model to predict carbon flux, the carbon cycle model comprising a photosynthesis model and a respiration model, the photosynthesis model being used to obtain a photosynthesis flux , and the respiration model being used to obtain a respiration flux ; wherein the optimization process of the carbon cycle model is: performing data assimilation on model parameters of a carbon cycle model by using an adaptive moment estimation algorithm, and performing constraint on carbon fluxes to obtain an initially optimized carbon cycle model; Carbon cycle model based on initial optimized model parameters to photosynthesis flux and respiration flux is fitted, and fluxes with poor fitting effect are automatically constrained according to fitting effect and to obtain the final optimized carbon cycle model.

2. The adaptive moment estimation based terrestrial ecosystem carbon sink source assimilation prediction method according to claim 1, characterized in that, a photosynthesis model is expressed by using a Michaelis-Menten model; where J is the photosynthetic flux A is the primary productivity of the ecosystem; A max A is the maximum photosynthetic rate; I is the irradiance; is the canopy light use efficiency.

3. The adaptive moment estimation based terrestrial ecosystem carbon sink-source assimilation prediction method according to claim 2, characterized in that, a respiration model is expressed by using a Lloyd & Taylor model or a Q10-SW model; where RE is the ecosystem respiration; B R is the ecosystem basal respiration at the reference temperature; Q 10 is the ecosystem respiration response factor to temperature, T is the soil temperature, T ref is the reference temperature.

4. The adaptive moment estimation based terrestrial ecosystem carbon sink-source assimilation prediction method according to claim 3, characterized in that, Carbon flux Determined from the following equation: 。 5. The adaptive moment estimation based terrestrial ecosystem carbon sink-source assimilation prediction method according to claim 4, characterized in that, performing data assimilation on model parameters of a carbon cycle model by using an adaptive moment estimation algorithm, and performing constraint on carbon fluxes to obtain an initially optimized carbon cycle model, the process being: setting initial values, value ranges, time steps, maximum iteration numbers and convergence thresholds for model parameters of each carbon cycle model, wherein the model parameters of the carbon cycle model include photosynthesis parameters and respiration parameters; obtaining meteorological data and flux data, the meteorological data being used as driving data for forward simulation of the carbon cycle model, and the flux data being used as constraint data for evaluating simulation errors of the carbon cycle model and constructing a loss function; At each parameter candidate value P (k), the carbon cycle model is run by time step to obtain the model predicted flux sequence at each time ( P (k)), wherein, j indicates that the different constraint fluxes are NEE, GEP, and RE, respectively, and k is the node at each running time. updating the parameters by using the following loss function: wherein, L represents a loss function, m represents a number of flux data of a constraint, n represents a data type of a data point in the j group of observations, represents a standard deviation of each data point in the j group of observations, x represents a measurement value, P represents a parameter to be trained; running the carbon cycle model by using initial values of each respiration parameter and photosynthesis parameter to obtain a set of simulation results, calculating the loss values according to the above loss function, and updating the parameters by using the adaptive moment estimation algorithm after the loss values are obtained, repeating the above process until each loss value does not change, and obtaining an optimal set of parameters.

6. The adaptive moment estimation based terrestrial ecosystem carbon sink-source assimilation prediction method according to claim 5, characterized in that, The meteorological data includes temperature, photosynthetically active radiation, relative humidity and soil water content; and the flux data includes primary productivity, net ecosystem exchange and ecosystem respiration.

7. The adaptive moment estimation based terrestrial ecosystem carbon sink-source assimilation prediction method according to claim 5, characterized in that, Carbon cycle model based on initial optimized model parameters to photosynthesis flux and respiration flux is fitted, and the flux with poor fitting effect is automatically constrained according to the fitting effect and to obtain the final optimized carbon cycle model, the process is: When using NEE observation data as a constraint variable to construct a loss function and updating the parameters by using the adaptive moment estimation algorithm, a preliminary optimized parameter set is obtained. After the preliminary optimization of NEE, the photosynthesis flux GEP and respiration flux RE in the carbon cycle model were fitted separately based on the initial optimization results. In the fitting process, the carbon cycle model compared the fitting degree R between the simulated values GEP, RE calculated by the model and the observed GEP, RE data 2 , automatically identified the carbon flux components with lower fitting accuracy, dynamically adjusted the assimilation strategy based on the error feedback mechanism, and included the fluxes with poor fitting effect into the optimization objective function. NEE and the joint loss function were formed. During the optimization process, the residual changes of each flux were automatically monitored, and the weights of each flux were dynamically adjusted according to the error feedback, realizing adaptive constraint fusion.

8. A device for predicting carbon source-sink assimilation in terrestrial ecosystems based on adaptive moment estimation, characterized in that, The method comprises: a data collection unit configured to collect a meteorological data set; The carbon flux prediction unit is configured to input a weather dataset into an optimized carbon cycle model to predict the carbon flux, the carbon cycle model comprising a photosynthesis model for obtaining a photosynthesis flux and a respiration model for obtaining a respiration flux ; wherein the optimization process of the carbon cycle model is: performing data assimilation on model parameters of a carbon cycle model by using an adaptive moment estimation algorithm, and performing constraint on carbon fluxes to obtain an initially optimized carbon cycle model; Carbon cycle model based on initial optimized model parameters to photosynthesis flux and respiration flux is fitted, and fluxes with poor fitting effect are automatically constrained according to fitting effect and to obtain the final optimized carbon cycle model.

9. An electronic device, comprising: The method comprises: at least one processor; and a memory connected in communication with the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions for: The one or more programs include computer instructions for causing a computer to perform the method according to any one of claims 1-7.

Images