A method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems

Abstract

A method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems is applied to iterative prediction of carbon sources and sinks in terrestrial ecosystems, including: collecting meteorological data and carbon data of a target area to determine a data set; training a meteorological model through the data in the data set, predicting meteorological data through the trained meteorological model, and determining predicted meteorological data; training a carbon cycle model through the data in the data set, inputting the predicted meteorological data into the trained carbon cycle model, predicting carbon flux and carbon storage, and determining carbon cycle assimilation data; continuously iteratively updating the meteorological model through the currently acquired meteorological data, re-predicting the meteorological data through the updated meteorological model; and updating the carbon cycle assimilation data based on the re-predicted meteorological data. This method assimilates the model parameters of the carbon cycle model through the observed carbon data, realizes model data fusion, and further realizes effective prediction of short-term carbon sources and sinks.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon cycle, and in particular to a method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems. Background Art

[0002] The impact of climate change on the carbon cycle of terrestrial ecosystems and the feedback of the carbon cycle of terrestrial ecosystems to climate change are important tasks in global climate change research. In the past 20 years, there have been many calls to make ecology a more predictive science. Traditional ecological predictions are based on one-off predictions of future scenarios, with time scales ranging from 10 to even 100 years, and their reliability is difficult to assess and test. With the increase in the frequency of ecological data collection, iterative prediction techniques combined with new observational data and model data fusion methods have gradually developed, which can significantly improve the prediction accuracy and realize the transformation of ecological prediction from long time scales to short time scales, and from one-off predictions to iterative predictions. In recent years, short-term iterative ecological predictions have become an increasingly powerful ecological prediction tool and have been successfully applied in many fields such as biodiversity, biological invasions, hydrological management, environmental management, etc. Compared with predicting future trends using only early data, predictions reconstructed using iterative models produce higher predictive power. This represents a paradigm shift in ecological prediction, focusing on iterative predictions of short-term conditions. Although short-term iterative predictions have great promise in improving the understanding of ecological processes and the predictive skills of ecological models, they have not yet been widely used in the prediction of ecosystem carbon sources and sinks. Summary of the invention

[0003] In order to solve the problems existing in the prior art, the embodiments of the present application provide a method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems, a computing device, a computer storage medium, and a product including a computer program, which can effectively predict short-term carbon sources and sinks.

[0004] In the first aspect, an embodiment of the present application provides a method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems, which is applied to the iterative prediction of carbon sources and sinks in terrestrial ecosystems. The carbon sources and sinks are characterized by carbon flux and carbon storage. The method includes: collecting meteorological data and carbon data of the target area to determine a data set; the carbon data includes carbon flux and carbon storage; training a meteorological model through the data in the data set, predicting the meteorological data through the trained meteorological model, and determining the predicted meteorological data; training a carbon cycle model through the data in the data set, inputting the predicted meteorological data into the trained carbon cycle model, and predicting carbon flux and carbon storage; continuously iteratively updating the meteorological model through the currently acquired meteorological data, and re-predicting the meteorological data through the updated meteorological model; updating carbon flux and carbon storage based on the re-predicted meteorological data.

[0005] In some possible implementations, determining the data set includes processing the collected data to determine the data set.

[0006] In some possible implementations, the meteorological data includes temperature, photosynthetically active radiation, relative humidity, and soil moisture content.

[0007] In some possible implementations, carbon stocks include leaf, fine root and xylem biomass, soil organic carbon; carbon fluxes include gross primary productivity, net primary productivity, net ecosystem productivity.

[0008] In some possible implementations, the carbon cycle model is the ecosystem carbon model DALEC. The DALEC model integrates a data assimilation function. The DALEC model is driven by meteorological data and constrained by carbon data when performing model data fusion.

[0009] In some possible implementations, the model data fusion process is performed using a Bayesian parameter estimation method combined with a Markov chain-Monte Carlo method and a likelihood function solution.

[0010] In some possible implementations, the likelihood function is calculated according to the following formula:

[0011]

[0012] In the formula, L represents the likelihood function, m represents the number of carbon data, n represents the data type of the data point in the jth group of observations, σ j Characterizes the standard deviation of each data point in the jth group of observations, x j,i Characteristic measurement value, μ j,i (P) represents the modeling result of parameter P, and parameter P represents the parameter to be trained.

[0013] In some possible implementations, the meteorological model is a long short-term memory network (LSTM) deep learning model that considers an attention mechanism, including an input layer, an attention layer, two LSTM layers, and a fully connected output layer.

[0014] In a second aspect, an embodiment of the present application provides a computer-readable storage medium, comprising computer-readable instructions. When a computer reads and executes the computer-readable instructions, the computer executes the method as described in any one of the first aspects.

[0015] In a third aspect, an embodiment of the present application provides a computing device, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method described in any one of the first aspects is executed.

[0016] In a fourth aspect, an embodiment of the present application provides a product comprising a computer program, which, when the computer program product runs on a processor, enables the processor to execute the method as described in any one of the first aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems provided in an embodiment of the present application;

[0019] Figure 2 This is a schematic diagram showing the comparison of the prediction accuracy of a long short-term memory network (LSTM) deep learning meteorological model considering the attention mechanism and the original LSTM algorithm provided in an embodiment of the present application;

[0020] Figure 3 This is a schematic diagram of the relationship between a meteorological data forecast and a historical data period used, provided in an embodiment of the present application;

[0021] Figure 4 This is a diagram showing the effect of meteorological data prediction and the predicted days provided in an embodiment of the present application;

[0022] Figure 5 This is a comparison diagram of the effects of a Metropolis algorithm and an MH-BFGS algorithm provided in an embodiment of the present application;

[0023] Figure 6 It is a schematic diagram of a site-scale forest ecosystem assimilation prediction business process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] The term "and / or" in this article is a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The symbol " / " in this article indicates that the associated objects are in an or relationship, for example, A / B means A or B.

[0026] The terms "first" and "second" in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of the objects. For example, a first response message and a second response message are used to distinguish different response messages rather than to describe a specific order of the response messages.

[0027] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0028] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc.; multiple elements refer to two or more elements, etc.

[0029] To facilitate the understanding of the embodiments of the present application, the following will further explain and illustrate specific embodiments in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0030] First, the technical terms involved in the embodiments of the present application are introduced:

[0031] 1. Carbon source refers to the process, activity or mechanism that releases carbon into the atmosphere. In layman's terms, carbon source refers to the origin of carbon dioxide, which includes natural or man-made processes. For example, the burning of fossil fuels, deforestation and land use changes are all carbon sources because they lead to carbon dioxide (CO 2 ) concentration increases.

[0032] 2. Carbon sink refers to the process, activity or mechanism of reducing the concentration of greenhouse gases in the atmosphere by absorbing carbon dioxide from the atmosphere through measures such as afforestation and vegetation restoration. Vegetation-covered areas such as forests are the main carbon sinks, which can absorb CO through photosynthesis. 2 .

[0033] Next, the technical solution provided by the embodiments of the present application is introduced.

[0034] Figure 1The flowchart of a terrestrial ecosystem carbon source and sink iterative assimilation prediction method provided in an embodiment of the present application is shown, and the method is applied to the prediction of carbon source and sink in terrestrial ecosystem. The carbon source and sink are characterized by carbon flux and carbon storage. Figure 1 As shown, the prediction method may include the following steps:

[0035] S11: Collect meteorological data and carbon data of the target area and determine the data set.

[0036] In this embodiment, meteorological data is an important basic data for describing regional climate characteristics. Carrying out local long-term meteorological observations is of far-reaching significance for improving the accuracy of regional meteorological forecasts and disaster prevention and mitigation capabilities, and for understanding and responding to climate change. Meteorological data may include temperature, photosynthetically active radiation, relative humidity, and soil moisture content. Carbon data for terrestrial ecosystems generally refers to a series of data on carbon storage, carbon flux, carbon cycle, and carbon exchange in terrestrial ecosystems. These data are of great significance for understanding how ecosystems affect the global carbon cycle, climate change, and biodiversity. Carbon storage data include leaf, fine root and xylem biomass, soil organic carbon data, etc., and carbon flux includes gross primary productivity (GPP), net primary productivity (NPP), net ecosystem productivity (NEP), etc. In this embodiment, meteorological data may also be referred to as driving data, and carbon data may also be referred to as constraint data. Among them, the driving data can be obtained daily, and the acquisition cycle of the constraint data can be longer. For example, NEE flux data can be obtained every six months, litter data sets and canopy dynamic data can be obtained once a month, and soil organic carbon data can be obtained once every five years. When acquiring meteorological data of the target area, data can be collected through various sensors deployed in the target area, such as temperature sensors, humidity sensors, etc. Data acquisition can be passive acquisition, for example, a data collection instruction is sent to the sensor, and the sensor collects data after receiving the data collection instruction. Data acquisition can also be the periodic active acquisition of data by the sensor. After the data collection is completed, the collected data is integrated to determine the data set.

[0037] In some possible embodiments, the collected data may be processed.

[0038] In this embodiment, after the sensor acquires the data, the acquired data can be processed. Since the sensor may have a sensor fault light when collecting data, the collected data may be abnormal, making the acquired data illegal data. Therefore, it is necessary to process the collected data. Data processing may include cleaning up abnormal data, identifying and correcting (or deleting) errors, duplications or incomplete records in the data, etc. Data processing can achieve quality control of data.

[0039] S12: Train a meteorological model using the data in the data set, and predict the meteorological data using the trained meteorological model.

[0040] In this embodiment, after determining the data set, the meteorological model is trained based on the meteorological data in the data set. The meteorological model can be a neural network model, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), etc. In order to further enhance the prediction effect of the neural network, an attention mechanism can also be introduced during the neural network training.

[0041] In some possible embodiments, the meteorological model introduces an attention mechanism in the LSTM algorithm, which may include an input layer, an attention layer, two LSTM layers, and a fully connected output layer. For example, please refer to Figure 2 , Figure 2 This is a schematic diagram of the comparison of the prediction accuracy of a meteorological model and a native LSTM algorithm provided in an embodiment of the present application. Figure 2 As shown in the figure, the blue dots represent the prediction results of the meteorological model, and the red dots represent the prediction results of the native LSTM. It can be seen that the prediction results of the meteorological model combined with the attention mechanism are more concentrated, while the prediction results of the native LSTM are relatively discrete. It can be seen that the prediction effect of this meteorological model is better.

[0042] Through multiple rounds of neural network training, it is found that using historical data longer than 2 months has limited improvement in prediction accuracy, but significantly increases running time. For example, please refer to Figure 3 , Figure 3 The figure shows the relationship between meteorological data forecast and time. Figure 3 As shown in Figure 2, after 60 days, as the number of days increases, the improvement in prediction accuracy does not change much, but the running time (time) increases greatly. 2 _Ta represents the temperature determination coefficient, R 2 _PAR represents the coefficient of determination of photosynthetically active radiation, R 2 _SWC represents the soil moisture determination coefficient, R 2_RH represents the coefficient of determination of relative humidity. Therefore, in the embodiment of the present application, the historical data within two months stored in the data set is used to train the neural network.

[0043] For example, Figure 4 The following is a diagram showing the effect of a meteorological data forecast over time provided by an embodiment of the present application. Figure 4 As shown, as the number of days increases, the temperature determination coefficient R 2 _Ta, coefficient of determination of photosynthetically active radiation R 2 _PAR, changes with time, is relatively stable after the third day, and the soil moisture determination coefficient R 2 _SWC, relative humidity determination coefficient R 2 _RH, with the change of time, the prediction effect decreases significantly. The corresponding root mean square error is the opposite. After comprehensive consideration, predicting the meteorological data for the next 7 days is a more ideal choice. Therefore, the embodiment of the present application selects the data for the next 7 days predicted by using the historical data of the past 2 months as the best choice.

[0044] S13: The carbon cycle model is trained using the data in the data set, and the predicted meteorological data is input into the carbon cycle model to predict carbon flux and carbon storage.

[0045] In this embodiment, the carbon cycle model is trained according to the data in the data set. During training, the temperature, photosynthetically active radiation, relative humidity and soil moisture data in the climate data are used as driving data of the carbon cycle model, and the carbon data are used as constraint data of the carbon cycle model. After the prediction of the carbon cycle model is completed, the predicted meteorological data is used as the input of the carbon cycle model, and the prediction of carbon cycle assimilation is performed through the carbon cycle model.

[0046] In some possible embodiments, the carbon cycle model can select the data assimilation linked ecosystem carbon model (DALEC). The DALEC model is a medium-complexity ecosystem process model that can integrate the model data fusion function. The model covers the key processes of the ecosystem carbon cycle process, such as the carbon cycle, etc. with a relatively simple structure. The carbon cycle is the carbon source sink referred to in this embodiment, which may include carbon allocation and carbon turnover. The DALEC model integrates the data assimilation function, and the main structure of the model (carbon allocation, carbon turnover and other processes) is basically consistent with the international mainstream model. The relatively simple structure can avoid the overfitting phenomenon in the model data fusion. Its simple structure enables it to efficiently utilize computing resources when performing model data fusion calculations, so it has been widely used in the study of carbon cycle processes in various regions around the world. The DALEC model is a daily-scale box model that covers key ecosystem processes such as carbon absorption, carbon allocation, carbon turnover, and carbon output. In the DALEC model, it is driven by meteorological data and constrained by carbon data, thereby realizing model data fusion. The process of model data fusion is the process of adjusting parameters of the DALEC model. In the process of realizing model data fusion, the metropolis algorithm is mostly used. However, as the research on carbon cycle becomes more and more in-depth, the complexity of the model is also increasing. The efficiency of the metropolis algorithm greatly increases the time cost in the model fusion process. Therefore, in this embodiment, based on the metropolis algorithm, the fusion quasi-Newton method (BFGS) is used to perform model data fusion, and this method is called the MH-BFGS algorithm. In the algorithm, the Gaussian suggestion distribution is fitted to the target distribution, so that points in the high probability area are more likely to be proposed. For example, please refer to Figure 5 , Figure 5 : is a comparison chart of the effects of the metropolis algorithm and the MH-BFGS algorithm provided in the embodiment of the present application. Figure 5As shown in the figure, the parameters of the metropolis algorithm still did not converge after about 20,000 samples, while the MH-BFGS algorithm converged only after 2,000 samples. According to the Gelman-Rubin test, the metropolis algorithm converged in about 15,000 samples, while the MH-BFGS algorithm converged in about 1,400 samples. It can be seen that compared with the metropolis algorithm, the MH-BFGS algorithm converges faster, consumes less time, and has significantly improved computational efficiency. Due to the high-dimensional, nonlinear and non-Gaussian characteristics of ecosystem process models, it is difficult to represent the posterior distribution of parameters by analytical means. Therefore, it is necessary to use the likelihood function and Bayesian method, combined with the Markov chain-Monte Carlo method, to sample in the probability distribution space of the parameters to obtain parameter samples that are easy to apply directly, and then obtain the posterior distribution characteristics of the parameters. Specifically, a recommended parameter set is sampled from the prior distribution of the carbon cycle model parameters. This process can specify the sampling range through expert knowledge. Next, the forward model is used to generate simulated values ​​based on the proposed parameter set, and it is used together with the observed values ​​of the carbon data to estimate the likelihood and calculate the posterior probability density function. By comparing the two, if the value generated by the proposed parameter set is greater than the value of the current parameter set, then the proposed parameter is accepted and made the current parameter. On the contrary, the proposed parameter is accepted as the current parameter with a certain probability. Among them, the ratio of the new posterior probability to the current posterior probability probability is used as the probability of accepting the proposed parameter. These two steps are repeated until the acceptance rate converges to a specified threshold and falls within a small range of 50% (50±2.5%). In this embodiment, this process is repeated 200,000 times, and the last 150,000 parameter sets are selected for statistical analysis of the posterior distribution of the model parameters. Among them, the uncertainty in the constraint data has been taken into account through the likelihood function in the model data fusion as the weight of various observations in the integral likelihood method.

[0047] Specifically, the likelihood function is performed according to the following formula:

[0048]

[0049] In the formula, L represents the likelihood function, m represents the number of carbon data, n represents the data type of the data point in the jth group of observations, σ j Characterizes the standard deviation of each data point in the jth group of observations, x j,i Characteristic measurement value, μ j,i (P) represents the modeling result of parameter P, and parameter P represents the parameter to be trained.

[0050] S14: updating the meteorological model through the currently acquired meteorological data, re-predicting the meteorological data through the updated meteorological model, and updating the carbon cycle assimilation prediction data through the re-predicted meteorological data.

[0051] In this embodiment, the meteorological data is predicted by the trained meteorological model. In the next data acquisition cycle, after the meteorological data is acquired by the sensor, the acquired meteorological data is compared with the predicted meteorological data to determine the difference between the actual meteorological data and the predicted meteorological data. When the difference is greater than a preset threshold, the meteorological model is iteratively updated based on the latest acquired meteorological data. After the update is completed, the meteorological data is re-predicted by the updated meteorological model. The re-predicted meteorological data is used as the input of the cycle assimilation, and the carbon cycle assimilation prediction is re-performed.

[0052] In order to more clearly see the changes in prediction performance after iteratively adding new assimilated data, a test was conducted every two months, and the test results are shown in Table 1. From Table 1, it can be seen that iterative assimilation prediction can significantly reduce the uncertainty of optimization parameters and improve the prediction accuracy of net ecosystem exchange (NEE).

[0053] Table 1

[0054]

[0055] The above is an introduction to the iterative assimilation prediction method for carbon sources and sinks in terrestrial ecosystems provided in an embodiment of the present application. The meteorological model and carbon cycle assimilation are trained by acquiring the meteorological data and carbon data of the terrestrial ecosystem. The data predicted by the meteorological model is used as the input of the carbon cycle model to predict the carbon flux. In the next data acquisition cycle, the acquired data is compared with the predicted value. Based on the difference, the meteorological model is updated, and the meteorological data is re-predicted by the updated meteorological model. Based on the predicted meteorological data, the predicted carbon flux and carbon reserves are updated. Through this solution, the model parameters of the carbon cycle model can be assimilated through the observed carbon data to achieve model data fusion, thereby achieving effective prediction of short-term (next 7 days) carbon sources and sinks.

[0056] It is understandable that the size of the sequence number of each step in the above-mentioned embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. In addition, in some possible implementations, the steps in the above-mentioned embodiments can be selectively executed according to actual conditions, and can be partially executed or fully executed, which is not limited here. All or part of any feature of any embodiment of the present application can be freely and arbitrarily combined without contradiction. The combined technical solution is also within the scope of the present application.

[0057] Based on the method in the above embodiment, the embodiment of the present application provides a site-scale terrestrial forest ecosystem assimilation prediction business process. For example, Figure 6 The diagram shows the business process of assimilation prediction of forest ecosystem at the site scale. Relying on the high-frequency meteorological and carbon flux observation data of ChinaFLUX field ecological stations, we have developed a real-time assessment and prediction technology of carbon sinks in ecological stations based on artificial intelligence, process models, model data fusion and iterative prediction, and built a site-scale forest ecosystem assimilation prediction system. In this prediction system, we complete the following tasks: Figure 6 Assimilation forecasting business process shown in Figure 1. The system is connected to heterogeneous data collected by various sensors on site, including real-time meteorological and biometric observation data, which are archived after data cleaning and conversion.

[0058] like Figure 6 As shown, the entire business process includes data collection, data processing, execution process, and result display.

[0059] Among them, for data collection, since the system is connected to various sensors on site, the data of the field ecological station can be collected and transmitted in real time through the sensors.

[0060] During the data processing process, unexpected situations may occur when the sensor is collecting data, such as a sensor failure, which may lead to abnormal data collection. Using abnormally collected data will reduce the model prediction effect during model training. Therefore, it is necessary to process the collected data to ensure the quality of the data. The processed data is stored to form a historical database. The data stored in the historical database includes meteorological data and carbon data.

[0061] During the execution process, the meteorological model is trained through the meteorological data stored in the historical database. The gas phase model is a deep learning model. At the same time, the meteorological data is used as the driving data and the carbon data is used as the training data to train the carbon cycle model to achieve model data fusion. The meteorological data is predicted through the trained meteorological model to obtain future meteorological data. The future meteorological data is used as the driving input into the carbon cycle model to obtain the future carbon cycle dynamics, including carbon flux and carbon storage.

[0062] For result presentation, the predicted carbon flux and carbon storage can be displayed on the computing device.

[0063] At present, the system has been successfully applied at Dinghushan Station, Qianyanzhou Station, Huitong Station, Jianfengling Station and other stations, and has been deployed on the National Ecological Science Data Center website.

[0064] Based on the methods in the above embodiments, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the methods in the above embodiments.

[0065] Based on the methods in the above embodiments, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the methods in the above embodiments.

[0066] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0067] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0068] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)), etc.

[0069] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

Claims

1. A method for iterative assimilation prediction of carbon sources and sinks in terrestrial ecosystems, characterized in that: It is applied to iterative prediction of carbon sources and sinks in terrestrial ecosystems, where the carbon sources and sinks are characterized by carbon fluxes and carbon stocks, including leaf, fine root and xylem biomass, and soil organic carbon; The carbon flux includes gross primary productivity, net primary productivity, and net ecosystem productivity, and the method includes: Collecting meteorological data and carbon data of the target area to determine a data set; the carbon data includes carbon flux and carbon storage; the meteorological data includes temperature, photosynthetically active radiation, relative humidity and soil moisture content; The meteorological model is trained by the data in the data set, and the meteorological data is predicted by the trained meteorological model to determine the predicted meteorological data; the meteorological model is a long short-term memory network LSTM deep learning model considering the attention mechanism, including an input layer, an attention layer, two LSTM layers and a fully connected output layer; The carbon cycle model is trained by the data in the data set, and the predicted meteorological data is input into the trained carbon cycle model to predict carbon flux and carbon storage; the carbon cycle model is an ecosystem carbon model DALEC, and the DALEC model is integrated with a data assimilation function. When the DALEC model performs model data fusion, it is driven by the meteorological data and constrained by the carbon data; the model data fusion process adopts a Bayesian parameter estimation method, combined with a Markov chain-Monte Carlo method and a likelihood function solution; Iteratively updating the meteorological model through currently acquired meteorological data, and re-predicting the meteorological data through the updated meteorological model; The carbon flux and the carbon stock are updated based on the re-forecasted meteorological data.

2. The method according to claim 1, characterized in that The determining of the data set includes processing the collected data to determine the data set.

3. The method according to claim 1, characterized in that The likelihood function is calculated according to the following formula: In the formula, L represents the likelihood function, m represents the number of carbon data, and n j Characterizes the data type of the data point in the jth group of observations, σ j Characterizes the standard deviation of each data point in the jth group of observations, x j,i Characteristic measurement value, μ j,i (P) represents the modeling result of parameter P, and parameter P represents the parameter to be trained.

Images