A grid-scale ecosystem carbon and water flux assessment method based on flux observation

Abstract

The grid-scale ecosystem carbon and water flux assessment method based on flux observation disclosed in the present invention relates to the field of data processing. The present invention explores the potential of carbon flux data of surface ecosystems based on meteorological grids, constructs a carbon and water flux data set of meteorological grids, and conducts in-depth research on the theory, method and technology of carbon flux construction of meteorological grids. Utilizing global eddy covariance flux stations and remote sensing, ecological and geographical big data, based on machine learning, multiple sampling technology, remote sensing and geographic information and other technologies, hundreds of thousands of carbon flux machine learning models based on flux stations in ecosystems are developed and formed. An adaptive assessment framework and method system for migrating and applying carbon flux inversion models to meteorological grids is formed, and suitable carbon flux simulation models that can be migrated to meteorological grids are screened, thereby obtaining a carbon flux data set of meteorological grids, which can significantly improve the accuracy and reliability of carbon dynamic process assessment of regional / global scale ecosystems while deepening the research on carbon cycle of terrestrial ecosystems.

Description

Technical Field

[0001] The invention relates to the technical field of electrical digital data processing, and in particular to a grid-scale ecosystem carbon and water flux assessment method based on flux observation. Background Art

[0002] The spatiotemporal distribution of terrestrial ecosystem productivity and carbon sources and sinks is still not very clear. First, there is a serious lack of ecosystem carbon observation data and insufficient support for carbon monitoring. Eddy covariance flux towers are the mainstream channel for obtaining ecosystem carbon flux observation data, but the construction, observation, management and data processing costs of flux towers are high, resulting in a scarcity of global flux sites, only 721, and very uneven distribution. Carbon flux observation sites are scarce, unable to fully and effectively cover complex and diverse ecosystems, and even more unable to fully meet the needs of ecosystem carbon monitoring and carbon accounting, and unable to fully understand the driving mechanism of ecosystem carbon dynamics, which has become one of the main bottlenecks restricting the study of global terrestrial ecosystem carbon cycle. Second, traditional ecosystem carbon accounting technologies and methods face many limitations and challenges. Biogeochemical models / ecosystem models, atmospheric carbon dioxide inversion models, land surface process models and empirical statistical models, either alone or in combination, are the mainstream methods for carbon accounting and assessment. These technical methods or models are basically based on the integration and coupling of known mechanism processes or empirical knowledge, and they do not adequately consider or simplify the mechanisms and processes that are being studied or are not yet mature, and lack sufficient carbon observation data to support verification; the input data of these model technical methods are mostly spatial grid data with large uncertainties, such as reanalysis data, and the spatial resolution is usually 25-100km, which results in high uncertainty in the estimation results and is completely unable to meet the requirements for the accuracy of ecosystem carbon accounting.

[0003] Compared with the very rare eddy covariance flux stations, meteorological stations are more abundant and widely distributed, with nearly 30,000 publicly available worldwide, more than 40 times the number of flux stations; but the potential of meteorological stations in ecosystem carbon monitoring has not yet been tapped.

[0004] With the development of data mining theory, machine learning, remote sensing big data and other advanced technologies, it is possible to use limited and precious eddy covariance flux tower observation data to mine carbon flux data from widely distributed meteorological stations around the world, including building carbon flux artificial intelligence simulation models and carbon accounting models based on precious eddy covariance flux stations; migrating these models built based on eddy covariance flux stations to apply to the suitability assessment framework and flux simulation model selection technology and methods of meteorological stations; using the selected flux simulation models, rich climate stations and multi-source remote sensing data to build meteorological station carbon flux data sets with quasi-observational properties, etc. This is a major breakthrough in methods and technologies for ecosystem carbon monitoring and carbon accounting that lacks carbon flux observation data, which can significantly promote global carbon cycle research and significantly reduce the uncertainty of ecosystem carbon monitoring and carbon accounting due to insufficient carbon flux observation data in the past.

[0005] At present, there are relevant literatures on carbon and water flux inversion or simulation based on meteorological grids, such as the literature "M. Junge et al.: Synthesis and evaluation of the FLUXCOM approach. Biogeosciences, 17, 1343-1365, 2020", but the number of flux inversion models constructed by this type of technology is very limited. Among them, only one carbon and water flux model is established for all flux stations, or different carbon and water flux machine learning inversion models are constructed for different ecosystem types. These models cannot effectively characterize the basic characteristics of surface heterogeneity, and when these models are migrated and applied to meteorological grids, they have not undergone rigorous applicability and accuracy evaluation. Each grid can generate carbon and water flux values, which makes the accuracy and reliability of these data cannot be guaranteed. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a grid-scale ecosystem carbon and water flux assessment method based on flux observation.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A grid-scale ecosystem carbon and water flux assessment method based on flux observation, including:

[0009] Obtaining a flux station data set; the flux station data set includes: global eddy covariance flux station observation data and soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding stations;

[0010] Based on eddy covariance flux stations of different ecosystem types, different surface landscape combinations are constructed based on the flux station data sets according to different ecosystem types, regional scales and climate types; each constructed surface landscape combination includes at least one eddy covariance flux station; in each surface landscape combination, the flux station combination data set is composed of the corresponding eddy covariance flux station observation data and the soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding station;

[0011] Different training sets and test sets are constructed for each flux station combined data set to obtain a data combination partitioning scheme;

[0012] A carbon and water flux simulation model for each flux station is constructed for each data combination partitioning scheme;

[0013] The constructed carbon and water flux simulation model of the flux station was migrated and applied to the flux stations one by one to conduct suitability assessment, and the determination coefficient of the carbon and water flux simulation model and the Euclidean distance of each explanatory variable in the training set and test set of the carbon and water flux simulation model were obtained, as well as the comprehensive Euclidean distance between the training set and the test set, to generate a model evaluation data set;

[0014] constructing an estimation model of the determination coefficient of the carbon-water flux simulation model based on the model evaluation data set;

[0015] Determine the Euclidean distance between the influencing factors in the training set of the carbon and water flux simulation model and the corresponding influencing factors in the meteorological grid dataset, and construct the Euclidean distance dataset of the influencing factors of the meteorological grid and the comprehensive Euclidean distance set of the meteorological grid;

[0016] Using the estimation model of the determination coefficient of the carbon-water flux simulation model and the Euclidean distance data set of the meteorological grid influencing factors, predict the determination coefficient value of the carbon-water flux when the constructed flux station carbon-water flux simulation model is applied to the target regional grid or the global land grid;

[0017] The carbon-water flux model corresponding to the maximum determination coefficient of the grid prediction and greater than or equal to the determined threshold is selected as the grid applicable model, or the carbon-water flux model corresponding to the minimum comprehensive Euclidean distance between the training set of the carbon-water flux model and the meteorological grid data set and less than or equal to the determined threshold is determined as the grid applicable model;

[0018] The determined carbon-water flux simulation model is used to simulate the carbon-water flux of the target area meteorological grid at multiple time scales or the carbon-water flux of the global land meteorological grid at multiple time scales to construct a carbon-water flux dataset of the target area meteorological grid or a carbon-water flux dataset of the global meteorological grid.

[0019] Optionally, a carbon-water flux model corresponding to a maximum determination coefficient predicted by the grid and greater than or equal to a determined threshold is selected as a model applicable to the grid, specifically including:

[0020] Screening out the model with the determination coefficient value greater than or equal to 0.5 from all the constructed carbon and water flux simulation models of the flux stations;

[0021] Among the screened models, the model corresponding to the maximum determination coefficient value is determined as the carbon and water flux simulation model suitable for meteorological grid points in the target regional grid or the global land grid.

[0022] Optionally, determining a corresponding carbon-water flux model with a minimum comprehensive Euclidean distance between the training set of the carbon-water flux model and the meteorological grid data set and less than or equal to a determined threshold as a model applicable to the grid specifically includes:

[0023] Determine the Euclidean distance of each explanatory variable, determine the weights of the Euclidean distances of different explanatory variables, and perform normalization based on the weights to generate a comprehensive Euclidean distance;

[0024] Extracting all comprehensive Euclidean distances corresponding to the determination coefficient values ​​between 0.5 and 0.55, determining the average value of all comprehensive Euclidean distances, and using the average value as the distance threshold;

[0025] Obtaining a Euclidean data set between different explanatory variables of the training set of the carbon and water flux model of the meteorological grid point and the flux station, and determining the comprehensive Euclidean distance of each record in the Euclidean data set between the meteorological station and the flux station, and using the comprehensive Euclidean distance as the judgment distance;

[0026] The carbon-water flux simulation model corresponding to the determination distance that is less than the distance threshold is determined as a carbon-water flux simulation model suitable for meteorological grid points in a target area grid or a global land grid.

[0027] Optionally, for each flux station combination data set, a k-fold cross-validation method is used to construct different training sets and test sets to obtain the data combination partitioning scheme.

[0028] Optionally, based on the model evaluation data set, a machine learning / deep learning algorithm is used to construct an estimation model for the determination coefficient of the carbon-water flux simulation model.

[0029] A computer-readable storage medium is used to store a computer program, and the computer program is used to implement the above-mentioned grid-scale ecosystem carbon and water flux assessment method based on flux observation.

[0030] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0031] The grid-scale ecosystem carbon and water flux assessment method based on flux observation provided by the present invention explores the potential of carbon flux data of surface ecosystems based on meteorological grids, constructs a carbon and water flux data set of meteorological grids, and conducts in-depth research on the theory, method and technology of carbon flux construction of meteorological grids. Utilizing global eddy covariance flux stations and remote sensing, ecological and geographical big data, based on machine learning, multiple sampling technology, remote sensing and geographic information and other technologies, we develop hundreds of carbon flux machine learning models based on flux stations in ecosystems and form an adaptive assessment framework and method system for carbon flux inversion model migration and application to meteorological grids, screen suitable carbon flux simulation models that can be migrated to meteorological grids, and then obtain carbon flux data sets of meteorological grids, which can significantly improve the accuracy and reliability of carbon dynamic process assessment of regional / global scale ecosystems while deepening the research on carbon cycle of terrestrial ecosystems. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 A flow chart of a grid-scale ecosystem carbon and water flux assessment method based on flux observation provided by the present invention;

[0034] Figure 2 A schematic diagram of the implementation principle of the grid-scale ecosystem carbon and water flux assessment method based on flux observation provided by the present invention;

[0035] Figure 3 Pathway diagram for the comprehensive Euclidean distance threshold provided by the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0037] The purpose of this invention is to provide a grid-scale ecosystem carbon and water flux assessment method based on flux observation, which can deepen the carbon cycle of terrestrial ecosystems, significantly improve the accuracy and reliability of regional / global scale ecosystem carbon dynamic process assessment, and thus provide favorable technical support for ecosystem carbon monitoring and carbon accounting.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] With the development of data mining theory, machine learning, remote sensing big data and other advanced technologies, it is possible to use limited and precious eddy covariance flux tower observation data to mine carbon flux data from widely distributed meteorological stations around the world, including building carbon flux artificial intelligence simulation models and carbon accounting models based on precious eddy covariance flux stations, and migrating these models built based on eddy covariance flux stations to apply to the suitability assessment framework of meteorological grids and flux simulation model selection technology and methods. Using the selected flux simulation models, rich climate grids and multi-source remote sensing data to build meteorological grid carbon flux data sets, this is a major breakthrough in methods and technologies for ecosystem carbon monitoring and carbon accounting that lacks carbon flux observation data, which can significantly promote global carbon cycle research and significantly reduce the uncertainty of ecosystem carbon monitoring and carbon accounting due to insufficient carbon flux observation data in the past.

[0040] Based on this, Figure 1 As shown, the grid-scale ecosystem carbon and water flux assessment method based on flux observation provided by the present invention includes:

[0041] Step 100: Obtain flux station data set. The flux station data set includes: global eddy covariance flux station observation data and soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding stations. The grid meteorological spatial data set includes: grid meteorological data and soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding grid. The grid meteorological data here refers to ERA5 reanalysis data or grid meteorological data based on spatial difference of meteorological stations. Soil, vegetation and terrain data include soil texture, land cover type and location terrain, etc. Remote sensing biophysical parameter data include vegetation index, band spectrum, water index, surface temperature and surface humidity, etc.

[0042] Step 101: Based on a limited number of eddy covariance flux stations of different ecosystem types, different surface landscape combinations are constructed based on flux station data sets according to different ecosystem types, regional scales and climate types. Each constructed surface landscape combination contains at least one eddy covariance flux station. For example, each constructed surface landscape combination has 1, 2, ... or all flux stations. In each surface landscape combination, the flux station combination data set is composed of the corresponding eddy covariance flux station observation data and the soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding station.

[0043] Step 102: construct different training sets and test sets for each flux station combined data set to obtain a data combination partitioning scheme. The training set and test set in each data combination partitioning scheme may be used to characterize a surface landscape or scenario.

[0044] In the actual application process, for each flux station combination data set, the k-fold cross-validation method can be used to construct different training sets and test sets to obtain a data combination partitioning scheme. For example, k = 10, that is, each flux station combination data set is randomly divided into 10 parts, one of which can be taken as a test set, and the remaining 9 parts can be taken as training sets, so there are 10 combination schemes. You can also take 2 of them as test sets and the remaining 8 as test sets, so there are 45 such combination schemes. Continue in this way until 5 parts are taken as test sets and the remaining 5 parts are taken as training sets, and there are 252 such combinations. There are a total of 537 permutation and combination schemes. If k = 20, according to the above combination partitioning scheme, more combination schemes can also be obtained.

[0045] Step 103: Construct a carbon-water flux simulation model for each flux station for each data combination partitioning scheme.

[0046] In the actual application process, remote sensing big data and machine learning / deep learning technology are integrated, and three machine learning models with the highest accuracy are selected based on the meta-analysis of the literature. For each permutation and combination scheme, a flux station carbon and water flux simulation model is constructed. For example, if the value of k is 10, the number of flux station carbon and water flux simulation models constructed is: 537 combinations × 3 (machine learning) × n (different ecosystem types, regional scales, climate types), which can reach tens of thousands or even tens of thousands. The inputs of these constructed models include temperature, precipitation, saturated water vapor pressure difference, air pressure, downward shortwave radiation, atmospheric relative humidity, wind speed, wind direction and sunshine hours. The output is carbon and water flux at different time scales.

[0047] Step 104: The constructed flux station carbon and water flux simulation model is migrated and applied to the flux sites one by one to perform suitability assessment, and the determination coefficient of the carbon and water flux simulation model and the Euclidean distance of each explanatory variable in the training set and test set of the carbon and water flux simulation model are obtained, as well as the comprehensive Euclidean distance between the training set and the test set to generate a model evaluation data set.

[0048] In this step, the carbon and water flux simulation models of these flux stations are migrated and applied to the meteorological grids one by one, and the suitability assessment is performed using a suitability assessment framework and method system similar to that of the migration to the meteorological station (see Chinese patent announcement number CN116050163B). This assessment process mainly calculates the determination coefficient R of the carbon and water flux simulation model. 2and the Euclidean distance of each explanatory variable in the model training and test sets.

[0049] Step 105: Construct an estimation model for the determination coefficient of the carbon-water flux simulation model based on the model evaluation data set. This step mainly uses machine learning / deep learning algorithms to construct the determination coefficient R of the carbon-water flux simulation model based on the model evaluation data set. 2 The estimation model is denoted as R 2 -M, which is convenient for the flux station to verify the accuracy and applicability of the carbon and water flux simulation model.

[0050] Step 106: Determine the Euclidean distance between the influencing factors in the training set of the carbon and water flux simulation model and the corresponding influencing factors in the meteorological grid dataset, and construct the Euclidean distance dataset of the influencing factors of the meteorological grid and the comprehensive Euclidean distance set of the meteorological grid.

[0051] Step 107: Using the estimation model of the determination coefficient of the carbon and water flux simulation model and the Euclidean distance dataset of the meteorological grid influencing factors, predict the determination coefficient value of the carbon and water flux when the constructed flux station carbon and water flux simulation model is applied to the target regional grid or the global land grid.

[0052] Step 108: Select the carbon-water flux model corresponding to the maximum determination coefficient of the grid prediction and greater than or equal to the determined threshold as the grid applicable model, or determine the carbon-water flux model corresponding to the minimum comprehensive Euclidean distance between the training set of the carbon-water flux model and the meteorological grid data set and less than or equal to the determined threshold as the grid applicable model. Specifically:

[0053] If the carbon-water flux simulation model estimates the R 2 The threshold is set to 0.5, and the estimated R 2 ≥0.5, and select the largest R 2 The carbon and water flux simulation model corresponding to the value is used as the carbon and water flux simulation model of the grid point. By processing all meteorological grid points one by one in this way, the carbon and water fluxes at multiple time scales of the target area or the global meteorological grid can be simulated.

[0054] If a certain meteorological grid cannot be matched to a suitable carbon-water flux machine learning model, the meteorological grid will not be able to produce carbon-water fluxes with satisfactory accuracy.

[0055] Based on this, another solution is provided:

[0056] Determine the Euclidean distance of each explanatory variable, determine the weights of the Euclidean distances of different explanatory variables, and perform normalization based on the weights to generate a comprehensive Euclidean distance. Extract all comprehensive Euclidean distances corresponding to the coefficient of determination between 0.5 and 0.55, determine the average of all comprehensive Euclidean distances, and use the average as the distance threshold.

[0057] The Euclidean distance between the flux station influencing factor and the meteorological grid is calculated to construct a meteorological grid Euclidean distance dataset. For each record of the meteorological grid Euclidean dataset, the comprehensive Euclidean distance of each day is calculated. If a comprehensive Euclidean distance is less than the distance threshold, the carbon and water flux simulation model corresponding to the comprehensive Euclidean distance is considered to meet the accuracy requirements and can be extended to the relevant meteorological grid points to invert the carbon and water flux of this grid point.

[0058] If the comprehensive Euclidean distance between a model training set and a meteorological grid point is less than the distance threshold and is the smallest, the corresponding carbon-water flux simulation model can be used as the carbon-water flux simulation model of the grid point. By processing all meteorological grid points one by one according to this step, the carbon-water flux data of the target area or global meteorological grid at multiple time scales can be simulated.

[0059] Step 109: Use the determined carbon-water flux simulation model to simulate the carbon-water flux of the target area meteorological grid at multiple time scales or the carbon-water flux of the global land meteorological grid at multiple time scales to construct a carbon-water flux dataset of the target area meteorological grid or a carbon-water flux dataset of the global meteorological grid.

[0060] The following is a specific example to illustrate the specific implementation process of the grid-scale ecosystem carbon and water flux assessment method based on flux observation provided by the present invention. Figure 2 As shown, the specific implementation process in this embodiment is:

[0061] Step-1: Data collection and processing.

[0062] Specifically, the datasets that meet the carbon and water flux monitoring and evaluation requirements of flux stations in a given area are collected and processed. The given area can be a river basin, a region, a country or a continent, or the entire land. The datasets mainly include the eddy covariance flux station and meteorological grid data in the region, as well as the soil texture, land cover type, topography and corresponding remote sensing-based surface biophysical parameter data (such as vegetation index, slope spectrum, water index, surface temperature, surface humidity, etc.) that match the station or grid. The grid meteorological spatial datasets are usually derived from meteorological reanalysis data and spatial interpolation datasets of meteorological stations. The mainstream reanalysis datasets include ERA5 reanalysis dataset (the fifth generation atmospheric reanalysis dataset of global climate from January 1950 to the present by the European Center for Medium-Range Weather Forecasts), NCEP / NCAR reanalysis dataset (produced jointly by the National Center for Meteorological and Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR)), CFSR (global reanalysis data product inverted by the National Center for Environmental Prediction using the Global Forecast System) and NASA / MERRA reanalysis dataset. The above data were quality controlled and standardized, and the geographic information system and remote sensing technology were used to generate carbon and water flux datasets with remote sensing information and carbon and water flux datasets without remote sensing information in a given area, and meteorological grid datasets with remote sensing information and meteorological grid datasets without remote sensing information in a given area ( Figure 2 Dataset 2).

[0063] Step-2: Construction of digital simulation model of carbon and water flux in eddy covariance flux tower.

[0064] Specifically, using the carbon and water flux datasets containing remote sensing information and the carbon and water flux datasets without remote sensing in the given area, according to the overall (overall), vegetation function type (Plant Function Type, PFT), climate type (arid, semi-arid, semi-humid, humid) and other ecosystem combination types, according to the requirements of the K-fold cross-validation model, through random sampling, hundreds of thousands of carbon and water flux data subsets (day, 8-day, month and quarterly scales) including test sets and validation sets are constructed to characterize hundreds of thousands of ecosystem scenarios. Using machine learning algorithms and deep learning algorithms, through hyperparameter optimization and K-fold cross-validation, thousands of eddy covariance flux tower carbon and water flux machine learning or deep learning numerical simulation models are constructed. For example, taking the random forest model algorithm as an example, hundreds of thousands of eddy covariance flux tower carbon and water flux random forest simulation models (Random Forest Model, RFM) are constructed to simulate multi-time scale carbon and water fluxes in thousands of ecosystem scenarios. The determination coefficient R is used. 2 Or relative error characterizes the accuracy of the RFM model on the test set.

[0065] Here, the machine learning algorithms and deep learning algorithms that have been used are listed: Long Short-Term Memory deep learning model, Random Forest algorithm learning (RandomForest, RF), support vector machine (support vector machine, SVM), back-propagation artificial neural network (back-propagation artificial neural network, BP-ANN), ModellingAnimationAndRenderingSystem, MARS, eXtreme Gradient Boosting (eXtreme Gradient Boosting, XGB), Multi-Layer Perceptron (Multi-Layer Perceptron, MLP), Stacking model, Boosting method and Bagging algorithm, etc.

[0066] The main PFT types considered are: cropland (CRO), deciduous broadleaf forest (DBF), evergreen broadleaf forest (EBF), evergreen needleleaf forest (ENF), grassland (GRA), mixed forest (MF), savannah (SAV), shrubland (SHR), wetland (WET), etc.

[0067] Step-3: Migrate the RFM model to the grid and build a grid-based carbon-water flux database.

[0068] Specifically, these RFM models are validated by using spatial cross validation (SCV) to verify the simulation accuracy of each eddy covariance flux station in the validation set one by one, generating thousands of determination coefficients (R 2 ). 2 The level of R determines the applicability of these models at the flux station. 2 The higher the value, the closer the training set of the model is to the flux station dataset ( Figure 2 The more similar the datasets are (usually R 2 ≥0.5), otherwise it will be lower (usually R 2<0.5). The similarity between data sets composed of the same explanatory variables in the RFM model can be represented by the Euclidean distance, that is, R 2 can be represented as a function of these Euclidean distances, where the R 2 The value depends on the geographical and ecological differences between the carbon flux influencing factors in the model training set and the corresponding influencing factors in the validation set. The Euclidean distance between the influencing factors in the training set and the validation set can be represented as d(x, y):

[0069]

[0070] Where x is the carbon flux influencing factor in the model training set, i.e., the explanatory variable, such as daily average temperature. y is the corresponding influencing factor in the model validation set. T is the number of samples of the factor.

[0071] The Euclidean distance between the coefficient of determination and the corresponding explanatory variables can form a data set of thousands of samples (the structure of the data set is shown in Table 1). Based on this data set, the coefficient of determination R can be constructed using a multivariate linear regression model and a machine learning algorithm. 2 Before these models are transferred to the meteorological grid, they are first 2 The prediction model is used to predict the R 2 Value. Usually R 2 ≥0.5, the model can be migrated to the meteorological grid point. 2 If it is less than 0.5, it is considered that the migration accuracy of the model is low and cannot be applied to the meteorological grid.

[0072] Table 1 Determination coefficient (R) of flux inversion model 2 ) and its impact factor Euclidean distance database structure table

[0073] ID <![CDATA[R 2 ]]> <![CDATA[d1]]> <![CDATA[d2]]> ··· <![CDATA[d u-1 ]]> <![CDATA[d u ]]> <![CDATA[d kw ]]> 1 2 ···

[0074] Using the inversion model constructed above, R 2 and its influencing factor (Euclidean distance) database, and use multivariate regression model or machine learning algorithm to construct the determination coefficient (R 2 ) simulation model R 2 -M. The following is based on the multiple linear regression algorithm to build R 2 Simulation model R 2 -M, for:

[0075] R 2 -M=a0+a1d1+a2d2+…+a u-1 d u-1 +a u d u .

[0076] Where z represents the R of the inversion model 2 a0, a1, a2, …, a u-1 , a u are all regression parameters. d1, d2, …, d u-1 , d u They are all Euclidean distances between the carbon flux influencing factors (explanatory variables) in the model training set and the carbon flux influencing factors of the flux stations in the model validation set.

[0077] In addition, the R of carbon flux inversion model can also be constructed based on machine learning algorithm. 2 The construction process of the simulation model is similar to the above method and will not be repeated here.

[0078] Furthermore, based on R 2 Simulation Model R 2 -M, build flux inversion model and transfer it to meteorological grid determination coefficient R 2 Database, evaluate the applicability of all carbon and water flux RFM models migrated to meteorological grids.

[0079] Using the R of the carbon flux inversion model 2 Simulation Model R 2 -M and the Euclidean distances between the explanatory variables of the carbon and water flux RFM models and the same variables at the meteorological grid points (as shown in Table 2), and the determination coefficients R of different carbon flux RFM models migrated to the meteorological grid points were calculated. 2 (d in Table 2 kw According to R 2 Threshold standards are used to evaluate whether the constructed flux station carbon flux inversion model can be migrated and applied to meteorological grids, or which meteorological grids can be selected to be applicable to the flux station carbon flux inversion model.

[0080] Table 2 Euclidean distance d between the influencing factors of the flux inversion model and the influencing factors of the meteorological grid u The comprehensive Euclidean distance d kw Database structure table

[0081] ID <![CDATA[d1]]> <![CDATA[d2]]> ··· <![CDATA[d u-1 ]]> <![CDATA[d u ]]> <![CDATA[d kw ]]> 1 2 ···

[0082] At this point, a framework and methodological system for evaluating the applicability of the carbon-water flux RFM model for migration and application to meteorological grids has been established.

[0083] Further, as another implementation of this embodiment, a data set of thousands of samples can be formed by the Euclidean distance between the determination coefficient and the corresponding explanatory variables and the comprehensive Euclidean distance. Figure 3 As shown, the comprehensive Euclidean distance d of each model kw The calculation process is:

[0084] Take d2,…,du-1 , d u It represents the Euclidean distance between different influencing factors (explanatory variables) in the carbon flux model training set and the corresponding factors of a flux station in the model validation set. Based on this, the comprehensive Euclidean distance d of each model is kw For: kw =d1×a1+d2×a2+…+d u-1 ×a u-1 +d u ×a u Among them, a1, a2, ..., a u These are the importance values ​​of the influencing factors and can be derived from RFM.

[0085] Then take the value between R 2 The comprehensive Euclidean distances corresponding to all carbon-water flux models between values ​​0.5 and 0.55 are calculated, and the average of these comprehensive Euclidean distances is used as the threshold.

[0086] If the comprehensive Euclidean distance calculated when the model is migrated to a certain meteorological grid point is less than or equal to the threshold, the model can be migrated to the meteorological grid point for simulation of carbon and water flux. If the comprehensive Euclidean distance calculated when the model is migrated to a certain meteorological grid point is greater than the threshold, it is considered that the migration accuracy of the model is low and cannot be applied to the meteorological grid point.

[0087] Finally, based on the eddy covariance flux station and meteorological grid carbon and water flux dataset ( Figure 2 The dataset 5) was used to analyze the carbon and water dynamics and characteristics of carbon fluxes in different ecosystems and ecosystem combinations in the region at the interannual, seasonal, and daily scales.

[0088] Based on the above description, the present invention applies cutting-edge technical means to explore the potential of ground observation data based on meteorological grids, and constructs a high-precision carbon flux dataset based on meteorological grids.

[0089] Therefore, the present invention can deepen the carbon cycle of terrestrial ecosystems and has the potential to significantly improve the accuracy of assessment of carbon dynamic processes in ecosystems at regional / global scales. This is of great significance for supporting the formulation of climate change response policies and the prediction of future climate change, and provides favorable theoretical and technical support for ecosystem carbon monitoring and carbon accounting.

[0090] In addition, the computer program form of the above-mentioned grid-scale ecosystem carbon and water flux assessment method based on flux observation can be stored in a computer-readable storage medium when it is implemented through a software functional unit and sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

[0091] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0092] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A grid-scale ecosystem carbon and water flux assessment method based on flux observation, characterized in that: include: Obtain flux station datasets; The flux station data set includes: global eddy covariance flux station observation data and soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding stations; Based on eddy covariance flux stations of different ecosystem types, different surface landscape combinations are constructed based on the flux station data sets according to different ecosystem types, regional scales and climate types; each constructed surface landscape combination includes at least one eddy covariance flux station; in each surface landscape combination, the flux station combination data set is composed of the corresponding eddy covariance flux station observation data and the soil, vegetation, terrain data and remote sensing biophysical parameter data of the corresponding station; Different training sets and test sets are constructed for each flux station combined data set to obtain a data combination partitioning scheme; A carbon and water flux simulation model for each flux station is constructed for each data combination partitioning scheme; The constructed carbon and water flux simulation model of the flux station was migrated and applied to the flux stations one by one to conduct suitability assessment, and the determination coefficient of the carbon and water flux simulation model and the Euclidean distance of each explanatory variable in the training set and test set of the carbon and water flux simulation model were obtained, as well as the comprehensive Euclidean distance between the training set and the test set, to generate a model evaluation data set; constructing an estimation model of the determination coefficient of the carbon-water flux simulation model based on the model evaluation data set; Determine the Euclidean distance between the influencing factors in the training set of the carbon and water flux simulation model and the corresponding influencing factors in the meteorological grid dataset, and construct the Euclidean distance dataset of the influencing factors of the meteorological grid and the comprehensive Euclidean distance set of the meteorological grid; Using the estimation model of the determination coefficient of the carbon-water flux simulation model and the Euclidean distance data set of the meteorological grid influencing factors, predict the determination coefficient value of the carbon-water flux when the constructed flux station carbon-water flux simulation model is applied to the target regional grid or the global land grid; The carbon-water flux model corresponding to the maximum determination coefficient of the grid prediction and greater than or equal to the determined threshold is selected as the grid applicable model, or the carbon-water flux model corresponding to the minimum comprehensive Euclidean distance between the training set of the carbon-water flux model and the meteorological grid data set and less than or equal to the determined threshold is determined as the grid applicable model; The determined carbon-water flux simulation model is used to simulate the carbon-water flux of the target area meteorological grid at multiple time scales or the carbon-water flux of the global land meteorological grid at multiple time scales to construct a carbon-water flux dataset of the target area meteorological grid or a carbon-water flux dataset of the global meteorological grid.

2. The grid-scale ecosystem carbon and water flux assessment method based on flux observation according to claim 1 is characterized in that: The carbon and water flux model corresponding to the maximum determination coefficient predicted by the grid and greater than or equal to the determined threshold is selected as the model applicable to the grid, including: Screening out the model with the determination coefficient value greater than or equal to 0.5 from all the constructed carbon and water flux simulation models of the flux stations; Among the screened models, the model corresponding to the maximum determination coefficient value is determined as the carbon and water flux simulation model suitable for meteorological grid points in the target regional grid or the global land grid.

3. The grid-scale ecosystem carbon and water flux assessment method based on flux observation according to claim 1, characterized in that: The corresponding carbon-water flux model with the minimum comprehensive Euclidean distance between the training set of the carbon-water flux model and the meteorological grid data set and less than or equal to the determined threshold is determined as the model applicable to the grid, specifically including: Determine the Euclidean distance of each explanatory variable, determine the weights of the Euclidean distances of different explanatory variables, and perform normalization based on the weights to generate a comprehensive Euclidean distance; Extracting all comprehensive Euclidean distances corresponding to the determination coefficient values ​​between 0.5 and 0.55, determining the average value of all comprehensive Euclidean distances, and using the average value as the distance threshold; Obtaining a Euclidean data set between different explanatory variables of a training set of a carbon and water flux model of a meteorological grid point and a flux station, and determining a comprehensive Euclidean distance of each record in the Euclidean data set, and using the comprehensive Euclidean distance as a judgment distance; The carbon-water flux simulation model corresponding to the determination distance that is less than the distance threshold is determined as a carbon-water flux simulation model suitable for grid points in a target area grid or a global land grid.

4. The grid-scale ecosystem carbon and water flux assessment method based on flux observation according to claim 1, characterized in that: For each flux station combination data set, a k-fold cross-validation method is used to construct different training sets and test sets to obtain the data combination partitioning scheme.

5. The grid-scale ecosystem carbon and water flux assessment method based on flux observation according to claim 1, characterized in that: Based on the model evaluation data set, a machine learning / deep learning algorithm is used to construct an estimation model for the determination coefficient of the carbon-water flux simulation model.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to implement the grid-scale ecosystem carbon and water flux assessment method based on flux observation as described in any one of claims 1 to 5.

Images