Multi-scale ecological system carbon metering method, metering model establishing method and system, equipment and medium

Abstract

The invention relates to a multi-scale ecological system carbon metering method, a metering model establishing method and system, equipment and a medium, and belongs to the technical field of ecological environment monitoring. The establishment method comprises the following steps: collecting continuous data of the carbon flux of the net ecosystem with different scales, and corresponding environmental factors and spatial heterogeneity parameters; performing data preprocessing on the continuous data of the carbon flux of the net ecosystem with different scales; performing feature construction on the continuous data of the carbon flux of the net ecosystem with different scales after data processing, the corresponding environmental factors and the spatial heterogeneity parameters; performing training data set construction on the data after feature construction; and based on the constructed training data set and the integrated learning model, completing the establishment of the multi-scale ecological system carbon measurement model. According to the invention, fusion of continuous data of the carbon flux of the net ecosystem with different scales is realized, and refined monitoring of the carbon of the ecosystem at different spatial positions is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological environment monitoring, and particularly relates to a multi-scale ecosystem carbon measurement method, a measurement model establishment method, and a system, equipment, and medium. Background Art

[0002] As the global demand for clean energy grows, the construction of large-scale photovoltaic power plants is increasing in ecologically fragile areas such as deserts and steppes. The deployment of photovoltaic panels alters the local microenvironment, influencing soil moisture, vegetation growth, and carbon flux. While some studies have measured ecosystem carbon storage using field sampling methods, in the specific context of photovoltaic power plants, there is still a lack of integrated solutions for achieving real-time monitoring of carbon absorption and emissions across the entire photovoltaic park, conducting real-time and detailed monitoring of different spatial locations (under photovoltaic panels, between photovoltaic panels, open spaces within the power station, and control areas), and accurately distinguishing the contributions of various components to carbon storage and carbon flux.

[0003] In the context of building new power systems, environmental and carbon emissions are also key concerns. Some researchers have built real-time energy management systems that use binary particle swarm algorithms to minimize energy consumption, CO2 emissions, and other pollutants, but the algorithms are relatively inefficient.

[0004] In desertified grasslands, photovoltaic installations can improve surface vegetation cover to a certain extent, potentially providing a carbon sequestration benefit. However, because human disturbances such as grazing and mowing are less noticeable in some areas but more common in others, how to exclude or quantify the impact of management activities, reliably interpolate missing data segments, and ensure the accuracy of observations remain pressing technical challenges.

[0005] At present, the monitoring of grassland ecosystem carbon flux mainly relies on the eddy covariance (EC) method and the automatic light-dark chamber (AC) method. EC can obtain large-scale net ecosystem carbon exchange (NEE) in real time, but it is prone to missed measurements due to the influence of turbulence intensity and instrument maintenance. AC has the advantages of stable points, little influence from weather conditions, and high observation accuracy, but lacks sufficient spatial representativeness. How to complement the advantages of the two types of observations and complete high-precision interpolation in the missed sections is an urgent need for evaluating carbon sink effects and conducting ecosystem carbon sink accounting. Traditional interpolation methods such as empirical regression or MDS are difficult to fully utilize multi-source features. Random forests have become an ideal statistical learning fusion tool due to their high tolerance to nonlinear relationships and built-in integration mechanisms. Summary of the Invention

[0006] In response to the above problems, the present invention provides a multi-scale ecosystem carbon measurement method, a measurement model establishment method, and a system, equipment, and medium.

[0007] The first object of the present invention is to provide a method for establishing a multi-scale ecosystem carbon accounting model, comprising:

[0008] Collect continuous data on net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters;

[0009] Data preprocessing was performed on continuous data of net ecosystem carbon flux at different scales;

[0010] The continuous data of net ecosystem carbon flux at different scales, the corresponding environmental factors, and spatial heterogeneity parameters after data processing are used to construct features;

[0011] Construct a training data set for the data after feature construction;

[0012] Based on the constructed training dataset and integrated learning model, the multi-scale ecosystem carbon measurement model is established.

[0013] In a specific embodiment of the present invention, the net ecosystem carbon flux continuous data at different scales includes net ecosystem carbon flux continuous data at a first scale and net ecosystem carbon flux continuous data at a second scale, and the unit step size of the first scale is larger than the unit step size of the second scale.

[0014] In a specific embodiment of the present invention, the data preprocessing of the continuous data of net ecosystem carbon flux at different scales includes:

[0015] The first preprocessing is performed on the continuous data of net ecosystem carbon flux at the first scale;

[0016] Performing a second preprocessing on the continuous data of net ecosystem carbon flux at the second scale;

[0017] Performing a third preprocessing on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing according to the unit step size of the first scale;

[0018] Aligning the continuous data of net ecosystem carbon flux at the first scale after the first preprocessing and the continuous data of net ecosystem carbon flux at the second scale after the third preprocessing in time series;

[0019] Based on the misalignment results, the missing segments in the continuous data of net ecosystem carbon flux at the first scale are marked.

[0020] In a specific embodiment of the present invention, the first preprocessing includes abnormal data removal;

[0021] The second preprocessing includes abnormal data elimination and unit unification;

[0022] The third preprocessing includes: performing integrated calculation on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing within the unit step of the first scale.

[0023] In a specific embodiment of the present invention, the processed continuous data of net ecosystem carbon flux at different scales, the corresponding environmental factors, and the spatial heterogeneity parameters are used to construct features, including:

[0024] All continuous data of net ecosystem carbon flux and corresponding environmental factors were used as flux features, and the sequences of flux features were filled with missing values;

[0025] The spatial location parameter in the spatial heterogeneity parameter is used as the heterogeneous feature, and the contribution weight of each spatial location of the spatial location parameter is used as the weighting coefficient or weight feature;

[0026] The second scale or all continuous data of net ecosystem carbon flux after sliding average and difference processing according to the preset time step are used as historical lagged features;

[0027] All flux characteristics, heterogeneous characteristics, weighting coefficients or weight characteristics, and historical lag characteristics are taken as input vectors, and the continuous data of net ecosystem carbon flux at the second scale is taken as output vector.

[0028] In a specific embodiment of the present invention, constructing a training data set for the feature-constructed data includes:

[0029] The data after feature construction is randomly divided into training set, validation set and test set;

[0030] The data of the training set is balanced.

[0031] A second object of the present invention is to provide a system for establishing a multi-scale ecosystem carbon accounting model, comprising:

[0032] Data collection module: used to collect continuous data of net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters;

[0033] Data preprocessing module: used to preprocess the continuous data of net ecosystem carbon flux at different scales;

[0034] Feature construction module: used to construct features of the continuous data of net ecosystem carbon flux at different scales, corresponding environmental factors, and spatial heterogeneity parameters after data processing;

[0035] Dataset construction module: used to construct training datasets for feature-constructed data;

[0036] Model training module: used to complete the establishment of a multi-scale ecosystem carbon accounting model based on the constructed training dataset and integrated learning model.

[0037] In a specific embodiment of the present invention, the feature construction module includes a first submodule, a second submodule and a third submodule;

[0038] The first submodule is used to take the continuous data of the second scale or all net ecosystem carbon fluxes and the corresponding environmental factors as input features, and fill in missing values in the sequence of input features;

[0039] The second submodule is used to take the spatial location parameter in the spatial heterogeneity parameter as an independent variable and the contribution weight of each spatial location of the spatial location parameter as a weighting coefficient or an independent feature;

[0040] The third submodule is used to use the second scale or all the continuous data of net ecosystem carbon flux after sliding average and difference processing according to a preset time step as historical lag features.

[0041] In a specific embodiment of the present invention, an update module is also included;

[0042] The updating module is used to update the incremental learning samples according to the real-time net ecosystem carbon flux continuous data.

[0043] A third object of the present invention is to provide a multi-scale ecosystem carbon measurement method, characterized by comprising:

[0044] The multi-scale ecosystem carbon accounting model generates ecosystem carbon accounting data based on real-time net ecosystem carbon flux continuous data and corresponding environmental factors.

[0045] In a specific embodiment of the present invention, the real-time net ecosystem carbon flux continuous data includes real-time net ecosystem carbon flux continuous data at a first scale and real-time net ecosystem carbon flux continuous data at a second scale.

[0046] In a specific embodiment of the present invention, the multi-scale ecosystem carbon accounting model generates ecosystem carbon accounting data based on real-time net ecosystem carbon flux continuous data and corresponding environmental factors, including:

[0047] Determine whether there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data;

[0048] According to the judgment result that there is no missing segment in the real-time first-scale net ecosystem carbon flux continuous data, the real-time first-scale net ecosystem carbon flux continuous data is used as a subsequent incremental learning sample;

[0049] According to the judgment result that there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data, the multi-scale ecosystem carbon accounting model outputs ecosystem carbon accounting data based on the real-time second-scale net ecosystem carbon flux continuous data and corresponding environmental factors.

[0050] A fourth object of the present invention is to provide an electronic device, comprising: a processor, wherein the processor is coupled to a memory;

[0051] The memory is used to store computer programs;

[0052] The processor is configured to execute the computer program stored in the memory, so that the electronic device executes the above method.

[0053] A fifth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a program or instruction, and when the program or instruction is run on a computer, the computer executes the method as described above.

[0054] A sixth object of the present invention is to provide a computer program product, comprising a computer program / instruction, which implements the above method when executed by a processor.

[0055] Beneficial effects of the present invention:

[0056] The multi-scale ecosystem carbon measurement method, measurement model establishment method, system, equipment, and medium of the present invention process and construct features of continuous data of net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters, and use the feature-constructed data as sample data of an integrated learning model to establish a multi-scale ecosystem carbon measurement model. While taking into account spatial heterogeneity parameters, the integrated learning model automatically captures the complex nonlinear relationship between continuous data of net ecosystem carbon flux at different scales, realizes the fusion of continuous data of net ecosystem carbon flux at different scales, and realizes refined monitoring of ecosystem carbon at different spatial locations.

[0057] Moreover, the ecosystem carbon measurement model of the present invention has high noise resistance and robustness, ensemble learning reduces the risk of overfitting, and is insensitive to outliers; the feature importance reveals the contribution of various environmental factors and spatial heterogeneity parameters to the continuous data of net ecosystem carbon flux, and has good interpretability.

[0058] Moreover, the present invention utilizes continuous data of net ecosystem carbon flux with a shorter unit scale to achieve high-precision interpolation of missing segments of continuous data of net ecosystem carbon flux with a longer unit scale.

[0059] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 A flowchart of a method for establishing a multi-scale ecosystem carbon accounting model according to an embodiment of the present invention is shown;

[0062] Figure 2 A framework diagram of a system for establishing a multi-scale ecosystem carbon accounting model according to an embodiment of the present invention is shown;

[0063] Figure 3 The present invention shows ecosystem carbon flux data obtained by a multi-scale ecosystem carbon accounting method according to an embodiment of the present invention;

[0064] Figure 4 A framework diagram of an electronic device according to an embodiment of the present invention is shown;

[0065] In the figure: 10, data acquisition module; 20, data preprocessing module; 30, feature construction module; 40, data set construction module; 50, model training module; 300, electronic device; 301, processor; 302, memory. DETAILED DESCRIPTION

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings 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 making creative efforts shall fall within the scope of protection of the present invention.

[0067] like Figure 1 As shown, a method for establishing a multi-scale ecosystem carbon accounting model according to an embodiment of the present invention includes:

[0068] S1. Collect continuous data on net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters;

[0069] S2. Data preprocessing of continuous data of net ecosystem carbon flux at different scales;

[0070] S3, constructing features of the continuous data of net ecosystem carbon flux at different scales, the corresponding environmental factors, and spatial heterogeneity parameters after data processing;

[0071] S4, constructing a training data set for the data after feature construction;

[0072] S5. Based on the constructed training dataset and ensemble learning model, complete the establishment of a multi-scale ecosystem carbon accounting model.

[0073] In certain embodiments of the present invention, in step S1, the continuous data of net ecosystem carbon flux at different scales include continuous data of net ecosystem carbon flux at a first scale and continuous data of net ecosystem carbon flux at a second scale, and the unit step size of the first scale is larger than the unit step size of the second scale;

[0074] The corresponding environmental factors refer to the environmental factors when continuous data on net ecosystem carbon flux are collected;

[0075] The environmental factors include PAR (abbreviation for photosynthetically active radiation), air temperature, relative humidity, soil temperature and moisture content, wind speed and direction, precipitation, etc.

[0076] The spatial heterogeneity parameter refers to the spatial information parameter of the observation equipment or device of the environmental factor when collecting continuous data of net ecosystem carbon flux;

[0077] The spatial heterogeneity parameters include installation location information of the observation equipment or device;

[0078] The net ecosystem carbon flux continuous data include net ecosystem carbon exchange (hereinafter referred to as NEE), net ecosystem respiration (hereinafter referred to as ER) and net ecosystem gross primary productivity (hereinafter referred to as GPP).

[0079] In some embodiments of the present invention, step S2 includes:

[0080] S2-1, performing a first preprocessing on the continuous data of net ecosystem carbon flux at the first scale;

[0081] S2-2, performing a second preprocessing on the continuous data of net ecosystem carbon flux at the second scale;

[0082] S2-3, performing a third preprocessing on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing according to the unit step size of the first scale;

[0083] S2-4, aligning the continuous data of net ecosystem carbon flux at the first scale after the first preprocessing and the continuous data of net ecosystem carbon flux at the second scale after the third preprocessing in time series;

[0084] S2-5. Based on the misalignment results, mark the missing segments in the continuous data of net ecosystem carbon flux at the first scale.

[0085] In certain embodiments of the present invention, the first preprocessing includes abnormal data removal;

[0086] The second preprocessing includes abnormal data elimination and unit unification;

[0087] The third preprocessing includes: performing averaging or footprint-weighted integration calculation on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing within a unit step of the first scale. For example, the integration calculation is averaging or footprint-weighted.

[0088] The footprint weighting is a well-known method in the technical field and will not be described in detail in this embodiment.

[0089] In some embodiments of the present invention, step S3 includes:

[0090] S3-1. All continuous data of net ecosystem carbon flux and corresponding environmental factors are used as flux features, and missing values of the input feature sequences are filled.

[0091] S3-2. Using the spatial location parameter in the spatial heterogeneity parameter as the heterogeneous feature, and using the contribution weight of each spatial location of the spatial location parameter as the weighting coefficient or weight feature;

[0092] S3-3. The continuous data of net ecosystem carbon flux at the second scale after sliding average and difference processing according to the preset time step is used as the historical lag feature;

[0093] Step S3-4: taking all flux characteristics, heterogeneous characteristics, weighting coefficients or weight characteristics, and historical lag characteristics as input vectors, and taking the continuous data of net ecosystem carbon flux at the second scale as output vectors;

[0094] Among them, the sliding average and difference processing are numerical processing methods well known in the technical field and will not be described in detail here.

[0095] The preset time step is a plurality of unit steps, illustratively 1-3 unit steps.

[0096] For example, the missing value filling method uses sliding average or linear interpolation or KNN method (nearest neighbor interpolation method) to fill the missing value.

[0097] For example, the contribution weight of each spatial location is calculated using carbon flux footprint analysis, expert experience, or the percentage of surface area covered by the observation report. Both carbon flux footprint analysis and expert experience are well-known processing methods in the art and will not be described in detail here.

[0098] In some embodiments of the present invention, step S4 includes:

[0099] S4-1. Randomly divide the data after feature construction into training set, validation set and test set;

[0100] S4-2. The training set data is balanced to balance the day and night or seasonal differences and prevent the model from being biased towards the dominant season.

[0101] For example, the data ratio of the training set, validation set and test set is 70:15:15.

[0102] For example, the balancing process adopts the SMOTE-ENN method, which is a data balancing process method commonly known in the art and will not be described in detail in the present invention.

[0103] In some embodiments of the present invention, step S5 includes:

[0104] S5-1. Perform model training on the ensemble learning model (for example, a random forest model) based on the training set. The model training includes cross-validating the training data and selecting optimal model hyperparameters, specifically including:

[0105] The RandomForestRegressor function from the scikit-learn package was used on the training set. Key hyperparameters included the number of trees (n_estimators) (300–500), the maximum depth (max_depth) (automatic or 15–30), the minimum number of leaf nodes (min_samples_leaf) (1–3), and the maximum number of features per tree (max_features) (√p or log2p). Five-fold cross-validation followed by grid search or Bayesian optimization was used to determine the optimal hyperparameter combination.

[0106] S5-2. Validate the trained model based on the validation set, including:

[0107] Monitoring R using a validation set 2, RMSE, and MAE; if overfitting occurs, increase min_samples_leaf or reduce max_depth. Output the model and persist it as a pkl file, recording the feature importance ranking.

[0108] S5-3. Evaluate the validated model based on the test set, specifically including:

[0109] Calculate R on the test set 2 , RMSE, Bias, NSE and other indicators to ensure R 2 ≥0.8, RMSE ≤1 μmol·m -2 ·s -1 Statistical errors are calculated for nighttime, daytime, and different seasons to verify the stability of the model under various operating conditions. The prediction interval is estimated using the 1-σ rule or Monte Carlo method, and the uncertainty is output.

[0110] The above-mentioned multi-scale ecosystem carbon accounting model is applied to the photovoltaic ecosystem as follows:

[0111] In step S1, the continuous data of net ecosystem carbon flux at different scales is obtained through EC and AC collection terminals. Therefore, before establishing the ecosystem carbon measurement model, it is necessary to deploy EC and AC collection terminals in the photovoltaic field. Specifically:

[0112] An eddy covariance flux tower (EC) is built in the center of the photovoltaic park, and 1-2 automatic light and dark boxes (AC) are placed under the photovoltaic panels, between the photovoltaic panels, in the open space, and in the control area. Each collection terminal is connected to the edge gateway through a wireless communication module to achieve unified time synchronization.

[0113] EC records three-dimensional wind speed and CO2 concentration at a frequency of 10 Hz, and EC outputs NEE_EC every 30 minutes;

[0114] AC automatically completes the light box and dark box measurement once every hour, outputs NEE_AC and ER_AC every second, and calculates GPP_AC by differential calculation of NEE_AC and ER_AC;

[0115] At the same time, environmental factors (PAR, air temperature, relative humidity, soil temperature and moisture content, wind speed and direction, precipitation, etc.) and spatial heterogeneity parameters (i.e., the locations of EC and AC collection terminals, such as under photovoltaic panels, between photovoltaic panels, open space, control area, etc.) were recorded;

[0116] From the above, it can be seen that the unit step size of the continuous data of net ecosystem carbon flux observed by EC is 30 minutes, which is the continuous data of net ecosystem carbon flux at the first scale. The unit step size of the continuous data of net ecosystem carbon flux observed by AC is per second, which is the continuous data of net ecosystem carbon flux at the second scale.

[0117] For step S2:

[0118] S2-1. Continuous data of net ecosystem carbon flux at the EC end: complete coordinate rotation, WPL correction, rain and snow removal, friction wind speed screening, and generate a quality flag.

[0119] S2-2, AC end net ecosystem carbon flux continuous data: linear fitting of CO2 change rate (R 2 ≥0.95), exclude leaking boxes; uniformly convert to μmol·m -2 ·s -1 .

[0120] S2-3. Average or footprint-weight the AC multi-point observations within a 30-min time window to obtain NEE_AC_avg and ER_AC_avg.

[0121] S2-4. Perform time series alignment on all data to generate a merged data table with a step length of 30 minutes;

[0122] S2-5. Based on the comparison of the continuous data of net ecosystem carbon flux at the EC end and the continuous data of net ecosystem carbon flux at the AC end in the combined data table, mark the missing segments of the continuous data of net ecosystem carbon flux at the EC end.

[0123] For step S3:

[0124] S3-1. Take NEE_AC_avg, ER_AC_avg, GPP_AC_avg, PAR, Ta, Ts, VPD, SWC, wind speed, wind direction (sin-cosine coding), day / night identification, month number, and DOY (sin-cosine coding) as flux features, and perform missing value filling (nearest neighbor interpolation method) on the sequence of flux features to avoid loss of training samples.

[0125] S3-2, the locations of EC and AC collection terminals (e.g., under photovoltaic panels, between photovoltaic panels, in open space, and in control areas) are used as heterogeneous features; the contribution weight of each area calculated by the footprint model is used as the weighting coefficient or weight feature;

[0126] S3-3. Perform sliding averaging and differencing of the NEE and ER time series over 1–3 time steps to capture the autocorrelation structure as a historical lag feature.

[0127] S3-4. All flux features, heterogeneous features, weighting coefficients or weight features, and historical lag features are used as input vectors, and the NEE_EC sequence after the above processing is selected as the output vector, and finally the data sample after feature construction is obtained.

[0128] For step S4:

[0129] All samples were randomly divided into a training set (70%), a validation set (15%), and a test set (15%). Methods such as SMOTE-ENN were used to balance the day and night or seasonal differences to prevent the model from being biased towards the dominant season.

[0130] For step S5:

[0131] Random Forest Model Training

[0132] The RandomForestRegressor from the scikit-learn package was called on the training set; key hyperparameters included the number of trees n_estimators (300–500), the maximum depth max_depth (automatic or 15–30), the minimum number of leaf nodes min_samples_leaf (1–3), and the maximum number of features per tree max_features (√p or log2p). Five-fold cross-validation with grid search or Bayesian optimization was used to determine the optimal hyperparameter combination. The validation set was used to monitor R. 2 , RMSE, and MAE; if overfitting occurs, increase min_samples_leaf or reduce max_depth. Output the model and persist it as a pkl file, recording the feature importance ranking.

[0133] Model Evaluation

[0134] Calculate R on the test set 2 , RMSE, Bias, NSE and other indicators to ensure R 2 ≥0.8, RMSE ≤1 μmol·m -2 ·s -1 Statistical errors are calculated for nighttime, daytime, and different seasons to verify the stability of the model under various operating conditions. The prediction interval is estimated using the 1-σ rule or Monte Carlo method, and the uncertainty is output.

[0135] like Figure 2 As shown, a system for establishing a multi-scale ecosystem carbon accounting model according to an embodiment of the present invention includes:

[0136] Data collection module 10: used to collect continuous data of net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters;

[0137] Data preprocessing module 20: used for data preprocessing of continuous data of net ecosystem carbon flux at different scales;

[0138] Feature construction module 30: used to construct features based on the processed net ecosystem carbon flux continuous data at different scales, corresponding environmental factors, and spatial heterogeneity parameters;

[0139] Dataset construction module 40: used to construct a training dataset for the data after feature construction;

[0140] Model training module 50: used to complete the establishment of a multi-scale ecosystem carbon accounting model based on the constructed training data set and integrated learning model.

[0141] In some embodiments of the present invention, the feature construction module 40 includes a first submodule, a second submodule, and a third submodule;

[0142] The first submodule is used to take the continuous data of the second scale or all net ecosystem carbon fluxes and the corresponding environmental factors as input features, and fill in missing values in the sequence of input features;

[0143] The second submodule is used to take the spatial location parameter in the spatial heterogeneity parameter as an independent variable and the contribution weight of each spatial location of the spatial location parameter as a weighting coefficient or an independent feature;

[0144] The third submodule is used to use the second scale after sliding average and difference processing according to a preset time step as the historical lag feature.

[0145] In certain embodiments of the present invention, the ecosystem carbon accounting model further comprises an update module;

[0146] The updating module is used to update the incremental learning samples according to the real-time net ecosystem carbon flux continuous data.

[0147] A multi-scale ecosystem carbon accounting method according to certain embodiments of the present invention includes:

[0148] Sa, the multi-scale ecosystem carbon accounting model of the above embodiment generates ecosystem carbon accounting data based on real-time net ecosystem carbon flux continuous data and environmental factors;

[0149] That is, the real-time net ecosystem carbon flux continuous data and the corresponding environmental factors are input into the multi-scale ecosystem carbon accounting model of the above embodiment, and the ecosystem carbon accounting data is output to achieve online statistics.

[0150] In certain embodiments of the present invention, the real-time continuous data of net ecosystem carbon flux is real-time continuous data of net ecosystem carbon flux at a first scale and continuous data of net ecosystem carbon flux at a second scale.

[0151] In some embodiments of the present invention, step Sa includes:

[0152] Sa-1: Determine whether there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data;

[0153] Sa-2, based on the judgment result of whether the real-time first-scale net ecosystem carbon flux continuous data is negative, the real-time first-scale net ecosystem carbon flux continuous data subsequent incremental learning samples;

[0154] Sa-3. Based on the judgment result that there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data, the multi-scale ecosystem carbon measurement model of the above embodiment outputs ecosystem carbon measurement data based on the real-time second-scale net ecosystem carbon flux continuous data and corresponding environmental factors.

[0155] In some embodiments of the present invention, step Sa-1 includes:

[0156] Sa-1-1, perform the first preprocessing on the real-time first-scale continuous data of net ecosystem carbon flux;

[0157] Sa-1-2, perform the second preprocessing on the real-time second-scale continuous data of net ecosystem carbon flux;

[0158] Sa-1-3, performing a third preprocessing on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing according to the unit step size of the first scale;

[0159] Sa-1-4, aligning the real-time first-scale continuous net ecosystem carbon flux data after the first preprocessing and the real-time second-scale continuous net ecosystem carbon flux data after the third preprocessing in time series;

[0160] Sa-1-5: Based on the misalignment results, determine whether there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data. Specifically:

[0161] If the real-time first-scale net ecosystem carbon flux continuous data after the first preprocessing and the real-time second-scale net ecosystem carbon flux continuous data after the third preprocessing are completely aligned, then there is no missing segment in the real-time first-scale net ecosystem carbon flux continuous data;

[0162] If the real-time first-scale net ecosystem carbon flux continuous data after the first preprocessing and the real-time second-scale net ecosystem carbon flux continuous data after the third preprocessing are not aligned, there is a missing segment in the real-time first-scale net ecosystem carbon flux continuous data.

[0163] For example, the ecosystem carbon measurement method of the above embodiment is applied to a photovoltaic ecosystem, specifically:

[0164] During the operation of the ecosystem carbon accounting model of the above embodiment, the model system calls the integrated learning model every 30 minutes: if the EC net ecosystem carbon flux continuous data is normal, the measured EC net ecosystem carbon flux continuous data is recorded and used as a subsequent incremental learning sample; if the EC net ecosystem carbon flux continuous data is missing, the AC net ecosystem carbon flux continuous data and environmental factors are input, and the model outputs the predicted value NEE_pred, and marks it as "interpolation". The interpolated value and the measured value are written into the database together to ensure the integrity of the NEE time series. Synchronously output ER_pred and GPP_pred (which can be calculated from AC data or inferred by the model secondary level) for carbon sink assessment. Specifically, the statistical data before and after the interpolation of the ecosystem carbon accounting model of the above embodiment are as follows: Figure 3 shown.

[0165] In certain embodiments of the present invention, the ecosystem carbon accounting model of the above embodiment can also implement an adaptive update step D, which specifically includes:

[0166] D-1. Incremental training of newly added continuous data of net ecosystem carbon flux at different scales;

[0167] D-2, if the evaluation index drops below the threshold (e.g. the threshold is R 2 <0.75), triggering a full retraining and replacing the old model.

[0168] For example, in a photovoltaic ecosystem, the adaptive update step D is specifically as follows:

[0169] Incremental training is performed on the newly added “measured EC + concurrent features” data every week or month; if the evaluation index drops below the threshold (such as R 2 <0.75), triggering a full retraining and replacing the old model.

[0170] In certain embodiments of the present invention, the ecosystem carbon accounting model of the above embodiment can realize the result output and visualization function, specifically the result output and visualization, wherein the output ecosystem carbon accounting data includes NEE, ER, and GPP reports at daily, monthly, and annual scales; and provides flux (i.e., data collection end) diagrams for under-plate, between-plate, open space, and control areas;

[0171] Visualization specifically includes uploading real-time data to the monitoring platform, supporting API calls and graphical dashboard displays.

[0172] like Figure 4 As shown, some embodiments of the present invention provide an electronic device, the electronic device 300 including: a processor 301, the processor 301 coupled to a memory 302;

[0173] The memory 302 is used to store computer programs;

[0174] The processor 301 is configured to execute the computer program stored in the memory 302 , so that the electronic device executes the method described in the above embodiment.

[0175] In certain embodiments of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores a program or instruction. When the program or instruction is executed on a computer, the computer executes the method described in the above embodiments.

[0176] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, an electronic device, or a device.

[0177] Certain embodiments of the present invention provide a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the method described in the above embodiments is implemented.

[0178] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements 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 establishing a multi-scale ecosystem carbon accounting model, characterized in that: include: Collect continuous data on net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters; Data preprocessing was performed on continuous data of net ecosystem carbon flux at different scales; The continuous data of net ecosystem carbon flux at different scales, the corresponding environmental factors, and spatial heterogeneity parameters after data processing are used to construct features; Construct a training data set for the data after feature construction; Based on the constructed training dataset and integrated learning model, the multi-scale ecosystem carbon measurement model is established.

2. The method for establishing a multi-scale ecosystem carbon accounting model according to claim 1, characterized in that: The net ecosystem carbon flux continuous data at different scales include net ecosystem carbon flux continuous data at a first scale and net ecosystem carbon flux continuous data at a second scale, and a unit step length of the first scale is greater than a unit step length of the second scale.

3. The method for establishing a multi-scale ecosystem carbon accounting model according to claim 1, characterized in that: The data preprocessing of continuous net ecosystem carbon flux data at different scales includes: The first preprocessing is performed on the continuous data of net ecosystem carbon flux at the first scale; Performing a second preprocessing on the continuous data of net ecosystem carbon flux at the second scale; Performing a third preprocessing on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing according to the unit step size of the first scale; Aligning the continuous data of net ecosystem carbon flux at the first scale after the first preprocessing and the continuous data of net ecosystem carbon flux at the second scale after the third preprocessing in time series; Based on the misalignment results, the missing segments in the continuous data of net ecosystem carbon flux at the first scale are marked.

4. The method for establishing a multi-scale ecosystem carbon accounting model according to claim 3, characterized in that: The first preprocessing includes abnormal data elimination; The second preprocessing includes abnormal data elimination and unit unification; The third preprocessing includes: performing integrated calculation on the continuous data of net ecosystem carbon flux at the second scale after the second preprocessing within the unit step of the first scale.

5. The method for establishing a multi-scale ecosystem carbon accounting model according to any one of claims 1 to 4, characterized in that: The processed net ecosystem carbon flux continuous data at different scales, the corresponding environmental factors, and spatial heterogeneity parameters are used to construct features, including: All continuous data of net ecosystem carbon flux and corresponding environmental factors were used as flux features, and the sequences of flux features were filled with missing values; The spatial location parameter in the spatial heterogeneity parameter is used as the heterogeneous feature, and the contribution weight of each spatial location of the spatial location parameter is used as the weighting coefficient or weight feature; The second scale or all continuous data of net ecosystem carbon flux after sliding average and difference processing according to the preset time step are used as historical lagged features; All flux characteristics, heterogeneous characteristics, weighting coefficients or weight characteristics, and historical lag characteristics are taken as input vectors, and the continuous data of net ecosystem carbon flux at the second scale is taken as output vector.

6. A system for establishing a multi-scale ecosystem carbon accounting model, characterized in that: include: Data collection module: used to collect continuous data of net ecosystem carbon flux at different scales, as well as corresponding environmental factors and spatial heterogeneity parameters; Data preprocessing module: used to preprocess the continuous data of net ecosystem carbon flux at different scales; Feature construction module: used to construct features of the continuous data of net ecosystem carbon flux at different scales, corresponding environmental factors, and spatial heterogeneity parameters after data processing; Dataset construction module: used to construct training datasets for feature-constructed data; Model training module: used to complete the establishment of a multi-scale ecosystem carbon accounting model based on the constructed training dataset and integrated learning model.

7. The system for establishing a multi-scale ecosystem carbon accounting model according to claim 6, characterized in that: The feature construction module includes a first submodule, a second submodule and a third submodule; The first submodule is used to take the continuous data of the second scale or all net ecosystem carbon fluxes and the corresponding environmental factors as input features, and fill in missing values in the sequence of input features; The second submodule is used to take the spatial location parameter in the spatial heterogeneity parameter as an independent variable and the contribution weight of each spatial location of the spatial location parameter as a weighting coefficient or an independent feature; The third submodule is used to use the second scale or all the continuous data of net ecosystem carbon flux after sliding average and difference processing according to a preset time step as historical lag features.

8. A multi-scale ecosystem carbon measurement method, characterized by: include: The multi-scale ecosystem carbon accounting model described in any one of claims 1-5 generates ecosystem carbon accounting data based on real-time net ecosystem carbon flux continuous data and corresponding environmental factors.

9. A multi-scale ecosystem carbon accounting method according to claim 8, characterized in that: The real-time net ecosystem carbon flux continuous data includes real-time net ecosystem carbon flux continuous data at a first scale and real-time net ecosystem carbon flux continuous data at a second scale.

10. A multi-scale ecosystem carbon accounting method according to claim 9, characterized in that: The multi-scale ecosystem carbon accounting model generates ecosystem carbon accounting data based on real-time net ecosystem carbon flux continuous data and corresponding environmental factors, including: Determine whether there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data; According to the judgment result that there is no missing segment in the real-time first-scale net ecosystem carbon flux continuous data, the real-time first-scale net ecosystem carbon flux continuous data is used as a subsequent incremental learning sample; According to the judgment result that there are missing segments in the real-time first-scale net ecosystem carbon flux continuous data, the multi-scale ecosystem carbon accounting model outputs ecosystem carbon accounting data based on the real-time second-scale net ecosystem carbon flux continuous data and corresponding environmental factors.

11. An electronic device, characterized in that: include: a processor coupled to the memory; The memory is used to store computer programs; The processor is configured to execute the computer program stored in the memory, so that the electronic device performs the method according to any one of claims 1 to 5.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed on a computer, the computer is caused to perform the method according to any one of claims 1 to 5.

13. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images