Carbon flux data collaborative fusion method based on multi-dimensional perception data

Abstract

The invention relates to the technical field of data fusion, in particular to a carbon flux data collaborative fusion method based on multi-dimensional perception data, which comprises the following steps of: acquiring multi-source heterogeneous carbon flux related data through a plurality of acquisition terminals, and performing space-time interpolation preprocessing to obtain a standardized multi-source data set; carrying out data fusion to construct a carbon flux dynamic evaluation model, dynamically simulating and outputting a carbon flux fusion index value, and then judging whether a data collaborative fusion process meets a standard or not based on a comparison result of the index value and a threshold value; when the first judgment does not meet the standard, adjusting the sampling frequency or the number of sampling points of the acquisition terminal, and optimizing the data quality from the data source; and when the carbon flux data does not meet the standard after adjustment, the number of blocks of the computing resources is further adjusted based on a model residual value or the window size of a space-time interpolation algorithm based on a space correlation intensity value, deep optimization is performed from the aspects of computing resources and algorithm parameters, and the precision and reliability of carbon flux data fusion are further remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of data fusion technology, and in particular to a collaborative fusion method for carbon flux data based on multidimensional sensing data. Background Technology

[0002] While existing flux tower monitoring schemes based on eddy coherence technology are considered the "gold standard" for carbon flux observation, they are essentially single-point observations. The measurement results only represent spatial integral values ​​within a limited range and cannot provide continuous spatial distribution information of carbon flux. Furthermore, due to their high construction and maintenance costs, flux towers cannot be deployed on a large scale and at high density, resulting in significant monitoring gaps at regional or global scales. In addition, model inversion schemes based on single-satellite remote sensing rely primarily on two-dimensional information such as vegetation indices (e.g., NDVI) extracted from optical images, which cannot fully characterize the impact of three-dimensional vegetation structure on carbon flux, leading to estimation bias and saturation in complex underlying surface areas. Regarding spatiotemporal resolution, existing remote sensing schemes also present contradictions: high spatial resolution satellites have long revisit periods, making it difficult to capture diurnal variations or quickly respond to sudden events; while high temporal resolution satellites have coarse spatial resolution, failing to meet the needs of refined management. Another biomass estimation scheme based on UAV photogrammetry has a core output of static carbon storage. To estimate dynamic carbon flux, it is necessary to rely on indirect calculations from multiple observations, which cannot achieve direct flux acquisition from a single monitoring and has poor real-time performance.

[0003] Although the above-mentioned technical solutions have observed carbon flux or related parameters from the perspectives of point measurement, area inversion, and fine structure scanning, the current technology lacks an effective collaborative fusion framework to organically integrate and collaboratively process multi-source heterogeneous data such as the "point" accuracy of flux towers, the "area" coverage of satellites, the "fine structure" of UAVs, and the "high-frequency time series" of near-ground sensing, so as to leverage their respective advantages and make up for the deficiencies of single data, thus restricting the further improvement of the accuracy and efficiency of carbon flux monitoring. Summary of the Invention

[0004] To address this issue, the present invention provides a carbon flux data collaborative fusion method based on multidimensional sensing data, which solves the problem that existing technologies cannot effectively integrate and collaboratively process multi-source heterogeneous data, resulting in low accuracy and efficiency of carbon flux monitoring.

[0005] To achieve the above objectives, this invention provides a method for collaborative fusion of carbon flux data based on multidimensional sensing data, comprising: Multi-source heterogeneous carbon flux-related datasets were acquired by data acquisition terminals deployed on different platforms. These datasets included fixed-point flux observation data, vegetation environmental parameter data, vegetation three-dimensional structure data, and high-frequency time-series environmental data. The carbon flux-related dataset is preprocessed using a spatiotemporal interpolation algorithm to obtain a standardized multi-source dataset; The multi-source datasets are fused to construct a dynamic carbon flux assessment model; Dynamic simulation is performed based on the carbon flux dynamic assessment model to output carbon flux fusion index value. When the data collaborative fusion process does not meet the standard based on the comparison result between the carbon flux fusion index value and the fusion index threshold, the sampling frequency or the number of sampling points of the acquisition terminal is adjusted. After adjusting the sampling frequency or the number of sampling points, a new carbon flux fusion index value is determined and recorded as the adjusted fusion index value, and the adjusted fusion quality parameters are determined, wherein the fusion quality parameters include the contribution of the model residuals and the spatial autocorrelation strength of the cross-validation error. If the data collaborative fusion process is still determined to be non-compliant based on the comparison result between the adjusted fusion index value and the fusion index threshold, the number of computing resource blocks used to process multi-source datasets is adjusted based on the model residual value, or the window size in the spatiotemporal interpolation algorithm is adjusted based on the spatial correlation strength value. The model residual value is determined based on the dispersion of the contribution, and the spatial correlation strength value is determined by quantifying the spatial autocorrelation strength.

[0006] Furthermore, the process of constructing the dynamic calculation model for carbon flux includes: Using the fixed-point flux observation data as a benchmark, the primary carbon flux products derived from the vegetation environmental parameter data and the vegetation three-dimensional structure data are scaled and their accuracy corrected. The temporal continuity of the fusion results is constrained and optimized using the high-frequency time-series environmental data to generate spatiotemporally continuous carbon flux fusion data. The carbon flux dynamic assessment model is constructed based on the carbon flux fusion data to synchronously assimilate multi-source heterogeneous data and output the carbon flux fusion index value to evaluate the fusion quality of multi-source heterogeneous data in the target area.

[0007] Furthermore, the comparison between the carbon flux fusion index value and the fusion index threshold to determine the data collaborative fusion process includes: If the carbon flux fusion index value is less than or equal to the fusion index threshold, the data collaborative fusion process is determined to be non-compliant with the standard, and the sampling frequency or the number of sampling points is increased based on the difference in the fusion index. The fusion index difference is the difference between the fusion index threshold and the carbon flux fusion index value.

[0008] Furthermore, based on the comparison result between the fusion index difference and the preset fusion index difference, the sampling frequency or the number of sampling points is increased, wherein the increase in the sampling frequency and the number of sampling points are both positively correlated with the fusion index difference.

[0009] Further, based on the re-determined sampling frequency or the number of sampling points, the adjusted fusion index value is recalculated, and the data collaborative fusion process is re-determined by comparing the adjusted fusion index value with the fusion index threshold. If the adjusted fusion index value is less than or equal to the fusion index threshold, the data collaborative fusion process is determined to still not meet the standard, and corresponding processing is determined. Based on the comparison results of the model residual value being greater than the model residual threshold, it is determined to increase the number of computing resource blocks based on the residual difference; Based on the comparison results of the model residual values ​​being less than or equal to the model residual threshold, the window for adjusting the spatiotemporal interpolation algorithm based on the spatial autocorrelation strength value is determined; The residual difference is the difference between the model residual value and the model residual threshold.

[0010] Furthermore, the number of computing resource blocks is increased based on the comparison result between the residual difference and the preset residual difference, wherein the increase in the number of blocks is positively correlated with the residual difference.

[0011] Furthermore, the process of adjusting the window of the spatiotemporal interpolation algorithm based on the spatial correlation intensity value includes: If the spatial correlation strength value is less than the spatial correlation strength threshold, the window size of the spatiotemporal interpolation algorithm is determined to be reduced based on the correlation strength difference; If the spatial correlation strength value is greater than or equal to the spatial correlation strength threshold, determine to increase the window size of the spatiotemporal interpolation algorithm based on the correlation strength offset value; Wherein, the correlation intensity difference is the difference between the spatial correlation intensity threshold and the spatial correlation intensity value, and the correlation intensity offset is the difference between the spatial correlation intensity value and the spatial correlation intensity threshold.

[0012] Furthermore, the window size of the spatiotemporal interpolation algorithm is reduced based on the comparison result between the correlation intensity difference and the preset correlation intensity difference, wherein the reduction in the algorithm window size is positively correlated with the correlation intensity difference.

[0013] Furthermore, based on the comparison result between the relevant intensity offset value and the preset spatial relevant intensity value offset, the window size of the spatiotemporal interpolation algorithm is increased, wherein the increase in the algorithm window size is positively correlated with the relevant intensity offset value.

[0014] Furthermore, the process of preprocessing the carbon flux-related dataset using a spatiotemporal interpolation algorithm also includes: Based on the spatial distribution density and data integrity of the fixed-point flux observation data, the data quality of the vegetation environmental parameter data and the vegetation three-dimensional structure data is evaluated to obtain the data quality score of each data source. Based on the comparison results between the data quality score and the quality score threshold, the target interpolation algorithm is dynamically selected from a variety of alternative spatiotemporal interpolation algorithms. Among them, the alternative spatiotemporal interpolation algorithms include at least a first target interpolation algorithm suitable for high data quality scenarios and a second target interpolation algorithm suitable for low data quality scenarios or scenarios with severe data loss.

[0015] Compared with existing technologies, the carbon flux data collaborative fusion method based on multi-dimensional sensing data of the present invention has the following advantages: it collects multi-source heterogeneous carbon flux-related data by deploying acquisition terminals on different platforms, and performs spatiotemporal interpolation preprocessing on the dataset to obtain a standardized multi-source dataset. Based on the preprocessing, data fusion is performed to construct a dynamic carbon flux assessment model. The model dynamically simulates and outputs carbon flux fusion index values. Then, the comparison results between the index values ​​and thresholds are used to determine whether the data collaborative fusion process meets the standards. If it is initially determined that it does not meet the standards, the sampling frequency or the number of sampling points of the acquisition terminals is adjusted to optimize the data quality from the data source. If it still does not meet the standards after adjustment, the number of computing resource blocks is further adjusted based on the model residual value, or the window size of the spatiotemporal interpolation algorithm is adjusted based on the spatial correlation strength value. Deep optimization is carried out from the level of computing resources and algorithm parameters, thereby significantly improving the accuracy and reliability of carbon flux data fusion.

[0016] Furthermore, when the fusion process is finally determined to meet the standards, the present invention generates accurate regional carbon flux assessment data based on the high-quality fusion results. This effectively solves the technical bottlenecks of insufficient spatial coverage, single model input dimension, and spatiotemporal resolution in monitoring from a single data source, thereby achieving accurate and dynamic monitoring of carbon flux in terrestrial ecosystems.

[0017] Furthermore, when the data fusion process is determined to be non-compliant with the standard based on the comparison result between the carbon flux fusion index value and the fusion index threshold, the present invention precisely increases the sampling frequency or the number of sampling points based on the comparison result between the fusion index difference and the preset fusion index difference, thereby reducing the time gap of data or reducing the monitoring blind spot, so as to improve the reliability of the carbon flux dynamic assessment model.

[0018] Furthermore, when the present invention determines that the data collaborative fusion process does not meet the standard based on the comparison result of the adjustment fusion index value and the fusion index threshold, it determines, based on the comparison result of the model residual value and the model residual threshold, that it is necessary to precisely increase the number of computing resource blocks based on the comparison result of the residual difference value and the preset residual difference value, thereby expanding the system's processing capacity during the data collaborative fusion process; or it determines that it is necessary to precisely reduce the window size of the spatiotemporal interpolation algorithm or reduce the window size based on the comparison result of the spatial correlation strength value and the spatial correlation strength threshold, thereby helping to preserve local features to prevent over-smoothing or helping to smooth the differences between different data sources, so as to improve the accuracy of the interpolation results in heterogeneous regions; with such settings, the adjustment fusion index value is ultimately increased so that the collaborative fusion process of carbon flux data meets the standard, thereby improving the accuracy and efficiency of carbon flux monitoring. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a system module used in an embodiment of the present invention to implement a carbon flux data collaborative fusion method based on multi-dimensional sensing data; Figure 2 This is a flowchart illustrating the carbon flux data collaborative fusion method based on multi-dimensional sensing data according to an embodiment of the present invention. Figure 3 This is a logic diagram illustrating the data collaborative fusion process determined by comparing the carbon flux fusion index value with the fusion index threshold in an embodiment of the present invention. Figure 4 This is a logic diagram of the collaborative fusion process of data based on the comparison result of adjusting the fusion index value and the fusion index threshold in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] In this implementation, the target area includes, but is not limited to, typical terrestrial ecosystems such as rainforests, forests, grasslands, wetlands, farmland, and urban green spaces. The goal is to accurately monitor and dynamically assess the carbon dioxide exchange flux between vegetation and the atmosphere in these areas. Furthermore, the goal is to improve the accuracy and reliability of the estimation of the spatiotemporal dynamics of regional carbon flux by synergistically fusing and analyzing fixed-point flux observation data, vegetation environmental parameter data, vegetation three-dimensional structure data, and high-frequency environmental time-series data obtained in the target area within a preset monitoring period.

[0024] Please see Figure 1 As shown, it is a schematic diagram of a system module used in an embodiment to implement a carbon flux data collaborative fusion method based on multi-dimensional sensing data.

[0025] The system includes a data acquisition module, a data processing module, a model building module, a dynamic simulation module, an analysis and evaluation module, and an optimization decision-making module. The data acquisition module acquires multi-source heterogeneous carbon flux-related datasets through acquisition terminals deployed on different platforms; The data processing module, which is connected to the data acquisition module, is used to preprocess the carbon flux-related dataset using a spatiotemporal interpolation algorithm to obtain a standardized multi-source dataset. The data processing module includes a central data processor, which performs spatiotemporal registration, missing value processing, and standardization on the carbon flux-related dataset to remove incomplete, garbled, and other abnormal data, so as to retain the effective data that can be used for data analysis. A model building module, which is connected to the data processing module, is used to fuse multi-source datasets to build a dynamic carbon flux assessment model. The dynamic simulation module, which is connected to the data processing module, is used to perform dynamic simulation based on the carbon flux dynamic assessment model to output the carbon flux fusion index value. The analysis and evaluation module, which is connected to the dynamic simulation module, is used to determine the data collaborative fusion process based on the comparison results of the carbon flux fusion index value and the fusion index threshold. The optimization decision module, connected to the dynamic simulation module, data acquisition module, and data processing module, is used to determine, when the data fusion process fails to meet the standards based on the carbon flux fusion index value, to adjust the sampling frequency or number of sampling points of the acquisition terminal, to redetermine a new carbon flux fusion index value and record it as the adjusted fusion index value, and to determine the adjusted fusion quality parameters, including the contribution of the model residuals and the spatial autocorrelation strength of the cross-validation error. If the data fusion process still fails to meet the standards based on the adjusted fusion index value, the module determines to adjust the number of computational resource blocks used to process multi-source datasets based on the model residual value, or to adjust the window size in the spatiotemporal interpolation algorithm based on the spatial correlation strength value. The model residual value is determined based on the dispersion of the contribution, and the spatial correlation strength value is determined by quantifying the spatial autocorrelation strength.

[0026] Among them, the contribution of the model residual is used to diagnose whether the fusion problem is caused by a local high error region. The calculation process is as follows: (1) Calculate the spatial unit residual: After rerunning the model, the target region is divided into regular spatiotemporal grids. For each grid unit with real observations (such as vorticity covariance tower data), its model residual is calculated; (2) Calculate the contribution: Calculate the contribution of the residual of each grid unit to the total error. In order to avoid the cancellation of positive and negative residuals, the square residual or absolute residual is usually used for calculation. The square residual contribution is preferred; (3) Quantify the degree of dispersion: In order to measure the degree of concentration of contribution in space, the system calculates its Gini coefficient as the model residual value M. It should be noted that the range of the Gini coefficient is [0, 1]. The closer the Gini coefficient is to 0, the more uniform the contribution is and the widely distributed the error is. The closer the Gini coefficient is to 1, the more concentrated the contribution is, that is, most of the model error comes from a very small number of spatial units.

[0027] The spatial autocorrelation intensity of cross-validation error is used to diagnose whether the fusion problem stems from poor global continuity of the data space. The calculation process is as follows: (1) Generate cross-validation error field: When performing spatiotemporal interpolation in the preprocessing stage, the "leave-one-out method" or "k-fold cross-validation" is used; (2) Calculate spatial autocorrelation intensity: The system uses the global Moran index to quantify the spatial autocorrelation intensity of the error field and uses it as the spatial autocorrelation intensity value N. It should be noted that the Moran index generally ranges from -1 to 1.

[0028] Please see Figure 2 The diagram shown is a flowchart illustrating the carbon flux data collaborative fusion method based on multi-dimensional sensing data according to an embodiment of the present invention. The process includes at least the following steps: S1: Obtain multi-source heterogeneous carbon flux-related datasets through acquisition terminals deployed on different platforms. The related datasets include fixed-point flux observation data, vegetation environmental parameter data, vegetation three-dimensional structure data, and environmental high-frequency time series data. S2: Preprocess the carbon flux-related datasets using a spatiotemporal interpolation algorithm to obtain standardized multi-source datasets; S3: Fusion of multi-source datasets to construct a dynamic carbon flux assessment model; S4: Dynamic simulation based on carbon flux dynamic assessment model to output carbon flux fusion index value; S5: If the data fusion process does not meet the standard based on the comparison result between the carbon flux fusion index value and the fusion index threshold, determine to adjust the sampling frequency or the number of sampling points of the acquisition terminal. S6: After completing the adjustment of the sampling frequency or the number of sampling points, a new carbon flux fusion index value is determined and recorded as the adjusted fusion index value; S7: and determine the adjusted fusion quality parameters, which include the contribution of the model residuals and the spatial autocorrelation strength of the cross-validation error; S8: If the data collaborative fusion process is still determined to be non-compliant with the standard based on the comparison result between the adjusted fusion index value and the fusion index threshold, the number of computing resource blocks used to process multi-source datasets is adjusted based on the model residual value, or the window size in the spatiotemporal interpolation algorithm is adjusted based on the spatial correlation strength value. The model residual value is determined based on the degree of dispersion of contribution, and the spatial correlation strength value is determined by quantifying the spatial autocorrelation strength.

[0029] Among them, 1. The implementation process of dynamic simulation: Dynamic simulation is to let the constructed carbon flux dynamic assessment model be deduced in the spatiotemporal dimension, thereby generating a complete spatiotemporally continuous carbon flux dataset. The specific steps are: (1) Initialize the model state: On all spatiotemporal grids in the target region, initialize the driving variables and state variables of the model. For example, using the preprocessed multi-source dataset, the initial vegetation environment parameters (such as LAI), high-frequency environmental time series data (such as temperature and radiation), and primary carbon flux products retrieved from remote sensing are input for each grid point; (2) Set the simulation spatiotemporal range: Define the start time, end time, and time step of the simulation (e.g., 1 hour or 1 day), and clarify the spatial range of the simulation (i.e., all grid units); (3) Iterative simulation (loop execution): The model advances step by step according to the time step. For each time step, the following are executed on each spatial grid: (a) data-driven; (b) model operation; (c) assimilation of observation data; (d) state update; (e) output storage; (4) Generate fusion product: When the time series iteration is completed, the system generates a set of "carbon flux fusion data products" that cover the entire spatiotemporal range and have been collaboratively corrected and optimized by multi-source data.

[0030] 2. Calculation process of carbon flux fusion index value: The carbon flux fusion index value is used to quantitatively evaluate the reliability and fusion effect of the fusion data product generated by the above dynamic simulation. The specific steps are as follows: (1) Calculation formula: R=w×Wc+(1-w)×Wa, where w is a weight coefficient between 0 and 1, Wc is the correlation coefficient used to measure the temporal change of fusion data and independent observation data, and Wa is the consistency coefficient used to measure the closeness between model prediction value and observation value; (2) Calculation example process: If there are two independent verification sites (site A and site B), after completing a month of dynamic simulation; for site A, Wc(A)=0.92, Wa(A)=0.88 are calculated; for site B, Wc(B)=0.89, Wa(B)=0.85 are calculated; take the average of the two sites, Wc=0.905, Wa=0.865, take the weight w=0.6, and finally get R=0.889.

[0031] In this embodiment, the data acquisition terminals mainly include the following types: 1. Ground-based observation platform terminals, including eddy covariance systems for acquiring fixed-point flux observation data and meteorological and environmental sensors for acquiring vegetation environmental parameter data. 2. Near-Earth / Airborne remote sensing platform terminals, including lidar for acquiring three-dimensional vegetation structure data and multispectral imagers for assisting in acquiring vegetation environmental parameter data. 3. Satellite remote sensing platform terminals, including Earth observation satellites and lidar satellites. 4. Other auxiliary platform terminals, including unmanned aerial vehicles (UAVs).

[0032] In this embodiment, the process of obtaining multi-source heterogeneous carbon flux-related datasets includes: 1. Deployment and Initialization: Deploy the various data acquisition terminals (eddy covariance system, weather station, UAV, satellite, etc.) to the designated observation locations or platforms; calibrate and set parameters for the equipment to ensure they are in normal working order. 2. Automated Data Acquisition: Each acquisition terminal automatically and continuously collects raw data according to its preset sampling frequency (e.g., 10Hz for eddy covariance system, every 30 minutes for weather station, and every few days or weeks for satellite). 3. Data Transmission and Aggregation: The collected raw data is transmitted from the distributed terminals to a central data processor in one or more data centers via wired networks, wireless networks (e.g., 4G / 5G), satellite links, etc. 4. Raw Dataset Formation: All transmitted data is initially aggregated and stored in the data center to form the aforementioned multi-source heterogeneous carbon flux-related dataset.

[0033] In this embodiment, the spatiotemporal registration of the carbon flux-related dataset includes the registration of spatial and temporal references; for missing values ​​in the registered grid, different spatiotemporal interpolation algorithms are used to fill them; after the missing value filling is completed, all variables are standardized to eliminate the influence of different physical quantities and dimensions.

[0034] Specifically, the process of constructing a dynamic carbon flux calculation model includes: Using fixed-point flux observation data as a benchmark, the primary carbon flux products derived from vegetation environmental parameter data and vegetation three-dimensional structure data are scaled and accuracy corrected. High-frequency environmental time-series data are used to constrain and optimize the temporal continuity of the fusion results to generate spatiotemporally continuous carbon flux fusion data. Based on the carbon flux fusion data, a dynamic carbon flux assessment model is constructed. The model is used to synchronously assimilate multi-source heterogeneous data and outputs carbon flux fusion index values ​​to evaluate the fusion quality of multi-source heterogeneous data in the target area.

[0035] In this embodiment: 1. Establishing a baseline: Fixed-point flux observation data is used as a precise baseline. 2. Relationship modeling: At locations with ground stations, a machine learning model (such as a random forest or gradient boosting tree) is established to learn the relationship between ground-measured carbon flux, vegetation environmental parameter data, and vegetation 3D structure data. This model, through training, infers a carbon flux in a specific area that more closely approximates the actual situation based on the vegetation environmental parameter data and vegetation 3D structure data. 3. Scale expansion and accuracy correction: The trained machine learning model is applied to every spatiotemporal grid cell in the entire target area. For each cell without a ground station, the corresponding vegetation environmental parameter data, vegetation 3D structure data, and high-frequency time-series environmental data are input. The model calculates a corrected carbon flux estimate, "transferring" the accuracy of the ground points to the entire area, generating a preliminary, spatially continuous carbon flux fusion data layer. 4. Constraints and Optimization: High-frequency environmental time-series data (such as temperature and radiation) are used to drive a simplified ecological process model (such as a light energy utilization model). This model can simulate a smooth and physically reasonable time-series trajectory of carbon flux under normal weather conditions, which serves as prior knowledge. The obtained preliminary carbon flux fusion data is then fused with the time-series trajectory simulated by the ecological process model. The final output analysis value is the most reliable carbon flux fusion data after time-series optimization. 5. Construct a dynamic carbon flux assessment model: Integrate the aforementioned “benchmark correction-scale expansion-assimilation optimization” process to construct a hybrid model framework of data-driven and process-driven approaches, thereby obtaining a dynamic carbon flux assessment model; the specific steps include: (1) Constructing a model-driven engine: multi-source data input and feature engineering, the model establishes a unified spatiotemporal grid framework as the benchmark for all data fusion; (2) Constructing a model core processor: dual-path fusion calculation, the model contains two parallel calculation paths, the first calculation path is a data-driven path (machine learning corrector), and the second calculation path is a two-dimensional process-driven path (ecological process simulator); (3) Constructing a model fusion processor: use a data assimilation algorithm to perform dynamic data assimilation, in the assimilation loop, the fusion processor continuously compares the outputs of the two paths, when the difference is significant, the assimilation algorithm will dynamically adjust the key state variables or parameters of the ecological process simulator; (4) Constructing a model self-evaluator: by calculating the correlation coefficient Wc and the consistency coefficient Wa, the carbon flux fusion index value R is finally obtained.

[0036] Please see Figure 3 As shown, it is a logic determination diagram of the data collaborative fusion process based on the comparison result of carbon flux fusion index value and fusion index threshold in an embodiment of the present invention.

[0037] Specifically, a fusion index threshold R0 is set and compared with the carbon flux fusion index value R to determine the data collaborative fusion process. The fusion index threshold R0 is determined based on historical data analysis and statistical analysis methods. For example, R0 = 0.85 is set. The process of comparing R and R0 is as follows: If R is greater than R0, it indicates that the current carbon flux data fusion index for multi-source heterogeneous data meets the preset standard. The current data fusion process can be deemed compliant, and a regional carbon flux monitoring report and assessment conclusion can be generated based on the latest carbon flux fusion data and transmitted to the user terminal. If R is less than or equal to R0, it indicates that the current carbon flux data fusion index does not meet the preset standard. The current data fusion process is deemed non-compliant, and adjustment instructions need to be generated to optimize the fusion process. This can be achieved by calculating the difference between the fusion index threshold R0 and the carbon flux fusion index value R to obtain the fusion index difference F. Based on the fusion index difference F, the sampling frequency of the acquisition terminal or the corresponding number of sampling points can be increased.

[0038] Specifically, the sampling frequency or the number of sampling points is increased based on the comparison results between the fusion index difference F and the preset fusion index difference F0. The increase in both the sampling frequency and the number of sampling points is positively correlated with the fusion index difference F. Increasing the sampling frequency (e.g., from once every 30 minutes to once every 5 minutes) can more accurately capture these diurnal variation characteristics, which is especially crucial for high-frequency environmental time-series data. This avoids missing key peaks or troughs due to insufficient sampling and reduces the time gaps in the data, providing a more continuous and smoother data foundation for subsequent spatiotemporal interpolation and model simulation. Increasing the number of sampling points (e.g., deploying more ground sensors or increasing UAV aerial survey routes) can effectively reduce monitoring blind spots, making it more suitable for areas with high spatial heterogeneity (such as fragmented forests and mountainous areas with complex terrain), thereby improving the reliability of the carbon flux dynamic assessment model.

[0039] In this embodiment, taking the adjustment of the sampling frequency of the meteorological station in the acquisition terminal as an example, a preset fusion index difference value F0 corresponding to the fusion index difference value F is set, and compared with the fusion index difference value F to determine the increase of the sampling frequency. Since the fusion index difference value F is the difference between the fusion index threshold R0 and the carbon flux fusion index value R, the larger the fusion index difference value F is, the smaller the corresponding carbon flux fusion index value R is, which indicates that the fusion quality problem is more serious, and the magnitude of the adjustment measures (such as increasing the sampling frequency or the number of sampling points) taken by the subsequent optimization decision module is also greater. Therefore, the increase of the sampling frequency and the number of sampling points are positively correlated with the fusion index difference value F.

[0040] To more accurately determine the magnitude of the increase in sampling frequency, the preset fusion index difference F0 can be divided into a first preset fusion index difference F1 and a second preset fusion index difference F2. For example, F1=0.1 and F2=0.2. The process of comparing F with F1 and F2 is as follows: If F is less than or equal to F1, a first sampling frequency adjustment command is generated, which increases the meteorological station's sampling frequency by 15% based on the original frequency. Specifically, if the original sampling frequency is once every 30 minutes, the increased sampling frequency will be once every 25.5 minutes. If F is greater than F1 and less than or equal to F2, a second sampling frequency adjustment command is generated, which increases the meteorological station's sampling frequency by 30% based on the original frequency. If F is greater than F2, a third sampling frequency adjustment command is generated, which increases the meteorological station's sampling frequency by 60% based on the original frequency.

[0041] Understandably, the increase in sampling frequency can also be set to other values ​​that meet the requirements. For example, when F is greater than F2, the sampling frequency can be increased by 65% ​​based on the original sampling frequency. It should be noted that the principle of increasing the number of sampling points is the same as that of increasing the sampling frequency, and the specific increase adjustment process will not be elaborated here.

[0042] Please see Figure 4 As shown, it is a logic decision diagram of the collaborative fusion process of the present invention based on the comparison result of adjusting the fusion index value and the fusion index threshold.

[0043] Specifically, after adjusting the sampling frequency or the number of sampling points, the adjusted fusion index value G is recalculated, a fusion index threshold R0 is set, and compared with the adjusted fusion index value G. The comparison process is as follows: If G is greater than R0, the data fusion process after adjusting the sampling frequency or the number of sampling points is deemed to meet the standard. If G is still less than or equal to R0, the data fusion process is deemed to still not meet the standard. In this case, adjusted fusion quality parameters can be obtained, including the contribution of model residuals and the spatial autocorrelation strength of cross-validation errors. Specifically, the contribution of model residuals in different vegetation types or terrain complexities in the multi-source dataset is obtained from the fusion analysis module. Then, the contribution is discretized to obtain the model residual value M. A model residual threshold M0 is set and compared with the model residual value M. Based on the comparison result, the number of computational resource blocks used to process the multi-source dataset is adjusted. The spatial autocorrelation strength of the cross-validation error generated during the operation of the spatiotemporal interpolation algorithm is obtained from the preprocessing module. Then, the spatial autocorrelation strength value N is obtained through quantification. A spatial autocorrelation strength threshold N0 is set and compared with the spatial autocorrelation strength value N. Based on the comparison result, the window size in the spatiotemporal interpolation algorithm is adjusted.

[0044] When discretizing the contribution of model residuals, it is preferable to use the coefficient of variation or the Gini coefficient for characterization. The former can eliminate the influence of dimensions and robustly reflect the relative dispersion, while the latter can intuitively measure the inequality of error distribution, thereby obtaining the model residual value M.

[0045] First, after performing spatiotemporal interpolation, the cross-validation error at each spatial location is retained to form an error spatial distribution field. Then, the spatial structure of the error field is quantified using methods such as the Moran index or the spatial autocorrelation statistic of the semivariogram to obtain a scalar index characterizing the spatial autocorrelation strength, which is denoted as the spatial correlation strength value N.

[0046] In this embodiment, based on the model residual dataset generated from multiple historical fusion experiments, a specific statistical quantile is calculated, for example, the 90th percentile is used as the model residual threshold M0. The process of comparing M and M0 is as follows: If M is greater than M0, it indicates that the model fits very poorly in some regions, with "outliers" or "blind spots" in understanding. Therefore, the reason why the data fusion process still does not meet the standard is due to the existence of local high-error regions. Concentrating computing power to solve these high-error regions results in insufficient computational resources. In this case, the number of computational resource blocks can be increased non-uniformly based on the residual difference E. By increasing the number of computational resource blocks, computational power can be concentrated to solve these high-error regions. Here, the residual difference E is the difference between the model residual value M and the model residual threshold M0.

[0047] If M is less than or equal to M0, it indicates that the model does not have relatively large local errors. In this case, it can be determined that the spatiotemporal interpolation algorithm is not well adapted to the current data features. Therefore, the adjustment method of the window size of the spatiotemporal interpolation algorithm can be determined based on the spatial correlation strength value N.

[0048] Specifically, the number of computational resource blocks is increased based on the comparison result between the residual difference E and the preset residual difference E0. The increase in the number of blocks is positively correlated with the residual difference E. As the model residual value increases, the system can dynamically and correspondingly increase the number of computational resource blocks, thereby expanding the system's processing capacity during data collaborative fusion. At this time, whether the monitoring scope expands or new data sources are connected, the system can effectively cope with the increasing model residual value.

[0049] In one specific embodiment, a preset residual difference value E0 is set and compared with the residual difference value E. The residual difference value E is the difference between the model residual value M and the model residual threshold M0. When the residual difference value E is larger, the corresponding model residual value M is larger, which indicates that the severity of the current problem exceeds the baseline by a larger amount. Therefore, the increase in the number of computing resource blocks is positively correlated with the residual difference value E.

[0050] To more accurately determine the increase in the number of computational resource blocks, the preset residual difference E0 can be divided into a first preset residual difference E1 and a second preset residual difference E2. For example, E1 is set to the 2nd quantile and E2 to the 4th quantile. The comparison process between E and E1 and E2 is as follows: If E is less than or equal to E1, a first block quantity adjustment instruction is generated. Based on this instruction, the central data processor is controlled to increase the number of blocks by one. For example, if the original number of blocks is four, the increased number of blocks will be five. If E is greater than E1 and less than or equal to E2, a second block quantity adjustment instruction is generated. Based on this instruction, the central data processor is controlled to increase the number of blocks by two. If E is greater than E2, a third block quantity adjustment instruction is generated. Based on this instruction, the central data processor is controlled to increase the number of blocks by four.

[0051] Understandably, the number of additional blocks can also be set to other numbers that meet the requirements. For example, if E is less than or equal to E1, two more blocks can be added to the original number of blocks.

[0052] Specifically, when using the Moran index to characterize the spatial correlation strength value N, the spatial correlation strength threshold N0 can be set to 0.4 for example. The comparison process between the spatial correlation strength value N and the spatial correlation strength threshold N0 is as follows: If N is less than N0, it indicates that the error is spatially distributed randomly and discretely, lacking a continuous pattern. In this case, reducing the window size of the spatiotemporal interpolation algorithm can avoid introducing irrelevant noise from distant locations, better preserve local details, and prevent the interpolation process from becoming overly smooth. The window size of the spatiotemporal interpolation algorithm is determined based on the correlation strength difference Q.

[0053] If N is greater than or equal to N0, it indicates that the error exhibits significant spatial clustering and continuity, with strong spatial dependence. In this case, increasing the window size of the spatiotemporal interpolation algorithm is reasonable because a larger window can more effectively utilize this spatial correlation, obtaining more information from neighboring regions to smooth the error and improve the stability of the interpolation. The window size of the spatiotemporal interpolation algorithm is determined based on the correlation strength offset value W. Specifically, when N equals N0, the window size of the spatiotemporal interpolation algorithm is also increased to avoid frequent oscillations between window enlargement and reduction states caused by small numerical fluctuations near the threshold, thereby enhancing the stability of the system.

[0054] Wherein, the correlation intensity difference Q is the difference between the spatial correlation intensity threshold N0 and the spatial correlation intensity value N, and the correlation intensity offset W is the difference between the spatial correlation intensity value N and the spatial correlation intensity threshold N0.

[0055] In one specific embodiment, a preset correlation strength difference Q0 is set and compared with the correlation strength difference Q. Based on the comparison result, the reduction in the window size of the spatiotemporal interpolation algorithm is determined. Here, the correlation strength difference Q is the difference between the spatial correlation strength threshold N0 and the spatial correlation strength value N. A larger correlation strength difference Q corresponds to a smaller spatial correlation strength value N, indicating weak spatial autocorrelation of the current data. This suggests that the cross-validation error exhibits a random and discrete spatial distribution, lacking an effective spatial dependency pattern. Using a large window for interpolation introduces a large amount of irrelevant noise, leading to severe distortion of the interpolation results. Therefore, the window size of the spatiotemporal interpolation algorithm should be reduced. This limits the search range of the interpolation, making it more dependent on the nearest few data points, thereby better capturing and preserving local variations, preventing over-smoothing, and improving the accuracy of the interpolation results in heterogeneous regions. Therefore, the reduction in the algorithm window size is positively correlated with the correlation strength difference Q.

[0056] To more accurately determine the reduction in window size, the preset correlation intensity difference Q0 can be divided into a first preset correlation intensity difference Q1 and a second preset correlation intensity difference Q2. For example, Q1 = 0.05 and Q2 = 0.08. The process of comparing Q with Q1 and Q2 is as follows: If Q is less than or equal to Q1, a first window adjustment instruction is generated, reducing the original spatial window size by 2×2 and the original time window size by 2 units. Specifically, if the original spatial window size was 7×7 and the original time window size was 5 units of time, the adjusted spatial window size will be 5×5 and the time window size will be 3 units of time. If Q is greater than Q1 and less than or equal to Q2, a second window adjustment instruction is generated, reducing the original spatial window size by 3×3 and the original time window size by 3 units. If Q is greater than Q2, a third window adjustment instruction is generated, reducing the original spatial window size by 4×4 and the original time window size by 4 units.

[0057] In one specific embodiment, a preset correlation intensity offset value W0 is set and compared with the correlation intensity offset value W. Based on the comparison result, the increase in the window size of the spatiotemporal interpolation algorithm is determined. Here, the correlation intensity offset value W is the difference between the spatial correlation intensity value N and the spatial correlation intensity threshold N0. A larger correlation intensity offset value W corresponds to a larger spatial correlation intensity value N, indicating that the spatial autocorrelation of the current data is very strong, and the cross-validation error exhibits high spatial clustering and continuity. Stable, large-scale spatial patterns exist in the data, and a small window cannot effectively utilize this long-range correlation. Therefore, the window size of the spatiotemporal interpolation algorithm should be increased; this allows the interpolation algorithm to draw information from a broader region, fully utilizing the strong spatial dependence to smooth out local random fluctuations and generate a more stable, continuous interpolation surface that conforms to the actual spatial pattern. Therefore, the increase in the algorithm window size is positively correlated with the correlation intensity offset value W.

[0058] To more accurately determine the magnitude of the window size increase, the preset spatial correlation intensity deviation value W0 can be divided into a first preset correlation intensity deviation value W1 and a second preset correlation intensity deviation value W2. For example, W1=2 and W2=3. The process of comparing W with W1 and W2 is as follows: If W is less than or equal to W1, a fourth window adjustment instruction is generated, increasing the spatial window size by 2×2 and the time window size by 2 units of time. Specifically, if the original spatial window size was 7×7 and the original time window size was 5 units of time, the adjusted spatial window size is reduced to 9×9 and the time window size to 7 units of time. If W is greater than W1 and less than or equal to W2, a fifth window adjustment instruction is generated, increasing the spatial window size by 4×4 and the time window size by 3 units of time. If W is greater than W2, a sixth window adjustment instruction is generated, increasing the spatial window size by 6×6 and the time window size by 4 units of time.

[0059] If the data fusion process still does not meet the standard when re-evaluating the fusion index value G, and the model residual value M is a comparison result that is less than or equal to the model residual threshold M0, the window size of the spatiotemporal interpolation algorithm can be adjusted by decreasing or increasing the spatial correlation strength value N, thereby increasing the fusion index value G so that it can be greater than the fusion index threshold R0, thus making the carbon flux data fusion process meet the standard.

[0060] Specifically, the preprocessing of carbon flux-related datasets using spatiotemporal interpolation algorithms also includes: Based on the spatial distribution density and data integrity of fixed-point flux observation data, data quality assessment is performed on vegetation environmental parameter data and vegetation three-dimensional structure data to obtain data quality scores for each data source. Based on the comparison results of data quality scores and quality score thresholds, a target interpolation algorithm is dynamically selected from a variety of candidate spatiotemporal interpolation algorithms. Among them, the candidate spatiotemporal interpolation algorithms include at least a first target interpolation algorithm suitable for high data quality scenarios and a second target interpolation algorithm suitable for low data quality scenarios or scenarios with severe data missing.

[0061] In assessing spatial distribution density, the ratio of the actual average nearest neighbor distance between all data points to the theoretical average nearest neighbor distance under a completely random distribution model is calculated. The target area is divided into regular grids (e.g., 1km x 1km), and the percentage of grids containing at least one data point is counted. The higher the percentage, the better the spatial coverage. This percentage is recorded as the spatial coverage. In assessing data integrity, the missing rate is obtained based on the calculation process of (fusion data volume - effective fusion data volume) / fusion data volume. The lower the missing rate, the higher the data integrity. The effective fusion data volume is the number of data points that have passed preprocessing, as counted during preprocessing.

[0062] Indicators with different dimensions are uniformly mapped to the [0,1] interval; generally, indicators with larger values ​​(such as spatial coverage) can be directly normalized; indicators with smaller values ​​(such as missing rate) need to be subtracted from the missing rate value by 1; the importance of each quality dimension is different under different data sources and different application scenarios, so different weight coefficients are assigned to different parameters, and the data quality score is obtained by using a weighted summation method.

[0063] In this embodiment, the first target interpolation algorithm can be the Kriging interpolation algorithm, and the second target interpolation algorithm can be the inverse distance weighted interpolation algorithm. The effectiveness of the selected target interpolation algorithm can also be verified based on the adjustment fusion index value output by the carbon flux dynamic assessment model, thereby optimizing the algorithm selection strategy in subsequent preprocessing. For example, if the data quality score is high (>0.8) and the adjustment fusion index value output by the carbon flux dynamic assessment model is greater than the fusion index threshold when the first target interpolation algorithm is selected, then the first target interpolation algorithm is selected as the spatiotemporal interpolation algorithm for preprocessing the subsequent carbon flux-related datasets; if the data quality score is low (<0.6) and the output adjustment fusion index value is less than or equal to the fusion index threshold when the first target interpolation algorithm is selected, then the current selection of the first target interpolation algorithm is deemed unsuitable, and therefore the second target interpolation algorithm can be preferentially selected; if the second target interpolation algorithm is still deemed unsuitable, a new algorithm can be directly replaced. The window of the spatiotemporal interpolation algorithm can be adjusted based on the spatial correlation strength value N of the multi-source heterogeneous data.

[0064] Any technologies not mentioned in the above embodiments are applicable to existing technologies.

[0065] It is understood that no specific limitation is made to any preset parameter or critical parameter in the embodiments of the present invention, and the above values ​​are not limited thereto. Those skilled in the art can make corresponding adjustments to the preset parameters or critical parameters according to actual needs, analysis of historical data, or equipment usage.

[0066] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for collaborative fusion of carbon flux data based on multidimensional sensing data, characterized in that, include: Multi-source heterogeneous carbon flux-related datasets were acquired by data acquisition terminals deployed on different platforms. These datasets included fixed-point flux observation data, vegetation environmental parameter data, vegetation three-dimensional structure data, and high-frequency time-series environmental data. The carbon flux-related dataset is preprocessed using a spatiotemporal interpolation algorithm to obtain a standardized multi-source dataset; The multi-source datasets are fused to construct a dynamic carbon flux assessment model; Dynamic simulation is performed based on the carbon flux dynamic assessment model to output carbon flux fusion index value. When the data collaborative fusion process does not meet the standard based on the comparison result between the carbon flux fusion index value and the fusion index threshold, the sampling frequency or the number of sampling points of the acquisition terminal is adjusted. After adjusting the sampling frequency or the number of sampling points, a new carbon flux fusion index value is determined and recorded as the adjusted fusion index value, and the adjusted fusion quality parameters are determined, wherein the fusion quality parameters include the contribution of the model residuals and the spatial autocorrelation strength of the cross-validation error. If the data collaborative fusion process is still determined to be non-compliant based on the comparison result between the adjusted fusion index value and the fusion index threshold, the number of computing resource blocks used to process multi-source datasets is adjusted based on the model residual value, or the window size in the spatiotemporal interpolation algorithm is adjusted based on the spatial correlation strength value. The model residual value is determined based on the dispersion of the contribution, and the spatial correlation strength value is determined by quantifying the spatial autocorrelation strength.

2. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 1, characterized in that, The process of constructing the dynamic carbon flux calculation model includes: Using the fixed-point flux observation data as a benchmark, the primary carbon flux products derived from the vegetation environmental parameter data and the vegetation three-dimensional structure data are scaled and their accuracy corrected. The temporal continuity of the fusion results is constrained and optimized using the high-frequency time-series environmental data to generate spatiotemporally continuous carbon flux fusion data. The carbon flux dynamic assessment model is constructed based on the carbon flux fusion data to synchronously assimilate multi-source heterogeneous data and output the carbon flux fusion index value to evaluate the fusion quality of multi-source heterogeneous data in the target area.

3. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 1, characterized in that, The data collaborative fusion process is determined by comparing the carbon flux fusion index value with the fusion index threshold. If the carbon flux fusion index value is less than or equal to the fusion index threshold, the data collaborative fusion process is determined to be non-compliant with the standard, and the sampling frequency or the number of sampling points is increased based on the difference in the fusion index. The fusion index difference is the difference between the fusion index threshold and the carbon flux fusion index value.

4. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 3, characterized in that, Based on the comparison result between the fusion index difference and the preset fusion index difference, the sampling frequency or the number of sampling points is increased, wherein the increase in the sampling frequency and the number of sampling points are both positively correlated with the fusion index difference.

5. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 4, characterized in that, Based on the newly determined sampling frequency or the number of sampling points, the adjusted fusion index value is recalculated, and the data collaborative fusion process is re-determined by comparing the adjusted fusion index value with the fusion index threshold. If the adjusted fusion index value is less than or equal to the fusion index threshold, the data collaborative fusion process is determined to still not meet the standard, and corresponding processing is determined. Based on the comparison results of the model residual value being greater than the model residual threshold, it is determined to increase the number of computing resource blocks based on the residual difference; Based on the comparison results of the model residual values ​​being less than or equal to the model residual threshold, the window for adjusting the spatiotemporal interpolation algorithm based on the spatial autocorrelation strength value is determined; The residual difference is the difference between the model residual value and the model residual threshold.

6. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 5, characterized in that, The number of computing resource blocks is increased based on the comparison result between the residual difference and the preset residual difference, wherein the increase in the number of blocks is positively correlated with the residual difference.

7. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 5, characterized in that, The process of adjusting the window of the spatiotemporal interpolation algorithm based on the spatial correlation intensity value includes: If the spatial correlation strength value is less than the spatial correlation strength threshold, the window size of the spatiotemporal interpolation algorithm is determined to be reduced based on the correlation strength difference; If the spatial correlation strength value is greater than or equal to the spatial correlation strength threshold, determine to increase the window size of the spatiotemporal interpolation algorithm based on the correlation strength offset value; Wherein, the correlation intensity difference is the difference between the spatial correlation intensity threshold and the spatial correlation intensity value, and the correlation intensity offset is the difference between the spatial correlation intensity value and the spatial correlation intensity threshold.

8. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 7, characterized in that, The window size of the spatiotemporal interpolation algorithm is reduced based on the comparison result between the correlation intensity difference and the preset correlation intensity difference, wherein the reduction in the algorithm window size is positively correlated with the correlation intensity difference.

9. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 7, characterized in that, The window size of the spatiotemporal interpolation algorithm is increased based on the comparison result between the relevant intensity offset value and the preset spatial relevant intensity offset value, wherein the increase in the algorithm window size is positively correlated with the relevant intensity offset value.

10. The carbon flux data collaborative fusion method based on multi-dimensional sensing data according to claim 1, characterized in that, The process of preprocessing the carbon flux-related dataset using a spatiotemporal interpolation algorithm also includes: Based on the spatial distribution density and data integrity of the fixed-point flux observation data, the data quality of the vegetation environmental parameter data and the vegetation three-dimensional structure data is evaluated to obtain the data quality score of each data source. Based on the comparison results between the data quality score and the quality score threshold, the target interpolation algorithm is dynamically selected from a variety of alternative spatiotemporal interpolation algorithms. Among them, the alternative spatiotemporal interpolation algorithms include at least a first target interpolation algorithm suitable for high data quality scenarios and a second target interpolation algorithm suitable for low data quality scenarios or scenarios with severe data loss.

Images