Atmospheric flux estimation method based on multi-scale dynamic correlation

Abstract

The invention relates to the technical field of atmospheric flux calculation, in particular to an atmospheric flux estimation method based on multi-scale dynamic correlation. In an existing multi-scale atmospheric flux calculation technology, multi-scale data integration depends on manual interpolation or step-by-step calculation, space-time correlation of different-scale atmospheric processes cannot be effectively captured, and remarkable scale connection errors are likely to be generated. And the traditional iterative calculation needs to repeatedly check parameters, so that the aging requirement is difficult to meet in the scenes of atmospheric pollution emergency monitoring, real-time carbon flux evaluation and the like. According to the method, space-time alignment and noise suppression are performed on multi-source and multi-scale observation data, and then a cross-scale dynamic association mechanism is constructed by adopting an AI model to replace a traditional artificial empirical formula. The method can effectively reduce the scale cohesion error, solves the problem that the precision and speed are difficult to consider in the traditional technology, and is suitable for the scenes of atmospheric pollution emergency response, regional carbon cycle dynamic monitoring and the like.

Description

Technical Field

[0001] This invention relates to the field of atmospheric flux calculation technology, and more specifically, to an atmospheric flux estimation method based on multi-scale dynamic correlation. Background Technology

[0002] Atmospheric flux estimation, as a core supporting technology in the field of atmospheric flux calculation, is directly related to the scientific nature of decision-making in scenarios such as emergency response to air pollution and dynamic monitoring of regional carbon cycle. Its estimation accuracy and response time are key indicators for measuring the practicality of the technology.

[0003] Current multi-scale atmospheric flux calculation techniques suffer from two major pain points: First, multi-scale data integration relies on manual interpolation or step-by-step calculations, failing to effectively capture the spatiotemporal correlations of atmospheric processes at different scales. This leads to significant scale-connection errors; for example, satellite remote sensing, ground observation, and numerical simulation data are prone to inconsistencies due to coarse spatiotemporal matching. Second, traditional iterative calculations require repeated parameter verification, which is insufficient for rapid response in time-sensitive scenarios such as emergency monitoring of air pollution and real-time carbon flux assessment. Furthermore, existing data fusion lacks a physically driven two-way verification mechanism, failing to dynamically quantify the representativeness and quality of data sources; cross-scale correlation construction relies on manual empirical formulas, lacking dynamically optimized structural support, further limiting accuracy improvement.

[0004] To address this, the present invention proposes an atmospheric flux estimation method based on multi-scale dynamic correlation, thereby overcoming the bottlenecks of existing technologies. Summary of the Invention

[0005] In view of the shortcomings of existing technologies, the purpose of this invention is to provide an atmospheric flux estimation method based on multi-scale dynamic correlation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The atmospheric flux estimation method based on multi-scale dynamic correlation includes the following steps: Step 1: Spatiotemporal alignment and noise suppression of multi-scale observation data.

[0007] S11. Data Acquisition: Acquire three types of multi-scale data sources: satellite remote sensing data, ground observation data, and numerical simulation data; S12. Spatiotemporal alignment: Based on the ground flux tower observation timestamp, perform temporal and spatial alignment on satellite remote sensing data and numerical simulation data; S13. Noise Suppression: Differentiated denoising strategies are adopted for satellite remote sensing data, ground observation data, and numerical simulation data, taking into account their different noise characteristics. S14. Physically driven multi-source data fusion: Based on bidirectional traceability, the adaptive confidence weighted fusion of multi-source data dynamically calculates the confidence weights of satellite remote sensing data, ground observation data, and numerical simulation data for each grid point that has completed spatiotemporal alignment and denoising, thereby realizing physically driven data fusion. Step 2: Construction of a cross-scale dynamic association mechanism based on AI models.

[0008] S21. AI Model Selection and Training: Dynamic graph neural network is selected as the association model. Graph node and edge weight functions are defined, message passing mechanism is designed, training set is constructed, and the model is trained. S22. Dynamic correlation calculation of pollutant flux: Based on the mass balance equation, the rationality of the initial flux field is verified. The edge contribution is calculated by the integral gradient method and the graph structure is optimized until the convergence condition is met, and the final pollutant flux distribution field is output.

[0009] Furthermore, in step S14, the physical-driven multi-source data fusion includes three sub-steps: S141. Upward source tracing: The spatial representativeness error between ground observation data and numerical simulation data is quantified using the Lagrange random particle diffusion model; S142, Downlink Assessment: Data quality diagnosis based on real-time meteorological conditions, quantifying data quality through real-time meteorological elements, and calculating dynamic quality factors for three types of multi-scale data sources; S143. Dynamic weighting and multi-source data fusion: Calculate the final fusion weight based on quantitative indicators of uplink tracing and downlink evaluation, and calculate the fusion value of the target point data.

[0010] Furthermore, the upstream tracing process is as follows: Calculate the representativeness of flux tower data: For flux tower sites, calculate their representativeness score of the underlying surface to the target grid points using the footprint function; Calculate the representativeness of numerical simulation data: For the model grid, the representativeness error is measured by the heterogeneity of the underlying surface; the model grid is the basic computational unit used for spatial discretization in numerical simulation.

[0011] Furthermore, the specific process of the downlink assessment is as follows: Satellite data quality factors are calculated by combining cloud cover ratio and ground visibility; numerical simulation data stability factors are calculated based on Richardson number; and ground station flow field correlation factors are calculated by combining the angle between flux tower stations and prevailing wind direction and horizontal attenuation.

[0012] Furthermore, the dynamic weight synthesis and multi-source data fusion process is as follows: First, the confidence factor of satellite remote sensing data is calculated based on the satellite data quality factor; the confidence factor of ground observation data is calculated based on the representativeness of flux tower data and the correlation factor of ground station flow field; and the confidence factor of numerical simulation data is calculated based on the representativeness of numerical simulation data and the stability factor of numerical simulation data. Second, the final weight of the target point data is calculated based on the three types of multi-scale data sources. Finally, the final fusion value of the target point data is calculated based on the initial fusion weight and the final weight.

[0013] Furthermore, the verification of the rationality of the initial flux based on the mass balance equation is carried out in the following specific process: The physical residuals between the initial flux value and the actual observed value are calculated using the discretized mass balance equation, and the average physical residuals within the region are also calculated.

[0014] Furthermore, the integral gradient method is used to calculate edge contributions and optimize the graph structure, as detailed below: Locating high residual nodes: Identifying high residual nodes by using the average physical residual within the region; Calculate edge contribution: For nodes with high residuals and their incoming edges, calculate the contribution using the integral gradient method; Dynamic optimization of graph structure: Contribution is normalized using the Sigmoid function, and edge weights are adjusted using the following formula: ; in, For learning rate, The current edge weight, For the Sigmoid function, The edge contribution of nodes with high residuals and their incoming edges.

[0015] Furthermore, in step S22, the convergence condition is as follows: ; in, For the first The area average physical residual was calculated once. The convergence threshold, This represents the number of iterations.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Two-way source tracing improves data fusion quality: A two-way source tracing mechanism of uplink and downlink evaluation is adopted. Uplink uses a Lagrange random particle diffusion model to quantify the spatial representativeness error of ground observation and numerical simulation data. Downlink combines real-time meteorological elements (cloud cover, Richardson number, etc.) to calculate dynamic quality factors of each data source, and finally dynamically generates confidence weights to achieve multi-source fusion. This method replaces traditional manual interpolation, effectively captures the physical correlation of data at different scales, significantly reduces scale transition errors, and provides a high-quality data foundation for flux estimation. 2. Dynamic Graph Structure Optimization Balancing Accuracy and Timeliness: After constructing a cross-scale correlation model using a dynamic graph neural network, the physical residuals are first calculated using the discretized mass balance equation to identify high residual nodes that meet preset conditions. Then, for high residual nodes and their incoming edges, the edge contribution is calculated using the integral gradient method (quantifying the degree of influence and positive or negative effect of the edge on the flux of high residual nodes). Subsequently, the contribution is normalized using the Sigmoid function, the edge weights are dynamically adjusted, and local forward inference is performed on high residual regions and neighboring nodes to correct the flux. Finally, the convergence of the regional average physical residuals is verified iteratively to output the global flux field. This process replaces the redundant operation of repeatedly verifying parameters and manually correcting abnormal flux values ​​in traditional methods. It improves the accuracy of cross-scale correlation through precise optimization of high residual regions while reducing ineffective iteration steps, solving the problem of difficulty in balancing accuracy and speed in traditional technologies, and meeting the needs of time-sensitive scenarios such as emergency monitoring of air pollution. Attached Figure Description

[0017] Figure 1 The flowchart shows an atmospheric flux estimation method based on multi-scale dynamic correlation. Figure 2 This is a schematic diagram illustrating the physical-driven multi-source data fusion of the present invention; Figure 3 This is a flowchart illustrating the dynamic correlation calculation of pollutant flux according to the present invention. Detailed Implementation

[0018] Example, refer to Figure 1 The atmospheric flux estimation method based on multi-scale dynamic correlation in this embodiment specifically includes the following steps: Step 1: Spatiotemporal alignment and noise suppression of multi-scale observation data.

[0019] S11. Data Acquisition: Acquire data from three types of multi-scale data sources; Satellite remote sensing data: For the scenario of estimating atmospheric pollutant fluxes, pollutant column concentration data, such as VOCs column concentration (Volatile Organic Compounds), from the Sentinel-5 Precursor (Sentinel-5P) satellite are used, with a spatial resolution of 3.5km and updated daily. Ground-based observation data: Based on the global flux observation network (FLUXNET) flux tower network, pollutant flux data with a spatial resolution of 50-200m and a temporal resolution of 30 minutes are collected; pollutant concentration data with a spatial resolution of 1-5km and a temporal resolution of 1 hour are obtained through ground-based automatic monitoring stations; Numerical simulation data: Meteorological field data with a spatial resolution of 10-50 km and a temporal resolution of 1 hour were generated using the Weather Research and Forecasting Model (WRF), including wind speed, temperature field, etc. S12, Spatiotemporal Alignment: Based on the observation timestamps of the ground flux tower, perform spatiotemporal matching on satellite data and numerical simulation data; Time alignment: Linear interpolation is used to unify the time resolution of satellite / simulated data to 30 minutes. For missing time nodes, data is supplemented by extrapolating the trends of adjacent time points. The extrapolation process specifically includes: using the least squares method to perform linear regression modeling on atmospheric pollutant flux data of multiple consecutive adjacent time points to obtain a trend equation; and predicting the atmospheric pollutant flux of the next time point based on the fitted trend equation. If data from multiple time points needs to be supplemented, iterative calculations are required sequentially. Spatial alignment: Construct a unified geographic grid (such as 500m×500m in this example), and use adaptive weighted interpolation to achieve data fusion, as detailed below: The interpolation weight W is dynamically adjusted based on the spatial resolution R of the data source. The calculation formula is as follows: ; in, For the first Each data source resolution, Total number of data sources; Based on a unified geographic grid, each grid cell is traversed. For data points from different scales that fall into the cell, a weighted average is calculated based on the weights. The weighted result is used as the final data value of the grid cell, thereby accurately mapping data of different scales to the same grid and avoiding spatial distortion caused by traditional fixed interpolation. S13. Noise Suppression: Differentiated denoising strategies are adopted based on the noise characteristics of different data sources; Satellite remote sensing data: Wavelet thresholding denoising method is used, Db4 wavelet basis is selected, and the threshold is calculated adaptively through data signal-to-noise ratio to eliminate outliers caused by cloud cover and atmospheric scattering; Ground observation data: Kalman filtering was used to filter out noise caused by instrument drift and transient interference based on the statistical characteristics of atmospheric pollutant flux fluctuation variance, mean offset, and outlier distribution frequency over 30 days of historical observation data. Numerical simulation data: First, using numerical models (such as WRF-Noah-MP, EC-Earth, and other atmosphere-land coupled models) based on meteorological reanalysis data, land use data, and topographic parameters, atmospheric pollutant fluxes at different altitudes are calculated by solving atmospheric dynamics and thermodynamic equations to obtain simulated values; then, the residual correction method is used to dynamically adjust the numerical simulation data by using the residuals between the simulated values ​​and the ground observations of the same period as correction terms to reduce systematic bias. S14, Physically driven multi-source data fusion: such as Figure 2 As shown, based on bidirectional traceability, multi-source data adaptive confidence weighted fusion dynamically calculates the confidence weights of satellite remote sensing data, ground observation data, and numerical simulation data for each grid point that has completed spatiotemporal alignment and denoising, thereby achieving physical-driven intelligent fusion and improving data quality. S141. Upward Source Tracing: Representativeness Assessment Based on Lagrange Footprint Model, using the Lagrange Random Particle Diffusion Model (LSM) to quantify the spatial representativeness error between ground observation data and numerical simulation data; Representativeness of flux tower data: For flux tower site s, the footprint function is used to represent the data. Calculate its representativeness score of the underlying surface for target grid point i. The formula is as follows: ; in, The valid source region for site s; For surface points The type of underlying surface, For target grid points Corresponding underlay type, This is an indicator function. The value of the indicator function is 1 when the condition inside the parentheses is true, and the value of the indicator function is 0 when the condition inside the parentheses is false. The higher the value, the better the representativeness; the underlying surface types include farmland, forest, urban areas, etc. Representativeness of numerical simulation data: For model grid m, the representativeness error is measured by the heterogeneity of the underlying surface. The calculation formula is as follows: ; in, For pattern grid Total number of high-resolution internal grids, Represents the m-th element within the pattern grid. A high-resolution grid of underlying surface types, The higher the value, the worse the representativeness; the model grid is the basic computational unit used for spatial discretization in numerical simulation. S142, Downlink Assessment: Data quality diagnosis based on real-time meteorological conditions, quantifying data quality through real-time meteorological elements, and calculating dynamic quality factors; Satellite data quality factor The calculation formula is as follows: ; in, The cloud cover ratio, which is an estimated proportion of the sky covered by clouds, is calculated using satellite sensors combined with algorithms. Ground visibility, in km. This is a reference threshold; Numerical simulation data stability factor The calculation formula is as follows: ; in, Here, 'a' is the Richardson number, and 'a' is an adjustment parameter determined through domain experience. In this embodiment... , The higher the value, the more unstable the atmosphere, and the higher the simulation confidence level. Ground station flow field correlation factor The calculation formula is as follows: ; in, The angle between the line connecting the flux tower site and the target point and the prevailing wind direction. Horizontal distance, attenuation scale These are empirical values; the prevailing wind direction refers to the wind direction with the highest frequency of occurrence in a certain region during a specific period, obtained through statistical methods, reflecting the dominant trend of atmospheric movement in that region. S143. Dynamic weight synthesis and multi-source data fusion: Calculate the final fusion weight using comprehensive quantitative indicators; Calculation of dynamic confidence factor: Confidence factor for satellite remote sensing data: ; Confidence factor for ground observation data: ; Number of sites within the affected area; Numerical simulation data confidence factor: ; Final weights and fusion values: The final weight of the data source src (containing satellite remote sensing data SAT, ground observation data GRD, and numerical simulation data MOD) at target point i. From static weights Dynamic confidence factor of target point i The calculation formula is as follows, jointly determined: ; Final fusion value of target point i for: ; in, For data source At point Observed or simulated values; The static weights Without introducing a two-way traceability mechanism, the initial fusion weights are determined solely based on the historical average accuracy of the data source. The specific process is as follows: For satellite remote sensing data (SAT), ground observation data (GRD), and numerical simulation data (MOD), their historical average accuracy in the target area is calculated respectively. (range of values) A higher value indicates better historical estimation accuracy (measured by long-term comparison of observed / simulated values ​​with actual values); static weights are calculated through normalization. ; Step 2: Construction of a cross-scale dynamic association mechanism based on AI models.

[0020] S21. AI Model Selection and Training: Dynamic Graph Neural Network (DyGNN) is selected as the correlation model to replace traditional manual empirical formulas. A three-layer message passing structure is adopted, and the specific design is as follows: Node definition: The multi-scale data grid cells processed in step one are used as graph nodes. Each node contains data features, such as Normalized Difference Vegetation Index (NDVI), surface temperature, and wind speed; and physical attributes, such as surface type and altitude. Dynamic calculation of edge weights: The edge weight function is designed based on atmospheric physical processes (such as turbulence intensity and water vapor gradient), and the formula is as follows: ; in, Let be the edge weight between node p and node q; The temperature difference between the two nodes; Spatial distance; The average wind speed is obtained through meteorological station monitoring equipment, meteorological satellite data, or numerical weather prediction models. The data covers wind speed information at different altitudes and is processed by time averaging to obtain the required average wind speed. The roughness length can be calculated based on field measurements or data obtained from a geographic information system, or by measuring terrain undulations using equipment such as lidar. The coefficients are dynamically adjusted and optimized using the gradient descent method based on real-time meteorological conditions. Message passing mechanism: The embedding update formula for the I-th layer node is: ; in, The LeakyReLU activation function is used. This is a function that aggregates the mean values ​​of neighboring features. and Let I be the trainable parameter matrix of the I-th layer. For nodes The neighbor node set, in this embodiment, is taken as the nodes within a spatial neighborhood radius of 5km. It is generally believed that meteorological elements within a 5km radius have strong spatial correlation. This represents the edge weight between node u and node v; This represents the feature embedding of node u in the l-th layer; This represents the feature embedding of node v in the l-th layer; This represents the updated feature embedding of node v in the (l+i)th layer; Model training: The training set was constructed using FLUXNET 2025 global flux observation data and corresponding multi-scale auxiliary data, with the measured values ​​of ground flux towers as labels. The model prediction error (root mean square error RMSE) was minimized using the Adam optimizer, and the data was divided into training and validation sets in an 8:2 ratio. Batch processing (batch size=128) was used for training, with 100 iterations. After each iteration, the RMSE was evaluated on the validation set. When the RMSE on the validation set did not decrease for 10 consecutive iterations, an early stopping mechanism was triggered, and the optimal model parameters were saved. Model output: Estimated atmospheric pollutant fluxes for each grid cell, with a time resolution of 30 minutes; S22. Dynamic correlation calculation for pollutant fluxes: This step uses an online dynamic cycle of simulation-evaluation-optimization to make the graph structure respond to the real-time atmospheric transport state, achieving dynamic correlation under physical constraints; such as Figure 3 As shown, the specific steps are as follows: S221. Initial Association Graph Construction and Reasoning: The preprocessed and fused multi-scale pollutant concentration and meteorological field data are input into the trained DyGNN model to initialize the global graph structure. ,in, For nodes, As an edge, This is the initial edge weight matrix; The multi-scale pollutant concentration and meteorological field data are organized in the form of a structured multidimensional tensor, with the dimension defined as [time dimension × spatial dimension × variable dimension]. If the study area covers a duration of... Hours, obtained after pretreatment A unified spatial grid, containing Given one pollutant and one meteorological variable, the data tensor shape is as follows: ; Among them, the time dimension ( ): Corresponds to minutes (or hours), reflecting the time scale; Spatial dimension ( ): This corresponds to the total number of grid cells after unified resampling, reflecting the fusion of multiple spatial scales; Variable dimensions ( This corresponds to the total number of pollutant concentration variables and meteorological field variables; for example, if it includes 5 pollutants and 6 meteorological parameters, then... ; Running forward inference yields the initial flux field. Where, the flux of node i , in, The node feature matrix; S222, Verification based on physical process feedback of mass balance: The rationality of the initial flux scenario is verified by discretizing the mass balance equation: ; in, The physical residual represents the difference between the initial flux value and the actual observed value. For future pollutant concentration monitoring, This represents the current pollutant concentration. Indicates time interval, This represents the pollutant flux estimated from the initial flux field. For wind speed vectors, The chemical degradation coefficient, It is the advection term; Calculate the average physical residual for the region: ; in, To verify the total number of nodes in the region; S223. Edge contribution calculation and graph optimization based on integral gradient: Optimize graph structure by quantifying the contribution of edges to nodes with high residuals; Locating high residual nodes: Identifying nodes that meet the requirements node combination ,in, To achieve a high residual coefficient, the optimization level of the graph structure is determined based on the specific scenario. In this embodiment... ; Calculate edge contribution: for nodes with high residuals and its incoming edge Contribution is calculated using the integral gradient method. : ; in, The current edge weight, This is the baseline weight (usually 0). For the flux output along the weighted path, To output the partial derivatives of the weights of opposite edges; The formula calculates It is a scalar, and its absolute value directly represents the edge. The degree of contribution to the pollutant flux at node i, with the symbol representing promotion or inhibition; Dynamic optimization of graph structure: normalizing contribution using the Sigmoid function and adjusting edge weights. ; in, The learning rate is typically set between 0.1 and 0.3. In this embodiment... By comparing different Indicators such as the degree of agreement between pollutant flux and measured values, and the convergence rate of physical residuals, are used to select the parameters that optimize the accuracy of the validation set. The degree of fit is determined by the coefficient of determination. As an evaluation metric, the convergence speed is evaluated using the number of iterations. For the Sigmoid function; right Local forward reasoning is performed on the flux and its neighboring nodes to obtain the corrected flux. ; S224, Output pollutant flux distribution field: Obtained by backfilling the corrected flux to the global field. Repeat steps S222~S223 until the convergence condition is met: ; in, The convergence threshold, This represents the number of iterations. Ultimately, the output is a pollutant flux distribution field that satisfies physical consistency. ; Through the detailed description of the above embodiments, the atmospheric flux estimation method based on multi-scale dynamic correlation of the present invention achieves efficient estimation through a progressive process of data preprocessing, physical fusion, and AI correlation optimization: First, spatiotemporal alignment and differential noise suppression are performed on three types of data: satellite remote sensing, ground observation, and numerical simulation, eliminating fundamental data biases; then, a physical-driven fusion mechanism is constructed based on bidirectional source tracing to dynamically quantify the quality of data sources and perform weighted fusion, thus solidifying the data foundation; finally, a cross-scale correlation is constructed using a dynamic graph neural network, and the graph structure is optimized by combining quality balance verification and integral gradient method, outputting an accurate flux distribution field after convergence judgment. This method, through its two core designs of bidirectional source tracing and graph structure optimization, effectively solves the problems of large scale connection errors and insufficient timeliness in existing technologies, balancing accuracy and speed, and providing reliable technical support for atmospheric pollution emergency response and regional carbon cycle monitoring.

[0021] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0022] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0023] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0024] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0025] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0026] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0027] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An atmospheric flux estimation method based on multi-scale dynamic correlation, characterized in that, The method flow is as follows: Step 1: Spatiotemporal alignment and noise suppression of multi-scale observation data: S11. Data Acquisition: Acquire three types of multi-scale data sources: satellite remote sensing data, ground observation data, and numerical simulation data; S12. Spatiotemporal alignment: Based on the ground flux tower observation timestamp, perform temporal and spatial alignment on satellite remote sensing data and numerical simulation data; S13. Noise Suppression: Differentiated denoising strategies are adopted for satellite remote sensing data, ground observation data, and numerical simulation data, taking into account their different noise characteristics. S14. Physically driven multi-source data fusion: Based on bidirectional traceability, the adaptive confidence weighted fusion of multi-source data dynamically calculates the confidence weights of satellite remote sensing data, ground observation data, and numerical simulation data for each grid point that has completed spatiotemporal alignment and denoising, thereby realizing physically driven data fusion. Step 2: Construction of a cross-scale dynamic association mechanism based on an AI model: S21. AI Model Selection and Training: Dynamic graph neural network is selected as the association model. Graph node and edge weight functions are defined, message passing mechanism is designed, training set is constructed, and the model is trained. S22. Dynamic correlation calculation of pollutant flux: Based on the mass balance equation, the rationality of the initial flux field is verified. The edge contribution is calculated by the integral gradient method and the graph structure is optimized until the convergence condition is met, and the final pollutant flux distribution field is output.

2. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 1, characterized in that, In step S14, the physical-driven multi-source data fusion includes three sub-steps: S141. Upward source tracing: The spatial representativeness error between ground observation data and numerical simulation data is quantified using the Lagrange random particle diffusion model; S142, Downlink Assessment: Data quality diagnosis based on real-time meteorological conditions, quantifying data quality through real-time meteorological elements, and calculating dynamic quality factors for three types of multi-scale data sources; S143. Dynamic weighting and multi-source data fusion: Calculate the final fusion weight based on quantitative indicators of uplink tracing and downlink evaluation, and calculate the fusion value of the target point data.

3. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 2, characterized in that, The upstream tracing process is as follows: Calculate the representativeness of flux tower data: For flux tower sites, calculate their representativeness score of the underlying surface to the target grid points using the footprint function; Calculate the representativeness of numerical simulation data: For the model grid, the representativeness error is measured by the heterogeneity of the underlying surface; the model grid is the basic computational unit used for spatial discretization in numerical simulation.

4. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 3, characterized in that, The specific process of the downlink assessment is as follows: Satellite data quality factors are calculated by combining cloud cover ratio and ground visibility; numerical simulation data stability factors are calculated based on Richardson number; and ground station flow field correlation factors are calculated by combining the angle between flux tower stations and prevailing wind direction and horizontal attenuation.

5. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 4, characterized in that, The dynamic weight synthesis and multi-source data fusion process is as follows: First, the confidence factor of satellite remote sensing data is calculated based on the satellite data quality factor; the confidence factor of ground observation data is calculated based on the representativeness of flux tower data and the correlation factor of ground station flow field; and the confidence factor of numerical simulation data is calculated based on the representativeness of numerical simulation data and the stability factor of numerical simulation data. Secondly, the final weights of the target point data are calculated based on three types of multi-scale data sources; Finally, based on the initial fusion weights and the final weights, the final fusion value of the target point data is calculated.

6. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 1, characterized in that, The specific process for verifying the rationality of the initial flux based on the mass balance equation is as follows: The physical residuals between the initial flux value and the actual observed value are calculated using the discretized mass balance equation, and the average physical residuals within the region are also calculated.

7. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 6, characterized in that, The integral gradient method is used to calculate edge contributions and optimize the graph structure, as detailed below: Locating high residual nodes: Identifying high residual nodes by using the average physical residual within the region; Calculate edge contribution: For nodes with high residuals and their incoming edges, calculate the contribution using the integral gradient method; Dynamic optimization of graph structure: Contribution is normalized using the Sigmoid function, and edge weights are adjusted using the following formula: ； in, For learning rate, The current edge weight, For the Sigmoid function, The edge contribution of nodes with high residuals and their incoming edges.

8. The atmospheric flux estimation method based on multi-scale dynamic correlation according to claim 1, characterized in that, In step S22, the convergence condition is as follows: ； in, For the first The area average physical residual was calculated once. The convergence threshold, This represents the number of iterations.

Images