Method and system for encrypting heterogeneous carbon sink nodes based on assimilation inversion feedback

Abstract

The application discloses a heterogeneous carbon sink node encryption method and system based on assimilation inversion feedback, and relates to the field of carbon sink monitoring networks.The method comprises the following steps: obtaining multi-source observation data and background driving data collected by ordinary nodes and core nodes, inputting the data into a surface carbon cycle model for data assimilation, obtaining posteriori carbon flux estimation results and a posteriori covariance matrix of each spatial grid; extracting local uncertainty and carbon flux spatial gradient based on the posteriori results, calculating a spatial heterogeneity risk index to determine candidate grids with higher monitoring risks; simulating the covariance updating process of the candidate grids after the addition of new core nodes or the upgrading of ordinary nodes, quantifying the expected reduction amount of global estimation uncertainty, and combining the risk index, the expected reduction amount and resource budget constraints to generate a heterogeneous node encryption strategy. Thus, the monitoring resources can be adaptively concentrated in high-uncertainty and high-heterogeneity areas, and the observation accuracy and configuration efficiency of the carbon sink monitoring network can be improved.

Description

Technical Field

[0001] This application relates to the field of carbon sink monitoring network technology, and in particular to a method and system for encrypting heterogeneous carbon sink nodes based on assimilation and inversion feedback. Background Technology

[0002] Carbon flux monitoring is a crucial technological foundation for global carbon cycle research, regional carbon source and sink assessment, and ecological environment change analysis. To obtain carbon flux information for target regions, current methods typically employ a combination of techniques, including ground-based flux tower networks, remote sensing observations, low-cost sensor networks, and related model estimations, to monitor, estimate, and characterize regional carbon exchange processes. Because different techniques have varying characteristics in terms of observation accuracy, spatial coverage, temporal resolution, and construction and maintenance costs, current carbon flux monitoring systems generally exhibit a technological structure supported by multi-source observation data and multiple monitoring methods.

[0003] In the construction of carbon flux monitoring networks, the type, number, and spatial location of sensor nodes are typically planned in advance based on the basic conditions of the target area, the monitoring range, and deployment requirements. After deployment, the overall network topology remains relatively stable. This deployment method is characterized by a clear implementation path, convenient engineering organization, and relatively direct operation and maintenance management, and is therefore widely used in relevant technical scenarios. However, the construction process of this type of monitoring network largely relies on prior judgments of regional distribution characteristics, resulting in a network structure with strong static configuration characteristics.

[0004] However, in practical applications, carbon flux distribution is often influenced by a combination of factors, including topography, vegetation type, soil moisture, local meteorological conditions, and human activities. It typically exhibits strong variability and heterogeneity in both temporal and spatial dimensions, especially in ecological boundary areas, regions with significant land cover changes, and areas with strong local disturbances, where carbon flux characteristics may vary considerably across different locations. Furthermore, current carbon flux monitoring systems typically consist of a front-end observation and acquisition unit and a back-end data processing unit. While the back-end can perform fusion analysis, state estimation, or inversion calculations based on observation results, current systems still primarily focus on the implementation of conventional monitoring and data processing, particularly regarding the configuration of different types of nodes, the adaptation and adjustment of network structures, and the coordinated utilization of multi-level observation resources. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for densifying heterogeneous carbon sink nodes based on assimilation and inversion feedback, which can at least solve the problem of the difficulty in rationally allocating observation resources in the process of monitoring carbon flux in complex areas in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for densifying heterogeneous carbon sink nodes based on assimilation and inversion feedback. The method includes: acquiring multi-source observation data and corresponding background driving data collected by a heterogeneous monitoring network deployed in a target area; wherein the target area is divided into multiple spatial grids, and the heterogeneous monitoring network includes ordinary nodes deployed in the spatial grids with an observation accuracy at a first accuracy level, and core nodes with an observation accuracy at a second accuracy level higher than the first accuracy level; the background driving data includes at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture data; inputting the multi-source observation data and the background driving data into a preset surface carbon cycle model to perform data assimilation processing to obtain posterior carbon flux estimation results and posterior covariance matrices characterizing the estimation uncertainty of each spatial grid; wherein the surface carbon cycle model is used to drive and simulate the carbon cycle physical process in the target area under the background driving data; based on the posterior carbon flux estimation results and the posterior covariance matrix, extracting the local uncertainty and spatial gradient information characterizing the spatial rate of change of carbon flux of each spatial grid, and utilizing each spatial grid The spatial heterogeneity risk index of each spatial grid is calculated based on the local uncertainty and the spatial gradient information to characterize the monitoring risk of each spatial grid. For candidate grids whose spatial heterogeneity risk index meets the preset triggering conditions, the covariance update process after adding the core node or upgrading the ordinary node deployed in the corresponding candidate grid to the core node is simulated using the posterior covariance matrix to quantify the expected reduction in the global estimation uncertainty corresponding to the candidate grid. The node selection decision is performed by combining the spatial heterogeneity risk index, the expected reduction, and the preset resource budget constraints of each candidate grid to generate a heterogeneous node encryption strategy. The heterogeneous node encryption strategy includes a core node addition instruction and / or an existing ordinary node capability upgrade instruction. The existing ordinary node capability upgrade instruction is used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level. The instruction indicated by the heterogeneous node encryption strategy is output to trigger the core node addition deployment operation and / or ordinary node upgrade operation in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

[0007] Secondly, embodiments of this application provide a heterogeneous carbon sink node encryption system based on assimilation and inversion feedback. The system includes: a multi-source data acquisition unit, used to acquire multi-source observation data and corresponding background driving data collected by a heterogeneous monitoring network deployed in a target area; wherein the target area is divided into multiple spatial grids, and the heterogeneous monitoring network includes ordinary nodes deployed in the spatial grids with an observation accuracy at a first accuracy level, and core nodes with an observation accuracy at a second accuracy level higher than the first accuracy level; the background driving data includes at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture. The system includes: a data assimilation processing unit, used to input the multi-source observation data and the background driving data into a preset surface carbon cycle model to perform data assimilation processing, so as to obtain the posterior carbon flux estimation results and the posterior covariance matrix characterizing the estimation uncertainty of each spatial grid; wherein, the surface carbon cycle model is used to drive the simulation of the carbon cycle physical process of the target area under the background driving data; and a spatial heterogeneity risk assessment unit, used to extract the local uncertainty and spatial gradient information characterizing the spatial change rate of carbon flux of each spatial grid based on the posterior carbon flux estimation results and the posterior covariance matrix, and to utilize the spatial grids... The system calculates the spatial heterogeneity risk index of each spatial grid based on the local uncertainty and the spatial gradient information, to characterize the monitoring risk of each spatial grid; the expected return quantification unit is used to simulate the covariance update process of adding the core node or upgrading the ordinary node deployed in the corresponding candidate grid to the core node using the posterior covariance matrix for candidate grids whose spatial heterogeneity risk index meets the preset trigger conditions, so as to quantify the expected reduction in the global estimation uncertainty corresponding to the candidate grid; the heterogeneous node selection decision unit is used to comprehensively consider the spatial heterogeneity risk index and the expected return quantification unit of each candidate grid. To reduce the amount of data and constrain the preset resource budget, the node selection decision is implemented to generate a heterogeneous node encryption strategy. This strategy includes instructions for adding core nodes and / or upgrading the capabilities of existing ordinary nodes. The upgrade instructions for existing ordinary nodes are used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level. A topology dynamic adjustment unit is used to output the instructions indicated by the heterogeneous node encryption strategy to trigger the execution of core node addition and / or ordinary node upgrade operations in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the heterogeneous carbon sink node encryption method based on assimilation inversion feedback according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the heterogeneous carbon sink node encryption method based on assimilation and inversion feedback according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the heterogeneous carbon sink node encryption method based on assimilation and inversion feedback according to any embodiment of this application.

[0011] The heterogeneous carbon sink node encryption method and system based on assimilation and inversion feedback provided in this application can achieve at least the following technical effects: Multi-source observational data and background driving data are jointly incorporated into the surface carbon cycle model for data assimilation. Based on the obtained posterior carbon flux estimates, the uncertainty information represented by the posterior covariance matrix and the spatial gradient information of carbon flux are further combined to construct a spatial heterogeneity risk index to characterize the monitoring risk of each spatial grid. This approach enables the monitoring system to not only focus on the carbon flux values ​​themselves but also to simultaneously identify key areas within the target region that have weaker estimation stability and are more sensitive to spatial changes, thereby more accurately depicting the spatial heterogeneity of carbon flux distribution within the target region. Consequently, the monitoring network possesses a stronger analytical capability for spatial changes caused by the combined effects of complex terrain, vegetation differences, and local disturbances, providing a reliable quantitative basis for the precise allocation of subsequent observation resources.

[0012] By simulating the covariance update process of candidate grids after the addition of core nodes or the upgrading of ordinary nodes to core nodes using the posterior covariance matrix, the expected reduction in global estimation uncertainty by the corresponding candidate grids is quantified. This is then combined with resource budget constraints to execute node optimization decisions. Consequently, before implementing physical deployment adjustments, the actual contribution of different candidate grids to the optimization of overall carbon flux estimation results can be pre-assessed, ensuring that node additions or upgrades not only match local spatial heterogeneity but also balance global estimation accuracy improvement with resource efficiency. Furthermore, the generated heterogeneous node densification strategy directly affects the adjustment of physical deployment topology and node capability configuration, effectively improving the rationality and implementation effectiveness of high-precision observation resource allocation.

[0013] This technical solution establishes a closed-loop feedback mechanism that directly transmits back-end assimilation and inversion results to the front-end heterogeneous monitoring network deployment optimization. This enables the carbon flux monitoring system to adaptively evolve based on regional spatial heterogeneity and global estimation requirements. Consequently, it not only improves the precision and stability of carbon flux monitoring and estimation results in the target area but also enhances the coverage of high-risk observation blind spots by high-precision node configuration, thus achieving a synergistic balance between improved monitoring accuracy and optimized utilization of limited observation resources. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart is shown as an example of a heterogeneous carbon sink node encryption method based on assimilation inversion feedback according to an embodiment of this application; Figure 2 A flowchart illustrating an example of determining the spatial heterogeneity risk index of each spatial grid in a method according to an embodiment of this application is shown. Figure 3 A flowchart illustrating an example of the expected reduction in the global estimation uncertainty corresponding to the candidate grid in the method according to an embodiment of this application is shown. Figure 4 This document illustrates an example of an operation flowchart for generating a heterogeneous node encryption strategy through multi-factor node optimization decision-making in a method according to an embodiment of this application. Figure 5 A schematic diagram illustrating the system operation mechanism of an example heterogeneous carbon sink node encryption method based on assimilation inversion feedback according to an embodiment of this application is shown. Figure 6 A comparative simulation diagram illustrating an example of the evolution trend of posterior uncertainty under different node encryption strategies is shown. Figure 7 This diagram illustrates an example of the spatial topology evolution and deployment distribution of a heterogeneous node encryption strategy generated using the method of this application within a target area. Figure 8 A Pareto front comparison and systems engineering boundary analysis diagram of different node encryption decision-making methods in the cost-effectiveness dimension is shown; Figure 9 A structural block diagram of an example heterogeneous carbon sink node encryption system based on assimilation inversion feedback according to an embodiment of this application is shown. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] It should be noted that, in current related technologies, some studies have attempted to integrate remote sensing observations, ground observations, and surface process models to improve the reliability of regional carbon flux estimation results. For example, CCDAS (Carbon Cycle Data Assimilation System) can assimilate remote sensing variables such as SMOS (Soil Moisture and Ocean Salinity) and FAPAR (Fraction of Absorbed Photosynthetically Active Radiation) into BETHY (Biosphere Energy Transfer Hydrology) to correct model parameters and improve carbon flux simulation results. Such methods can improve the consistency between model output and observational information to some extent and form posterior uncertainty information to characterize the reliability of the estimation; however, their focus is usually still on model parameter optimization, state estimation, or carbon flux product generation, and the relevant uncertainty information is more often used as a basis for evaluating model results, and less often used to guide the deployment and adjustment of front-end monitoring nodes or the configuration of observation capabilities.

[0018] With the development of low-cost carbon dioxide sensors and micrometeorological sensors, some studies have begun to construct high-density, low-cost gas monitoring networks. By deploying low-cost CO2 sensor nodes and combining them with environmental correction and drift correction, the spatial coverage of urban-scale CO2 concentration observations has been improved. Other studies have used low-cost CO2 sensors such as the SenseAir K30 to construct multi-point observation systems to reduce the deployment cost of high-density monitoring networks. These technologies demonstrate that low-cost nodes have certain advantages in expanding observation coverage, but their deployment methods typically still rely on pre-set locations and fixed network structures, and are mostly applied in urban areas, near-shore platforms, or scenarios with relatively stable operation and maintenance conditions. For carbon sink monitoring areas with significant topographic relief, complex ecological boundaries, or strong local micrometeorological disturbances, the ability of the network structure to dynamically adjust to changes in monitoring needs remains relatively limited.

[0019] In high-resolution carbon flux inversion, some studies employ ensemble filtering methods such as GEnSRF (Geostatistical Ensemble Square Root Filter) to fuse observational data and utilize covariance dilation, localization, and observation error settings to improve the inversion results. These methods can address high-dimensional state estimation problems to some extent and quantify the uncertainty of carbon flux estimation results. However, the inversion performance of ensemble filtering methods is typically closely related to the spatial distribution density of the front-end observation network, the accuracy of node observations, and the setting of observation errors. When the actual ground observation network is sparse and uneven, or when observation nodes cannot cover ecological transition zones, relying solely on the back-end algorithm is insufficient to fully compensate for the mismatch between the front-end observation structure and the spatial heterogeneity of regional carbon flux.

[0020] In addition, some data-driven carbon flux estimation methods attempt to combine flux tower observations with spatial variables such as PAR (Photosynthetically Active Radiation), NDVI (Normalized Difference Vegetation Index), temperature, precipitation, and soil properties to estimate NEE (Net Ecosystem Exchange) at different locations and time periods. These methods can reveal the spatial differences in carbon flux caused by the influence of climate gradients, vegetation status, and soil conditions, and also demonstrate that carbon flux distribution is not a simple smooth field. However, these methods typically rely on existing flux towers or monitoring points to provide training, calibration, and validation data. When the number of monitoring points is limited or their distribution is uneven, their ability to characterize ecological boundaries, areas of rapid land cover change, and areas of local disturbance remains affected by the observation layout.

[0021] Furthermore, some comprehensive carbon monitoring projects attempt to integrate satellite, airborne, and ground-based observation data to create high-resolution carbon flux products, emphasizing error assessment and uncertainty quantification. These systems possess strong comprehensive capabilities in multi-source data fusion, product generation, and scale conversion; however, their primary objectives are typically carbon monitoring product construction, carbon cycle process characterization, and uncertainty assessment. Currently, there is a lack of detailed processing methods tailored to carbon sink monitoring scenarios for dynamically adding, upgrading, or reallocating heterogeneous front-end monitoring nodes based on back-end uncertainty diagnosis results.

[0022] In summary, current technologies have laid a certain foundation in remote sensing variable assimilation, low-cost sensor networking, ensemble filtering inversion, and multi-source carbon monitoring product generation. However, these technologies primarily focus on observational data acquisition, model parameter optimization, inversion calculation, or product evaluation. In complex carbon sink monitoring scenarios, low-cost ordinary nodes offer coverage advantages but have relatively limited observation accuracy and stability, while high-precision core nodes have strong observational constraints but higher deployment and operating costs. Without technical means to integrate posterior uncertainty, spatial variation characteristics of carbon flux, and hardware resource constraints into the node configuration process, situations may arise where some key areas have insufficient observation capabilities while other areas have relatively redundant resource configurations. Therefore, how to make the backend assimilation and inversion results more effectively serve the dynamic configuration of the frontend heterogeneous monitoring network remains a technical issue that needs further consideration in the optimization of carbon sink monitoring networks.

[0023] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0024] Figure 1 A flowchart illustrating an example of a heterogeneous carbon sink node encryption method based on assimilation inversion feedback according to an embodiment of this application is shown.

[0025] Regarding the execution entity of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as a carbon sink monitoring network management controller, which implements the method of the embodiments of this application by running programs or instructions stored in a storage medium. In some examples, it can be integrated and configured in an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse.

[0026] like Figure 1 As shown, in step S110, multi-source observation data and corresponding background driving data are acquired by a heterogeneous monitoring network deployed in the target area.

[0027] In this embodiment, the target area can be a forest area, grassland area, wetland area, farmland area, urban fringe ecological transition area, or other geographical area requiring dynamic monitoring of regional carbon flux. Here, the target area is divided into multiple spatial grids. Since such areas are usually large in scale and vary in topography, vegetation cover type, surface humidity conditions, and human activity intensity, the target area can be pre-divided into multiple spatial grids according to a uniform spatial resolution, so that subsequent state estimation, monitoring risk identification, and node scheduling are all carried out around a unified spatial unit. For example, for the forest-farmland boundary area, the monitoring area can be gridded based on vegetation cover change characteristics and topographic relief; for urban fringe areas, a corresponding spatial grid structure can be established based on the distribution of construction land, green space, and ecological buffer zones.

[0028] The heterogeneous monitoring network comprises ordinary nodes deployed in a spatial grid with observation accuracy at the first level, and core nodes with observation accuracy at the second level, which is higher than the first level. Ordinary nodes primarily provide basic observation information under high spatial coverage density; their equipment can be low-cost carbon concentration monitoring units, basic micrometeorological monitoring units, or other sensing devices suitable for mass deployment. Core nodes primarily provide critical observation information with higher accuracy or higher reliability; their equipment can be high-precision carbon flux monitoring devices, flux tower observation equipment, or other enhanced observation equipment. By combining ordinary and core nodes, a basic monitoring system that balances coverage and observation accuracy can be formed.

[0029] Multi-source observation data can include carbon dioxide concentration data, temperature data, humidity data, wind speed and direction data, radiation data, and other environmental observation data that can reflect the local carbon exchange status, collected by ordinary nodes and core nodes. Background driving data is used to characterize the external driving conditions of the carbon cycle process in the target area, and can include at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture data. To facilitate subsequent data assimilation, the above data can be preprocessed by performing time alignment, spatial matching, outlier removal, missing value completion, and unit unification to form a standardized dataset that can be input into the surface carbon cycle model. In this way, the gridded representation of the target area and the data preparation of the heterogeneous observation system are completed, enabling the system to utilize both the wide-area basic observation information provided by ordinary nodes and the high-precision observation information provided by core nodes in subsequent processing, providing basic data support for carbon flux state estimation and monitoring network optimization.

[0030] In step S120, multi-source observation data and background driving data are input into a preset surface carbon cycle model to perform data assimilation processing, so as to obtain the posterior carbon flux estimation results of each spatial grid and the posterior covariance matrix characterizing the estimation uncertainty. Here, the surface carbon cycle model is used to drive the simulation of the carbon cycle physical processes in the target area under the background driving data.

[0031] In this embodiment, the surface carbon cycle model is used to describe the dynamic changes in carbon absorption, carbon release, and related ecological exchange processes in the target area under background driving conditions. Specifically, the meteorological forcing data, vegetation index data, and soil moisture data obtained in step S110 can be input into the surface carbon cycle model, enabling the model to make prior predictions on the carbon flux status of each spatial grid within the current assimilation cycle. For example, in areas with high vegetation cover, the model can characterize the vegetation photosynthetic process based on radiation conditions, temperature conditions, and vegetation indices; in areas with significant changes in soil moisture, the model can characterize soil respiration and carbon release processes based on soil moisture and temperature conditions.

[0032] After obtaining the prior prediction results, multi-source observation data collected by the heterogeneous monitoring network are introduced into the assimilation and update process as external observation constraints, allowing the model prediction results to be fused and corrected with the actual observation results. Since there are differences in observation accuracy between ordinary nodes and core nodes, the assimilation process can distinguish the observation error representations of nodes with different accuracy levels, thus giving high-precision observations a stronger constraint capability in state correction, while retaining the spatial coverage advantage of ordinary nodes. After the above processing, the posterior carbon flux estimation results for each spatial grid in the current period can be obtained.

[0033] Furthermore, a posterior covariance matrix corresponding to the posterior carbon flux estimation results is generated. This posterior covariance matrix is ​​used to characterize the uncertainty level of the state estimation results for each spatial grid and the correlation of estimation errors between different spatial grids. The main diagonal elements characterize the uncertainty of the state estimation of each spatial grid itself, while the off-diagonal elements characterize the correlation of errors between different spatial grids. Thus, not only are posterior carbon flux estimation results obtained for each spatial grid in the target area, but also the uncertainty characterization basis required for subsequent risk identification and node densification benefit assessment is simultaneously formed.

[0034] In step S130, based on the posterior carbon flux estimation results and the posterior covariance matrix, the local uncertainty of each spatial grid and the spatial gradient information characterizing the spatial rate of change of carbon flux are extracted. The spatial heterogeneity risk index of each spatial grid is calculated using the local uncertainty and spatial gradient information of each spatial grid to characterize the monitoring risk of each spatial grid.

[0035] In this embodiment, local uncertainty is used to reflect the reliability of the current carbon flux estimation results for a spatial grid. Higher local uncertainty indicates a significant deficiency in the understanding of the spatial grid's state under current monitoring network conditions. Spatial gradient information reflects the strength of carbon flux changes between the spatial grid and its surrounding grids. A larger spatial gradient indicates a less smooth spatial distribution of carbon flux in the region, with more pronounced local transition characteristics. For example, in areas where forests meet farmland, wetland edges, or urban green spaces meet construction land, the carbon exchange states between adjacent spatial grids typically show significant differences, and the corresponding spatial gradient information is also stronger.

[0036] Specifically, the uncertainty characterization of each spatial grid can be extracted from the posterior covariance matrix first, and this characterization can be defined as the local uncertainty of the corresponding spatial grid. Then, based on the difference in posterior carbon flux estimation results between each spatial grid and its neighboring spatial grids, the spatial gradient information of the corresponding spatial grid can be extracted. Furthermore, to avoid judging the monitoring risk solely based on the state of a single spatial grid, the uncertainty levels of other grids within the local spatial neighborhood of the target spatial grid can be combined to form a neighborhood background risk reference. Subsequently, the local uncertainty, spatial gradient information, and neighborhood background risk reference of the target spatial grid are comprehensively processed to obtain the spatial heterogeneity risk index of the corresponding spatial grid.

[0037] In this way, the resulting spatial heterogeneity risk index not only reflects the magnitude of the estimation error of a particular spatial grid, but also whether that spatial grid is located in a region highly sensitive to spatial changes. Consequently, subsequent node densification decisions can prioritize spatial grids that exhibit both high uncertainty and are situated in distinct spatial transition zones or regions of drastic local change, thereby improving the targeting of monitoring resource allocation.

[0038] In step S140, for candidate grids whose spatial heterogeneity risk index meets the preset triggering conditions, the covariance update process after adding a core node or upgrading an ordinary node deployed in the corresponding candidate grid to a core node is simulated using the posterior covariance matrix, so as to quantify the expected reduction in the global estimation uncertainty corresponding to the candidate grid.

[0039] It should be noted that the spatial heterogeneity risk index is used to identify candidate grids with high monitoring enhancement needs, but it is insufficient to directly determine whether to invest in second-level precision observation resources based solely on the spatial heterogeneity risk index. Therefore, in this embodiment, a pre-evaluation of the global improvement effect of introducing high-precision observation capabilities on each candidate grid is also performed. This high-precision observation capability can be manifested either as adding a core node at the candidate grid or as upgrading an existing ordinary node at the candidate grid to a core node.

[0040] Specifically, without actually deploying hardware, the state update effect of introducing a second-level precision observation to a candidate grid can be simulated based on the current posterior covariance matrix. Since the posterior covariance matrix contains the error correlations between spatial grids in the target region, introducing higher-precision observation capabilities at a candidate grid will not only change the uncertainty of that candidate grid itself, but also potentially shrink the uncertainties of other associated spatial grids. By comparing the differences in the global uncertainty characterization results before and after the simulation update, the expected reduction in uncertainty corresponding to that candidate grid can be quantified.

[0041] For example, for spatial grids located in the "forest-farmland transition zone" and significantly influenced by the prevailing wind direction, introducing core nodes in this area can typically enhance the grid's own state constraints and may also converge the estimation errors of downwind or adjacent grids. However, for spatial grids that are relatively isolated and weakly coupled with surrounding grids, even introducing core nodes may have limited global improvement effects. Thus, the benefits of densification for different candidate grids can be quantitatively evaluated before actual resource allocation, providing a basis for subsequent node selection decisions.

[0042] In step S150, a heterogeneous node optimization decision is generated by comprehensively considering the spatial heterogeneity risk index, expected reduction amount, and preset resource budget constraints of each candidate grid. Here, the heterogeneous node optimization strategy includes core node addition instructions and / or existing ordinary node capability upgrade instructions. The existing ordinary node capability upgrade instructions are used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level.

[0043] In this embodiment, resource budget constraints may include equipment purchase budget constraints, operating energy consumption constraints, implementation condition constraints, maintenance capability constraints, or other constraints used to limit the actual resource input boundary. Node selection decisions are not simply ranked from highest to lowest according to the spatial heterogeneity risk index, nor are target grids determined solely based on the expected reduction amount. Instead, they simultaneously consider local monitoring needs, global benefit potential, and actual resource conditions. For example, when multiple candidate grids have high spatial heterogeneity risk indices, if a candidate grid has a larger expected reduction amount and lower implementation cost, its priority can be higher; conversely, for candidate grids with high spatial heterogeneity risk indices but significantly high implementation costs and limited global improvement effects, their priority can be appropriately reduced.

[0044] In practice, each candidate grid can be individually assessed to determine whether adding a core node or upgrading the capabilities of ordinary nodes is more suitable. For candidate grids that did not previously deploy high-precision observation capabilities but meet the conditions for addition, a core node addition command can be generated. For candidate grids that have already deployed ordinary nodes and whose locations have high upgrade value, an ordinary node capability upgrade command can be generated, increasing the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level. In this way, the resulting heterogeneous node encryption strategy can not only determine the target grid set but also clarify the specific action type corresponding to each target grid.

[0045] For example, in a large farmland area, if some peripheral spatial grids already have ordinary nodes deployed and these nodes are located at the boundary between farmland and woodland, upgrading the capabilities of existing ordinary nodes can be prioritized. Conversely, for some high-risk spatial grids that have not yet deployed high-precision observation capabilities but are located at the edge of wetlands or in suburban ecological boundaries, adding new core nodes can be prioritized. Thus, the resulting heterogeneous node encryption strategy can balance the need to improve the overall accuracy of the monitoring network with the real-world resource boundaries, allowing high-precision observation capabilities to be prioritized for deployment in more worthwhile locations.

[0046] In step S160, the instructions indicated by the heterogeneous node encryption strategy are output to trigger the core node new deployment operation and / or ordinary node upgrade operation to be performed in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

[0047] In this embodiment, after the heterogeneous node encryption strategy is generated, it can be converted into scheduling instructions for the execution layer. For core node addition instructions, device deployment, access, and calibration operations at the target spatial grid can be triggered; for ordinary node upgrade instructions, hardware module replacement, sampling capability enhancement, observation accuracy improvement, or other capability adjustment operations corresponding to the second accuracy level can be triggered. Thus, the aforementioned digital layer decisions based on assimilation and inversion results, monitoring risk identification, and global benefit assessment can be implemented in the actual physical adjustment process of the monitoring network.

[0048] After node deployment or upgrades are completed, the state of the heterogeneous monitoring network will change. On the one hand, if a new core node deployment is performed, the spatial distribution pattern of high-precision observation nodes in the target area will change, and the physical deployment topology of the network will be adjusted. On the other hand, if a regular node upgrade is performed, the capability configuration relationship of nodes with different precision levels within the network will change, and the node capability configuration of the network will be optimized. After completing the above adjustments, the updated heterogeneous monitoring network can continue to collect new multi-source observation data.

[0049] For example, after adding core nodes in suburban ecological transition zones, the region can continuously provide more reliable carbon flux constraint information during subsequent monitoring cycles, thereby reducing the uncertainty of state estimation in the region and its surrounding areas. Similarly, after upgrading nodes in farmland edge areas where ordinary nodes have already been deployed, the observation quality can be improved without altering existing station locations. This creates a closed-loop dynamic optimization process for the entire system, encompassing data acquisition, assimilation and inversion, risk identification, benefit assessment, node densification, and feedback updates, thereby improving the accuracy, adaptability, and resource utilization efficiency of carbon flux monitoring in the target area.

[0050] Regarding the implementation details of obtaining the posterior covariance matrix through data assimilation processing in step S120, in some examples of embodiments of this application, in order to obtain the prior state of carbon flux in the target area during the current assimilation period, firstly, the net ecosystem exchange process of the target area under background driving data is simulated using a surface carbon cycle model to generate a prior forecast set. The prior forecast set contains multiple prior set members, each of which is used to characterize the prior state distribution of carbon flux in each spatial grid within the target area.

[0051] It should be noted that the surface carbon cycle model can be a vegetation photosynthesis model based on light energy use efficiency, a land surface process model, or other carbon cycle models that can characterize vegetation absorption, ecosystem respiration, soil release, and environmental driving effects. Background driving data can include meteorological forcing data, vegetation index data, and soil moisture data, etc., to characterize the impact of temperature, humidity, radiation, vegetation growth status, and soil moisture conditions on the carbon exchange process.

[0052] Considering the potential perturbations and uncertainties in the background driving data and model parameters, in practical implementation, perturbations can be applied to meteorological forcing data, vegetation physiological parameters, soil parameters, or other key inputs. Multiple parallel simulations can then be performed based on these perturbated inputs to drive the surface carbon cycle model, thereby generating a priori forecast set. Let... Given the total number of set members preset within the current assimilation period, the prior prediction set can be represented as:

[0053] Equation (1) in, Indicates the first There are 10 prior set members, each of which is used to characterize the prior state distribution of carbon flux in each spatial grid within the target area. Specifically, this state distribution refers to the spatial numerical arrangement of carbon absorption or carbon release intensity corresponding to each spatial grid, derived from the forward extrapolation of the surface carbon cycle model before the actual observation data of the current period is introduced as constraints.

[0054] In this way, the system outputs not a single prior forecast result, but a set of results that reflect the discrete distribution characteristics of the prior state. This set of results can characterize the prior state of carbon flux in the target area in the current cycle, and can also reflect the uncertainty level of the prior state through the degree of dispersion among the set members, so that the prior forecast result has a more complete statistical expression capability.

[0055] Then, the multi-source observation data is converted into actual observation vectors, and observation operators are constructed to map the state variables of the surface carbon cycle model on each spatial grid to the physical spatial location of each node.

[0056] Here, after generating the prior prediction set, multi-source observation data collected by the heterogeneous monitoring network within the corresponding time period can be extracted simultaneously, and the multi-source observation data can be organized into actual observation vectors according to node number, spatial location, or predetermined observation order. Let the actual observation vector be... Then, each observation component in the vector corresponds to the observation value output by the ordinary node or the core node within the current assimilation period.

[0057] Since the surface carbon cycle model outputs model state variables for each spatial grid, while the physical nodes in a heterogeneous monitoring network are typically distributed in discrete geographical locations, their spatial representations are not consistent. Therefore, it is also necessary to construct observation operators. Observation operators are used to map state variables in the model's state space to their corresponding positions in the observation space, achieving a one-to-one correspondence between the model output and physical observations. In one implementation, if the objects directly observed by the physical nodes have the same physical meaning as the model's state variables, the observation operators can primarily perform spatial interpolation and spatial mapping. In another implementation, if the physical nodes measure observations that have a physical transformation relationship with the model's state variables, the observation operators can also include corresponding state mapping relationships or physical quantity transformation relationships. Through the above processing, the prior prediction set and the actual observation vectors can be placed within a comparable and fusionable unified expression framework, thereby ensuring that the correspondence between the model's state variables and the physical observation data is clear, dimensionally consistent, and spatially well-defined during the assimilation and update process.

[0058] Subsequently, based on the first or second precision level of each node, an observation error variance matching its hardware physical characteristics is assigned to the corresponding node to construct an observation error covariance matrix with precision heterogeneity.

[0059] In this embodiment, ordinary nodes and core nodes in the heterogeneous monitoring network correspond to different observation accuracy levels. Therefore, not all observation components can be considered equally reliable during the assimilation process. To address this, observation error variance can be assigned to each observation component in the actual observation vector based on the accuracy level and hardware characteristics of each node. For ordinary nodes at the first accuracy level, a relatively large observation error variance can be assigned; for core nodes at the second accuracy level, a relatively small observation error variance can be assigned.

[0060] Let the variance of the observation error corresponding to the ordinary node be... The variance of the observation error corresponding to the core node is Then the error variances corresponding to each observation component can be organized into the observation error covariance matrix according to the order of the actual observation vectors. The observation error covariance matrix constructed in this way can mathematically reflect the differences in the observation reliability of nodes with different accuracy levels, giving high-precision observation data a stronger constraint capability in state correction, while the observation data provided by low-cost ordinary nodes mainly serves to expand the spatial coverage. In this way, the accuracy differences of physical nodes in heterogeneous monitoring networks are directly mapped to the error representation structure of assimilation processing, enabling observation data from different sources to participate in state updates according to their respective reliability levels during the fusion process, thereby improving the rationality and stability of the assimilation results.

[0061] Subsequently, based on the prior forecast set, observation operators, and observation error covariance matrix, the cross covariance between the prior forecast set and the observation space, as well as the forecast error covariance in the observation space, are calculated, and the Kalman gain matrix is ​​solved accordingly.

[0062] After obtaining the prior forecast set, actual observation vectors, observation operators, and observation error covariance matrix, the relative contributions of model predictions and observation information to state updates can be further evaluated. Specifically, the prior forecast set can first be mapped to the observation space using the observation operators, and the cross covariance between the model state space and the observation space, as well as the prediction error covariance of the prior forecast set in the observation space, can be calculated. Based on this, the Kalman gain matrix can be solved using the observation error covariance matrix:

[0063] Equation (2) In the formula, Represents the Kalman gain matrix. Denotes the prior covariance matrix. For the observation operator, Denotes the transpose matrix of the observation operator. Let be the observation error covariance matrix. Where is the prior covariance matrix. It can be calculated based on the sample distribution characteristics of each prior set member in the prior prediction set.

[0064] In the above expression, The Kalman gain matrix is ​​used to characterize the total uncertainty in the observation space, which includes both the uncertainty of prior forecasts and the uncertainty of observation noise. By inverting the total uncertainty and combining it with the cross-covariance, the Kalman gain matrix can be obtained. The Kalman gain matrix is ​​used to determine the degree of influence of model forecasts and actual observations on state updates within the current period. For observation components with smaller observation errors, the corresponding Kalman gain term is usually larger, thus making the observation have a stronger impact on state correction; for observation components with larger observation errors, their update impact is relatively weaker. Through this process, an adaptive trade-off based on error statistics can be achieved between model forecast results and heterogeneous observation results, so that the state update process does not rely on empirical fixed weights, but dynamically determines the correction intensity based on the relative magnitude of forecast and observation errors.

[0065] Next, using the Kalman gain matrix, the observation operator, and the actual observation vector, the state is updated for each prior set member in the prior prediction set, resulting in a posterior set composed of multiple posterior set members.

[0066] In some implementations, to maintain the reasonableness of the ensemble's statistical characteristics, perturbation observations conforming to predetermined statistical characteristics can be introduced into the actual observation vectors to generate observation updates corresponding to each prior ensemble member. Accordingly, the... The state update of each prior set member can be represented as:

[0067] Equation (3) In the formula, Indicates the first A posterior set member, Indicates the first A priori set of members, Indicates the relationship with the first The number of observation updates corresponding to each set member This represents the innovation term in the observation space. This innovation term is used to characterize the deviation between the model's equivalent observations and the actual observations, and the Kalman gain matrix is ​​used to feed this deviation back into the model's state space according to a predetermined error statistical relationship.

[0068] Thus, each prior set member, after being updated, forms a corresponding posterior set member, and all posterior set members together constitute the posterior set: Equation (4) Through the above processing, state correction can be completed while preserving the characteristics of the set representation. This ensures that the assimilation result not only contains the updated optimal state information but also retains the statistical characteristics of the updated state distribution, providing a basis for the subsequent output of posterior estimation results and posterior covariance matrix.

[0069] Furthermore, the ensemble mean of the posterior set is determined as the posterior carbon flux estimate for each spatial grid, and the sample covariance matrix of the posterior set relative to the ensemble mean is calculated to generate the posterior covariance matrix.

[0070] Specifically, the ensemble average of all posterior set members can be taken to obtain the posterior carbon flux estimate for each spatial grid: Equation (5) In the formula, This represents the posterior carbon flux estimation result.

[0071] Simultaneously, the sample covariance matrix of the posterior set can be calculated based on the dispersion of each posterior set member relative to the set mean: Equation (6) In the formula, Let represent the posterior covariance matrix. The main diagonal elements of the posterior covariance matrix characterize the local uncertainty of each spatial grid after observation fusion, while the off-diagonal elements characterize the spatial correlation of residual errors among the spatial grids. This allows the system to output the posterior carbon flux estimation results for each spatial grid, along with the corresponding uncertainty characterization results. This enables the system to not only obtain state estimates but also quantify the reliability of the estimates, which improves both the completeness of the regional carbon flux state representation and the interpretability and credibility of the assimilation results.

[0072] Through the embodiments of this application, the prior state output by the surface carbon cycle model and the multi-source observation data collected by the heterogeneous monitoring network can be deeply coupled and corrected under a unified error statistics framework. In particular, the hardware precision difference between ordinary nodes and core nodes is transformed into a variance-dissimilarity mapping in the observation error covariance matrix, enabling observation information of different precision levels to adaptively match the corresponding constraint weights during assimilation and inversion. As a result, the output posterior carbon flux estimation results not only more closely approximate the true evolution law of carbon flux in the target area in terms of physical mechanism, but the synchronously output posterior covariance matrix also enables a quantitative diagnosis of the reliability of the entire monitoring area, thus providing a rigorous mathematical foundation for subsequent high-precision node selection and densification based on uncertainty characteristics.

[0073] Figure 2A flowchart illustrating an example of determining the spatial heterogeneity risk index of each spatial grid according to an embodiment of this application is shown. Figure 2 As shown in this example, based on the posterior carbon flux estimation results and posterior covariance matrix obtained from data assimilation processing, the local uncertainty, spatial gradient information, and neighborhood distance-weighted uncertainty of each spatial grid can be determined sequentially, and the spatial heterogeneity risk index of each spatial grid can be calculated accordingly. This spatial heterogeneity risk index is used to characterize the monitoring risk of each spatial grid under the current monitoring network configuration.

[0074] like Figure 2 As shown, in step S210, the main diagonal elements of the posterior covariance matrix are extracted, and the main diagonal elements are determined as the local uncertainty corresponding to each spatial grid, so as to characterize the posterior error variance of the corresponding spatial grid.

[0075] In practice, the posterior covariance matrix obtained after data assimilation is used to characterize the error statistics of the carbon flux estimation results for each spatial grid. For this posterior covariance matrix, the elements along its main diagonal correspond to the error variance of the state estimation results for each spatial grid. Therefore, the corresponding elements can be extracted along the main diagonal of the posterior covariance matrix, and the ... The main diagonal elements corresponding to each spatial grid are denoted as . This represents the local uncertainty of the spatial grid.

[0076] Here, local uncertainty is used to characterize the reliability of the carbon flux estimation results for the corresponding spatial grid within the current assimilation period. When the local uncertainty for a spatial grid is high, it indicates that the posterior carbon flux estimation results for that spatial grid still have a large degree of error dispersion; when the local uncertainty is low, it indicates that the state estimation results for that spatial grid are relatively stable. Through this processing, the statistical information in the posterior covariance matrix can be converted into a local risk characterization quantity for a single spatial grid.

[0077] In step S220, for each spatial grid, based on the difference between the posterior carbon flux estimation results of the spatial grid and the adjacent spatial grids in each preset orthogonal coordinate direction, the approximate value of the flux partial derivative in each preset orthogonal coordinate direction is determined, and the carbon flux gradient intensity of the corresponding spatial grid is determined based on the approximate value of the flux partial derivative in each preset orthogonal coordinate direction, as the corresponding spatial gradient information.

[0078] In some implementations, the spatial variability of carbon flux can be characterized by the difference in posterior carbon flux estimates between adjacent spatial grids. Taking a two-dimensional spatial grid as an example, this can be applied to the... For each spatial grid, posterior carbon flux estimates of adjacent spatial grids are extracted along two preset orthogonal coordinate directions. Approximate values ​​of flux partial derivatives in the corresponding directions are calculated based on central difference, forward difference, or backward difference methods. If the spatial grid is represented in three dimensions, differential calculations in the height direction can also be included.

[0079] In one example, let Let be the spatial distribution function of the carbon flux field, and let the two preset orthogonal coordinate directions be respectively direction and By determining the direction, we can obtain an approximate value for the flux partial derivative. and Furthermore, the approximate values ​​of the flux partial derivatives in each orthogonal direction can be combined to obtain the direction-independent carbon flux gradient intensity. For example, the intensity of the first flux gradient can be calculated according to the following formula. Carbon flux gradient intensity of each spatial grid (characterized using L2 norm):

[0080] Equation (7) in, Indicates the first The carbon flux gradient intensity of a spatial grid is physically used to characterize the intensity of spatial variations in carbon flux near the corresponding spatial grid. For example, in areas where forests and farmland meet, wetland edges, or urban green spaces meet built-up areas, the carbon exchange states between adjacent spatial grids may differ significantly, and the corresponding L2 norm of the carbon flux gradient is usually high. Therefore, it avoids judging spatial heterogeneity solely based on flux changes in a single direction, allowing spatial gradient information to more comprehensively reflect the overall intensity of carbon flux variations near the corresponding spatial grid.

[0081] In step S230, the distance attenuation weight is determined based on the physical spatial distance between the corresponding spatial grid and each grid in its local spatial neighborhood, and the local uncertainty of each grid in the local spatial neighborhood is weighted and aggregated using the distance attenuation weight to obtain the neighborhood distance weighted uncertainty of the corresponding spatial grid.

[0082] In some implementations, it is possible to target the first Each spatial grid determines its local spatial neighborhood set. This local neighborhood set can be determined based on preset physical distances, adjacent grid orders, or other spatial adjacency rules. For example, it can be determined by the first... A spatial local neighborhood set is defined by taking a spatial grid as the center and a spatial grid within a predetermined physical distance as the radius. Alternatively, the set can be defined by taking the first spatial grid as the center and a spatial grid within a predetermined physical distance as the radius. A set of several layers of grids that are directly or indirectly adjacent to each other in a spatial grid is defined as a local spatial neighborhood.

[0083] For spatial local neighborhood sets Any grid in According to the first Individual spatial grids and grids The physical spatial distance between them determines the distance attenuation weight. In one implementation, the distance attenuation weight can decrease as the physical spatial distance increases, giving neighboring grids closer to the target grid a greater influence weight and neighboring grids farther away a smaller influence weight. This physical spatial distance can be Euclidean distance or geodesic distance calculated based on geographic coordinates.

[0084] Furthermore, distance decay weights can be utilized. Local neighborhood set The local uncertainties of each grid are weighted and aggregated to obtain the first... The neighborhood distance weighted uncertainty corresponding to each spatial grid.

[0085] In one example, the neighborhood distance-weighted uncertainty can be expressed as: ,in, Represents a local neighborhood set in space Medium grid The local uncertainty. This neighborhood distance-weighted uncertainty is used to characterize the overall uncertainty level of the region surrounding the target spatial grid, enabling the calculation of the spatial heterogeneity risk index to simultaneously consider the state of the target spatial grid itself and the state of the surrounding spatial environment.

[0086] In step S240, the local uncertainty of the corresponding spatial grid is coupled with the carbon flux gradient intensity adjusted by the preset gradient weight, and the coupling result is calculated as a ratio to the neighborhood distance weighted uncertainty, which serves as the neighborhood risk benchmark, to determine the spatial heterogeneity risk index of each spatial grid.

[0087] In one example, the first The spatial heterogeneity risk index of a spatial grid can be calculated using the following formula: Equation (8) In the formula, Indicates the first Spatial heterogeneity risk index of a spatial grid; Indicates the first Local uncertainty of a spatial grid; Indicates the first The carbon flux gradient intensity of each spatial grid; This represents the maximum value of the carbon flux gradient intensity across all spatial grids within the target region; The preset gradient weight adjustment index is used, and ; Indicates the first A set of spatial local neighborhoods of a spatial grid; Indicates the first Individual spatial grid and neighboring grid Distance attenuation weights determined based on physical spatial distance; Represents the neighborhood grid The local uncertainty. By normalizing the carbon flux gradient intensity, the spatial heterogeneity risk index can be made more rigorous in terms of dimensions as a dimensionless relative risk comparison indicator between different spatial grids.

[0088] In the above equation (8), the numerator term is used to characterize the first term. The risk contribution of each spatial grid itself. Specifically, when the first... When a spatial grid exhibits high local uncertainty and a high L2 norm of the carbon flux gradient corresponding to that grid, it indicates that the grid possesses both significant state estimation uncertainty and is located at a point where carbon flux spatial variations are pronounced, thus increasing its spatial heterogeneity risk index. (Adjustment index) This adjustment index is used to adjust the degree of influence of spatial gradient information on the risk index. When it is necessary to increase the sensitivity to spatial transition regions, the adjustment index can be appropriately increased; when it is necessary to reduce the influence of local gradient fluctuations on the risk index, the adjustment index can be appropriately decreased.

[0089] In the above equation (8), the denominator term is used to characterize the first term. The spatial heterogeneity risk index is a benchmark for the neighborhood of a spatial grid. If the overall uncertainty of the neighborhood of a spatial grid is high, it indicates that the overall observation reliability of the region is relatively insufficient, and the relative prominence of a single spatial grid is moderately weakened. If the overall uncertainty of the neighborhood of a spatial grid is low, but the uncertainty and gradient L2 norm of the spatial grid itself are high, then the spatial heterogeneity risk index of the spatial grid increases accordingly. Thus, the spatial heterogeneity risk index can distinguish between two different situations: local prominence risk and overall regional uncertainty increase, making the risk assessment results more suitable for characterizing the distribution of monitoring risks under spatial heterogeneity conditions.

[0090] In one implementation, to avoid the calculation result being abnormally amplified due to excessively small neighborhood distance weighted uncertainty, a constraint of not less than a preset lower limit can be set on the neighborhood distance weighted uncertainty, or a preset regularization parameter can be added to the denominator. This regularization process does not change the comprehensive characterization logic of the spatial heterogeneous risk index for local uncertainty, spatial gradient information, and neighborhood uncertainty, but it can significantly improve the numerical stability of the risk index calculation under extreme data conditions.

[0091] Through the embodiments of this application, the system deeply and jointly characterizes the statistical uncertainty reflected by the posterior covariance matrix and the spatial physical change characteristics reflected by the posterior carbon flux estimation results. The Spatial Heterogeneity Risk Index (SHRI) cleverly uses a relative ratio model to simultaneously consider the absolute uncertainty of the target spatial grid itself, the drastic degree of spatial change in carbon flux, and the baseline uncertainty level of the surrounding neighborhood. This allows the monitoring risk assessment to no longer rely on a single absolute value of error, but to keenly capture the relative risk characteristics of the target spatial grid within the overall spatial structure of the region. Therefore, the resulting Spatial Heterogeneity Risk Index can accurately quantify the monitoring shortcomings of each spatial grid in complex ecological scenarios, and is particularly suitable for accurately identifying high-risk scenarios with strong spatial heterogeneity of carbon flux, such as forest-farmland boundary areas, wetland edge areas, and urban ecological transition zones, where missed measurements are highly likely.

[0092] Figure 3 A flowchart illustrating an example of the expected reduction in the global estimation uncertainty corresponding to the candidate grid in the method according to an embodiment of this application is shown. Figure 3 As shown in this example, the posterior covariance matrix can be localized based on the physical spatial distance between spatial grids. Then, a virtual observation operator is constructed for each candidate grid, and the covariance update result after introducing a second-precision-level observation at the candidate grid is simulated to determine the expected reduction amount corresponding to each candidate grid.

[0093] In step S310, a localized truncation matrix based on the Gaspari-Cohn function is constructed based on the physical spatial distance between each spatial grid and a preset correlation distance threshold.

[0094] It should be noted that, since the number of members in the ensemble is usually limited, the posterior covariance matrix estimated from the ensemble sample may contain some long-range spurious correlations. To reduce the impact of such long-range correlations on the evaluation of candidate grid gains, the covariance matrix can be localized based on the physical spatial distance between the spatial grids. Specifically, for any two spatial grids within the target region, the physical spatial distance between them can be calculated first. The physical spatial distance can be the Euclidean distance based on the coordinates of the grid center point, or the geodesic distance based on geographic coordinates.

[0095] In some implementations, the physical spatial distance can be input into the Gaspari-Cohn function and combined with a preset correlation distance threshold to obtain the distance decay coefficient between corresponding grid pairs. For spatial grids that are close together, the distance decay coefficient can be kept at a high value; for spatial grids that are far apart, the distance decay coefficient can be gradually reduced, or truncated to zero when it exceeds a preset effective range. The distance decay coefficients corresponding to each grid pair form a localized truncation matrix, which is used to limit the effective range of the covariance relationship between different spatial grids, making the covariance characterization more consistent with the spatial proximity constraint. In this way, while preserving the main correlation between adjacent or near-neighbor spatial grids, the impact of weak or spurious correlations at long distances on the simulation update results can be reduced, thus making the calculation of the expected reduction of candidate grids more focused on the error propagation structure consistent with the physical spatial relationship.

[0096] In step S320, a localized truncation operation is performed on the posterior covariance matrix using the localized truncation matrix to generate a localized posterior covariance matrix.

[0097] In practical implementation, let the posterior covariance matrix output by the data assimilation process be... The localized truncation matrix is The localized posterior covariance matrix can then be obtained by multiplying corresponding elements of the matrix. For example, It can be expressed by the following formula:

[0098] Equation (9) In the formula, This represents the Hadamard product operation, which is the element-wise multiplication of two matrices of the same type at corresponding positions. This represents the posterior covariance matrix output by the data assimilation process. This represents the localized posterior covariance matrix.

[0099] Through the aforementioned localization truncation operation, the distant correlations in the posterior covariance matrix that do not match the physical spatial distance can be weakened or truncated, while the main correlations between nearby spatial grids are preserved. Therefore, the resulting localized posterior covariance matrix can serve as the error statistics basis for subsequent virtual observation updates, making the candidate grid benefit assessment more consistent with the actual constraints of spatial correlation.

[0100] In step S330, for each candidate grid, a virtual observation operator is constructed to map the model state variables when the second precision level observation is set at the candidate grid to the corresponding virtual observation space.

[0101] In practice, candidate grids are spatial grids determined based on the spatial heterogeneity risk index meeting preset triggering conditions. For any candidate grid, it can be assumed that a second level of observation capability is introduced at that grid. This introduction can include adding a core node at the candidate grid, or upgrading an existing ordinary node in the candidate grid to a core node. To assess the potential uncertainty changes caused by this action without actually altering the physical monitoring network, a corresponding virtual observation operator can be constructed.

[0102] This virtual observation operator is used to map state variables associated with candidate grids in the model state space to the virtual observation space. In one implementation, if the virtual observation directly corresponds to the carbon flux state variable of the candidate grid, the virtual observation operator can primarily be used to extract the state variable corresponding to that candidate grid. In another implementation, if there is a spatial interpolation, scale transformation, or physical quantity transformation relationship between the virtual observation and the model state variables, the virtual observation operator can include the corresponding state mapping relationship. By constructing the virtual observation operator, the effect of introducing second-precision level observations at the candidate grid can participate in the covariance update simulation in a unified matrix form.

[0103] In step S340, a virtual Kalman gain is generated by combining the preset observation error variance corresponding to the second precision level, the virtual observation operator, and the localized posterior covariance matrix. This gain is used to characterize the update effect on the model state estimation after introducing the second precision level observation at the candidate grid.

[0104] In specific implementation, let the first... The virtual observation operator corresponding to each candidate grid is: The preset observation error variance corresponding to the second accuracy level is The localized posterior covariance matrix is Candidate meshes can then be generated as follows: Corresponding virtual Kalman gain : Equation (10) In the formula, Indicates in the candidate grid When introducing the virtual Kalman gain corresponding to the second level of accuracy observation, Represents the virtual observation operator The transpose of the matrix, This is the preset observation error variance corresponding to the second accuracy level, and .

[0105] In the expression of equation (10), The virtual Kalman gain is used to characterize the total uncertainty in the virtual observation space, which includes the prediction error obtained by mapping the localized posterior covariance matrix and the observation error corresponding to the second-level precision observations. By combining the total uncertainty with the covariance relationship between the model state space and the virtual observation space, the virtual Kalman gain can be obtained. This virtual Kalman gain is used to characterize the update effect on the model state estimate after introducing second-level precision observations into the candidate grid. Through the above processing, the potential update intensity of different candidate grids after introducing high-precision observation capabilities can be compared within a unified error statistics framework, enabling the evaluation of candidate grid gains to simultaneously consider the observation error level and spatial covariance structure.

[0106] In step S350, the localized posterior covariance matrix is ​​simulated and updated using virtual Kalman gain to obtain the virtual updated covariance matrix corresponding to the candidate grid.

[0107] In practical implementation, the covariance change after introducing second-level precision observations can be simulated using virtual Kalman gain without actually deploying the hardware. In one example, the candidate grid... The corresponding virtual updated covariance matrix It can be calculated using the following formula:

[0108] Equation (11) In the formula, Indicates in the candidate grid The virtual updated covariance matrix after introducing second-precision level observations is then introduced. This represents the covariance adjustment resulting from the virtual observation update. Applying this adjustment to the localized posterior covariance matrix yields the corresponding virtual-updated covariance matrix.

[0109] Through the above processing, the covariance simulation results of each candidate grid after introducing the second level of accuracy observation capability can be obtained. These results are used to characterize the potential impact of the candidate grid observation enhancement actions on the uncertainty of the target area estimation, so that different candidate grids have comparable error shrinkage characteristics.

[0110] In step S360, the expected reduction amount corresponding to the candidate grid is determined based on the difference between the localized posterior covariance matrix and the virtual updated covariance matrix in terms of global uncertainty measure, wherein the global uncertainty measure is characterized by the trace of the covariance matrix.

[0111] Here, the trace of the covariance matrix can be represented by the sum of the elements on the main diagonal of the matrix, used to characterize the overall level of local error variance of each spatial grid within the target region. Let... To represent the matrix trace operation, the first... Expected reduction amount corresponding to each candidate grid It can be calculated using the following formula:

[0112] Equation (12) In the formula, Indicates the first The expected reduction amount corresponding to each candidate grid The trace of the localized posterior covariance matrix is ​​represented. This represents the trace of the covariance matrix after the virtual update. In the above expression, if... A larger value indicates that the candidate grid is larger. After introducing observations of the second precision level, the localized posterior covariance matrix shows a significant decrease in the overall variance level; if A smaller value indicates that the observation enhancement actions corresponding to the candidate grid have a relatively limited effect on improving the overall uncertainty. This scalarization process allows the changes in the covariance matrix corresponding to different candidate grids to be converted into comparable benefit evaluation metrics.

[0113] Through the embodiments of this application, before adding core nodes or upgrading ordinary nodes at candidate grids, the covariance change after introducing second-level precision observations can be simulated based on the current posterior covariance matrix. A localized truncation matrix is ​​used to reduce the impact of long-range spurious correlations on the simulation results. The virtual observation operator and virtual Kalman gain are used to describe the state update relationship of the candidate grid after introducing high-precision observation capabilities. Expected Variance Reduction (EVR) is used to convert the covariance change results into comparable scalar evaluation values. This helps to quantitatively compare the observation enhancement value of different candidate grids before resource investment, ensuring that the selection of candidate grids considers not only their current monitoring risks but also their potential impact on the overall estimation uncertainty after introducing second-level precision observations.

[0114] In some examples of embodiments of this application, the heterogeneous node encryption strategy may include not only core node addition instructions but also existing ordinary node capability upgrade instructions. For candidate grids where ordinary nodes have already been deployed, a comprehensive evaluation can be conducted based on the long-term observation deviation of the ordinary node, the uncertainty level of the spatial grid, and the ecological scene attributes to determine whether to use the ordinary node as the upgrade target node. This allows for the improvement of the observation accuracy level of the corresponding spatial grid without changing the existing site locations. Regarding the decision-making process for generating the existing ordinary node capability upgrade instructions included in the heterogeneous node encryption strategy, in some examples of embodiments of this application, firstly, for each ordinary node deployed in the candidate grid, the assimilation residual sequence of the ordinary node within a consecutive preset number of data assimilation processing cycles is extracted. The assimilation residual of a single data assimilation processing cycle is the difference between the actual observation data collected by the ordinary node and the forecast data output by the surface carbon cycle model at the corresponding physical spatial location of the ordinary node.

[0115] It should be noted that the observation bias within a single assimilation cycle may be affected by instantaneous meteorological disturbances, instrument random noise, communication jitter, or short-term local disturbances. Judging whether to upgrade a normal node based solely on a single observation bias may lead to decisions influenced by chance. Therefore, assimilation residuals can be extracted over multiple consecutive assimilation cycles to form a sequence of assimilation residuals for the corresponding normal node. Let the number of consecutive assimilation cycles be... For the first The nth ordinary node, which is in the nth... The assimilation residual within one assimilation period can be expressed as:

[0116] Equation (13) In the formula, Indicates the first The ordinary node at the ... Assimilation residuals within one assimilation cycle This indicates that the ordinary node is at the th . Actual observation data collected within one assimilation period The surface carbon cycle model is in the first The ensemble average of the prior predicted states output within each assimilation period (i.e., the highest prior estimate). This represents the observation mapping operator corresponding to the physical spatial location of the ordinary node. This represents the equivalent forecast data of the model at the physical location corresponding to the common node. Through the above processing, the observation deviation of the common node can be obtained over multiple consecutive assimilation periods. This assimilation residual sequence can reflect the persistence and volatility of the deviation between the observation results of the common node and the model forecast results, and is more suitable for node upgrade judgment than a single residual.

[0117] Then, the variance of the assimilation residual sequence for each ordinary node is calculated as the residual variance characterizing the degree of deviation of the observations of the corresponding ordinary node.

[0118] In some implementations, the first A normal node in continuous The assimilation residual sequences within each assimilation cycle are statistically calculated to obtain the residual variance. For example, the residual variance can be calculated as follows:

[0119] Equation (14) In the formula, Indicates the first The residual variance of each ordinary node This indicates that the ordinary node is in The average value of the assimilation residuals within each assimilation cycle.

[0120] Residual variance characterizes the fluctuation of observational deviations at ordinary nodes over consecutive assimilation periods. A high residual variance for an ordinary node indicates a strong periodic deviation or fluctuation between the actual observations and the model's equivalent forecasts. This may reflect complex local carbon flux variations within the spatial grid containing the node, or insufficient constraint of the node's first-level accuracy observation capability on regional state changes. By introducing residual variance, the upgrade judgment of ordinary nodes can be made not only dependent on spatial location or single-instance errors, but also further consider the observational stability of the node over time.

[0121] Next, the ecological scene weight of the spatial grid where each ordinary node is located is obtained. The ecological scene weight is quantitatively assigned based on whether the corresponding spatial grid is located at the boundary between carbon source and carbon sink and whether it belongs to the preset ecological sensitive area.

[0122] In practice, the ecological scenario attributes of each ordinary node's spatial grid can be determined based on land use classification data, vegetation cover data, ecological function zoning data, topographic data, or other geospatial data of the target area. For example, when an ordinary node is located in an area where forests and farmland meet, wetland edges, urban green spaces and construction land meet, or other areas where carbon source and carbon sink alternation is relatively obvious, the spatial grid where the ordinary node is located can be assigned a higher ecological scenario weight; when an ordinary node is located in an area where carbon flux spatial changes are relatively stable, a relatively lower ecological scenario weight can be assigned.

[0123] Ecological scenario weights are used to incorporate spatial scenario importance information into upgrade assessments. This weighting allows for the simultaneous consideration of data assimilation results and the ecological spatial attributes of the target area in ordinary node upgrade decisions, avoiding reliance solely on residual statistical characteristics to determine upgrade targets. This approach helps make upgrade assessments more aligned with the observation needs of carbon sequestration monitoring operations for ecological transition zones and sensitive areas.

[0124] Subsequently, combining the local uncertainty of the spatial grid where each ordinary node is located, the residual variance of the ordinary node, and the ecological scenario weight of the spatial grid where the ordinary node is located, a node upgrade evaluation index is calculated for each ordinary node using a normalized weighted fusion method. The node upgrade evaluation index is positively correlated with the local uncertainty of the corresponding spatial grid where the ordinary node is located, positively correlated with the residual variance of the corresponding ordinary node, and positively correlated with the ecological scenario weight of the corresponding spatial grid where the ordinary node is located. Furthermore, the local uncertainty and residual variance are normalized using the maximum value of the corresponding index within the target area as the benchmark.

[0125] For example, the first The node upgrade evaluation index for a regular node can be calculated as follows: Equation (15) In the formula, Indicates that for the first The node upgrade evaluation index for a regular node. For the first Local uncertainty of the spatial grid where a typical node is located. The maximum local uncertainty of all spatial grids within the target region. For the first The residual variance of each ordinary node The maximum residual variance of all ordinary nodes within the target area. For the first The ecological scene weight of the spatial grid where each ordinary node is located. , and This is the preset adjustment coefficient.

[0126] In the above expression (15), local uncertainty is used to reflect the uncertainty of the state estimation of the spatial grid where the ordinary node is located, residual variance is used to reflect the fluctuation of the observation deviation of the ordinary node in the continuous assimilation period, and ecological scene weight is used to reflect the importance of the ecological scene of the spatial grid where the ordinary node is located. Since the dimensions and numerical ranges of local uncertainty and residual variance may be different, they can be normalized by using the maximum value of the corresponding index, so that different indices can participate in weighted fusion under a relatively uniform scale. The adjustment coefficient is used to adjust the influence of each factor in the node upgrade evaluation, and can be preset according to the degree of attention of the monitoring task to uncertainty, observation stability and ecological scene attributes.

[0127] Through the above processing, the spatial estimation uncertainty of the location of a common node, the observation deviation fluctuation of the common node, and the spatial scene attributes can be combined into a single evaluation metric. This evaluation metric can be used to compare the upgrade priority of different common nodes.

[0128] Furthermore, when the target ordinary node's node upgrade evaluation index When the preset evaluation threshold is exceeded, the target ordinary node is identified as the upgrade target node, and an upgrade instruction for the existing ordinary node capabilities is generated for the upgrade target node.

[0129] In some implementations, the node upgrade evaluation index of each ordinary node can be compared with a preset evaluation threshold. If the node upgrade evaluation index of an ordinary node is greater than the evaluation threshold, it indicates that the ordinary node has a high upgrade demand under the comprehensive evaluation of local uncertainty, residual variance, and ecological scenario weight, and can be identified as an upgrade target node. The existing ordinary node capability upgrade instruction can include the node identifier of the upgrade target node, the spatial grid identifier, the target accuracy level, the hardware module information to be replaced or enhanced, and access configuration parameters.

[0130] At the physical execution level, upgrading existing ordinary nodes can include replacing them with higher-precision sensing modules, adding high-precision auxiliary observation modules, improving sampling accuracy or sampling frequency, enhancing local data processing capabilities, or integrating the ordinary node into the calibration and quality control process corresponding to the second accuracy level. Through such upgrades, the observation capabilities of the corresponding spatial grid can be improved while reusing existing site locations, power supply, and communication conditions.

[0131] This application provides an embodiment that allows for the evaluation and upgrade of existing ordinary nodes deployed in candidate grids, generating capability upgrade instructions based on the evaluation results. The evaluation process simultaneously considers the local uncertainty of the spatial grid where the ordinary node resides, the residual variance of the ordinary node itself, and the ecological scenario weight of the spatial grid, ensuring that node upgrade decisions do not rely on a single indicator. This helps to select more suitable nodes for upgrade from existing ordinary nodes and provides quantifiable criteria for upgrading ordinary nodes from a first precision level to a second precision level. Furthermore, performing capability upgrades on existing sites facilitates the reuse of deployed site locations, power supply conditions, and communication conditions, thereby reducing the engineering implementation pressure brought by adding new sites to a certain extent.

[0132] Figure 4 A flowchart illustrating an example of generating a heterogeneous node encryption strategy through multi-factor node optimization decision-making in the method according to an embodiment of this application is shown. Figure 4 As shown in this example, the target set of grids that meet the resource budget constraints can be determined based on the spatial heterogeneity risk index, expected reduction amount, unit execution cost and unit operating energy consumption of the candidate grids, and a heterogeneous node encryption strategy can be generated accordingly.

[0133] In step S410, for each candidate grid, the unit execution cost and unit operating energy consumption of that candidate grid are determined according to the candidate instruction type corresponding to that candidate grid. The candidate instruction types include core node addition instructions and existing ordinary node capability upgrade instructions. Core node addition instructions correspond to equipment purchase cost and total operating energy consumption, while existing ordinary node capability upgrade instructions correspond to hardware module replacement cost and incremental energy consumption cost.

[0134] In some implementations, if the candidate instruction type corresponding to the candidate grid is a new instruction for the core node, the unit execution cost of the candidate grid may include the core node equipment purchase cost, installation cost, access cost, calibration cost, and necessary on-site implementation cost; its unit operating energy consumption may include the energy consumption of the core node sensing equipment, data acquisition unit, and communication unit during operation. If the candidate instruction type corresponding to the candidate grid is an upgrade instruction for the capabilities of existing ordinary nodes, the unit execution cost of the candidate grid may include the sensor module replacement cost, auxiliary hardware installation cost, and on-site debugging cost; its unit operating energy consumption can be determined based on the energy consumption difference between the upgraded equipment and the original ordinary node equipment.

[0135] By determining the cost and energy consumption of different candidate instruction types, the operation of adding core nodes and upgrading ordinary nodes can be compared under the same resource evaluation framework. This helps to reflect the differences in the engineering implementation cost and operating load of different node adjustment methods, so that node selection decisions are not based solely on risk or benefit indicators.

[0136] In step S420, a node selection optimization model is constructed with the objective of maximizing the weighted comprehensive return and satisfying the upper limits of the total hardware budget and the total energy consumption budget. The weighted comprehensive return is determined by normalizing the spatial heterogeneity risk index and expected reduction amount of each candidate grid according to preset weights, and both the spatial heterogeneity risk index and the expected reduction amount contribute positively to the weighted comprehensive return.

[0137] In practical implementation, the spatial heterogeneity risk index and expected reduction of candidate grids can be used as benefit evaluation factors in the node selection optimization model. The spatial heterogeneity risk index characterizes the monitoring risk level of the candidate grid, while the expected reduction characterizes the potential reduction in global estimation uncertainty after introducing second-level precision observations into the candidate grid. Since the spatial heterogeneity risk index and expected reduction may have different dimensions and numerical ranges, they can be normalized separately and then weighted and fused using preset weights.

[0138] In one example, the node selection optimization model can be represented as: Equation (16) And satisfy:

[0139] as well as:

[0140] In the formula, This represents the set of candidate target meshes to be optimized (which becomes the final selected target mesh set after the solution is completed); Indicates the first Spatial heterogeneity risk index of candidate grids; Indicates the first The expected reduction in the number of candidate grids; This represents the maximum value of the spatial heterogeneity risk index among all candidate grids; This represents the maximum expected reduction among all candidate grids; and These represent the preset risk tolerance weight and return preference weight, respectively; Indicates the first The individual execution cost of each candidate grid; Indicates the first Energy consumption of each candidate grid unit during operation; Indicates the upper limit of the total hardware budget; This indicates the upper limit of the total energy consumption budget.

[0141] In the above model, the objective function is used to evaluate the overall benefits of the candidate grid set, while the constraints are used to limit the cumulative execution cost and cumulative operating energy consumption of the target grid set. This model allows for the simultaneous consideration of the candidate grid's risk level, potential uncertainty reduction effect, and resource input boundaries in the same evaluation process.

[0142] In step S430, a weighted greedy algorithm is used to solve the node selection optimization model, and the unit resource comprehensive benefit score of each candidate grid is calculated based on the weighted comprehensive benefit of each candidate grid and the corresponding unit execution cost and unit operating energy consumption.

[0143] In practical implementation, the weighted comprehensive benefit of each candidate grid can be calculated first based on the normalized spatial heterogeneity risk index and the normalized expected reduction. Further, a unit resource comprehensive benefit score can be calculated by combining the unit execution cost and unit operating energy consumption of the candidate grid. This score represents the comprehensive benefit level of the candidate grid under unit resource consumption conditions.

[0144] In one example, the first The unit resource comprehensive benefit score of each candidate grid can be expressed as: Equation (17) In the formula, Indicates the first The comprehensive resource benefit score per unit of each candidate grid; and These represent the normalized adjustment weights corresponding to execution cost and operating energy consumption, respectively. The numerator represents the weighted overall return of the candidate grid, and the denominator represents the resource consumption level of the candidate grid. This scoring method transforms the differences in risk-reward and resource consumption among different candidate grids into ranking evaluation metrics.

[0145] It should be noted that the scoring process described in equation (17) above is only one example. In other implementations, other scoring methods that can simultaneously reflect the weighted comprehensive benefits, individual execution costs, and individual operating energy consumption can also be used, as long as they can be used to compare the selection priority of each candidate grid under resource constraints.

[0146] In step S440, under the condition of satisfying the upper limit of the total hardware budget and the upper limit of the total energy consumption budget, each candidate grid is iteratively selected in descending order of the comprehensive benefit score per unit resource until the addition of the next candidate grid will cause the cumulative cost or cumulative energy consumption to exceed the corresponding upper limit constraint, at which point the iterative selection stops.

[0147] In some implementations, candidate grids can be first sorted in descending order according to their unit resource comprehensive benefit score to form a candidate grid sequence. The system then selects candidate grids sequentially from this sequence and adds them to the target grid set. Each time a candidate grid is added, the cumulative execution cost and cumulative operating energy consumption corresponding to the current target grid set are updated.

[0148] If adding the next candidate grid would cause the cumulative execution cost to exceed the total hardware budget limit, or cause the cumulative operating energy consumption to exceed the total energy consumption budget limit, then the iterative selection process stops. The resulting target grid set satisfies the preset resource budget constraints, and its members are selected according to the priority order of their comprehensive unit resource benefit score. This process allows for candidate grid selection under relatively controllable computational complexity.

[0149] In step S450, the target mesh set selected when the iteration stops and the candidate instruction type corresponding to each target mesh are output to generate a heterogeneous node encryption strategy.

[0150] In some implementations, each target grid in the target grid set corresponds to a candidate instruction type. If the target grid corresponds to a core node addition instruction, the heterogeneous node encryption strategy may include information such as the spatial location of the target grid, the type of core node to be added, deployment requirements, and access parameters. If the target grid corresponds to an existing ordinary node capability upgrade instruction, the heterogeneous node encryption strategy may include information such as the ordinary node identifier, target accuracy level, hardware module to be upgraded, and configuration update parameters. By outputting the target grid set and its corresponding candidate instruction types, the node selection optimization results can be converted into executable strategy content. This strategy can simultaneously cover both core node addition and ordinary node upgrade actions, thus giving the physical deployment topology adjustment and node capability configuration adjustment of the heterogeneous monitoring network a clear execution target.

[0151] Through the embodiments of this application, execution cost and operational energy consumption constraints can be introduced on the basis of spatial heterogeneity risk index and expected reduction amount to comprehensively optimize candidate grids. This ensures that the selection of candidate grids not only considers the monitoring risk and potential uncertainty reduction effect, but also the implementation cost and operational load corresponding to different candidate instruction types. Furthermore, through a unit resource comprehensive benefit scoring and iterative selection process, multiple candidate grids can be transformed into a set of target grids that meet budget constraints, and the node encryption actions corresponding to each target grid can be clearly defined. This helps to form a more suitable heterogeneous node encryption strategy under resource-constrained conditions.

[0152] In some examples of embodiments of this application, after outputting the instructions indicated by the heterogeneous node encryption strategy and triggering the corresponding spatial grid to perform core node addition deployment operations and / or ordinary node upgrade operations, it can be further determined whether the underlying hardware at the spatial grid can complete deployment and generate valid data return within a predetermined time. If the underlying hardware deployment or upgrade process takes a long time, temporary mobile nodes can be scheduled to perform supplementary observations of the spatial grid before the fixed nodes are connected, so as to reduce the impact of insufficient observation data on assimilation processing during the implementation delay.

[0153] First, for each spatial grid that triggers a new deployment operation for a core node or an upgrade operation for a regular node, the estimated implementation delay time from receiving the execution command to the completion of the underlying hardware deployment at that spatial grid and the generation of valid backhaul data is calculated based on the terrain accessibility and hardware engineering cycle of that spatial grid.

[0154] Here, terrain accessibility can be determined based on slope, road connectivity, vegetation cover density, transportation conditions, or other spatial characteristics that affect the difficulty of reaching the site; the hardware engineering cycle can be determined based on the time required for the installation, power supply, communication access, and equipment calibration of new core nodes, or the time required for module replacement, debugging, and access during the upgrade of ordinary nodes. Through the above processing, the differences in engineering implementation across different spatial grids can be transformed into time parameters that can be used for judgment.

[0155] When the estimated implementation delay exceeds the preset blind zone safety delay threshold, the three-dimensional spatial coordinates of the spatial grid are determined, and a temporary spatial scheduling instruction containing the three-dimensional spatial coordinates is generated.

[0156] Here, after obtaining the estimated implementation delay duration, it can be compared with a preset blind zone safety delay threshold. This threshold can be pre-set based on the target area's carbon flux change frequency, assimilation cycle, monitoring task's data continuity requirements, and regional ecological fluctuation characteristics. When the estimated implementation delay duration exceeds the blind zone safety delay threshold, it indicates that there may be a time interval requiring supplementary observations before the fixed hardware is connected to the spatial grid. In this case, the three-dimensional spatial coordinates of the spatial grid can be determined, and a temporary spatial scheduling command containing the three-dimensional spatial coordinates can be generated. The three-dimensional spatial coordinates can be determined based on the spatial grid's planar position, elevation data, target observation height, or spatial constraints to provide scheduling locations for temporary mobile nodes.

[0157] Subsequently, based on temporary space scheduling instructions, a swarm of UAVs carrying airborne carbon flux sensing payloads matching the second accuracy level were scheduled as temporary mobile nodes to fly to the target airspace corresponding to the three-dimensional spatial coordinates to perform temporary high-frequency data acquisition.

[0158] Here, airborne carbon flux sensing payloads may include high-precision carbon dioxide concentration sensors, micro-meteorological sensors, gas sampling modules, or other airborne observation equipment capable of reflecting carbon flux-related states. In one implementation, a swarm of UAVs can perform fixed-point hovering observations within the target airspace; in another implementation, they can also conduct short-term cruise observations over the target grid and its adjacent areas according to a preset route. This method allows for the acquisition of temporary observation data of the target area before the deployment of fixed nodes is completed.

[0159] Next, the temporary high-frequency data collected by the temporary mobile node is input into the preset surface carbon cycle model in real time as new multi-source observation data to perform data assimilation processing, and a dynamic virtual observation operator is constructed based on the spatial location of the temporary mobile node in the target airspace.

[0160] It should be noted that since the spatial position of a UAV as a temporary mobile node may change over time, a dynamic virtual observation operator can be constructed based on the spatial position of the temporary mobile node in the target airspace. Specifically, the dynamic virtual observation operator can be determined based on the spatial coordinates, flight altitude, observation field of view, airflow conditions, and observation payload type of the temporary mobile node at each sampling time. This dynamic virtual observation operator is used to map the state variables of the surface carbon cycle model on each spatial grid to the observation space at the location of the temporary mobile node. If the airborne observation data is atmospheric concentration data and the model state variable is surface carbon flux, the dynamic virtual observation operator can also include mapping relationships related to atmospheric transport, mixing, or source region footprint to achieve a correspondence between the model state variables and the airborne observations.

[0161] Then, based on temporary high-frequency data and dynamic virtual observation operators, compensatory data assimilation updates are performed to update the posterior carbon flux estimation results and posterior covariance matrix of each spatial grid until the underlying hardware deployment at that spatial grid is completed and connected to the heterogeneous monitoring network.

[0162] Here, after obtaining temporary high-frequency data and dynamic virtual observation operators, the temporary high-frequency data collected by the temporary mobile node can be input into the preset surface carbon cycle model in real time or near real time as multi-source observation data, and compensatory data assimilation updates can be performed. In one example, for sampling times that include temporary mobile node observations, the state update can be performed according to the following formula:

[0163] Equation (18) In the formula, Indicates time The posterior carbon flux state estimation results. Indicates time The prior state estimation results of carbon flux. Indicates the temporary mobile node at time Collected temporary high-frequency data, Indicates time The corresponding dynamic virtual observation operator, This represents the Kalman gain calculated based on the dynamic virtual observation operator and the variance of the observation error of the temporary moving node.

[0164] In the expression of the above formula (18), Used to characterize the difference between observations from temporary mobile nodes and model-equivalent observations. This is used to feed the difference back to the model state space. Through this compensatory data assimilation update, data collected by temporary mobile nodes can be incorporated into the current assimilation process even when fixed hardware has not yet been connected, so as to update the posterior carbon flux estimation results and posterior covariance matrix of each spatial grid.

[0165] It should be noted that in the specific implementation framework based on ensemble Kalman filtering (EnKF), the update process shown in equation (18) above is executed independently for each set member. That is, Corresponding to each independent prior set member, and In actual computation, it is processed as a set of perturbed observations superimposed with the corresponding observation error distribution, thereby generating a new posterior set after the compensatory data assimilation update is completed.

[0166] In practice, compensatory data assimilation updates can continue until the underlying hardware deployment at that spatial grid is completed and connected to the heterogeneous monitoring network. Once the fixed core node or upgraded ordinary node begins to generate valid data feedback, the temporary spatial scheduling of that spatial grid can be stopped, and the temporary mobile node can be removed from the supplementary observation task corresponding to that spatial grid. Depending on the task requirements, the temporary mobile node can return to its standby position or be scheduled to other spatial grids that require supplementary observations.

[0167] Through the embodiments of this application, temporary mobile nodes can be used to supplement the observation data of the target spatial grid during the period when fixed hardware deployment or upgrades are not yet completed. The dynamic virtual observation operator enables the spatial position changes of the temporary mobile nodes to correspond to the spatial state expression of the surface carbon cycle model, which helps to incorporate temporary high-frequency observation data into the data assimilation process. Therefore, it can alleviate the problem of insufficient observation data during the implementation delay period to a certain extent, and provide temporary observation support for the continuous updating of posterior carbon flux estimation results and posterior covariance matrix.

[0168] Figure 5 A schematic diagram illustrating the system operation mechanism of an example heterogeneous carbon sink node encryption method based on assimilation inversion feedback according to an embodiment of this application is shown.

[0169] like Figure 5 As shown, this application establishes a closed-loop mechanism for the dynamic evolution of the front-end monitoring network driven by back-end data assimilation and inversion. First, the system front-end acquires input information such as observation data from ordinary nodes, high-frequency observation data from core nodes (e.g., flux tower data), and remote sensing and meteorological forcing data. Ordinary node observations and meteorological forcing data are input into a surface carbon cycle model for forward physical simulation to generate prior forecasts. Subsequently, combining high-precision actual observations provided by core nodes, multi-source data fusion is performed within an ensemble Kalman filter (EnKF) assimilation framework, thereby outputting updated carbon flux state estimation results and a posterior covariance matrix characterizing the spatial error distribution of the system to the subsequent stages.

[0170] Next, based on the output posterior covariance matrix, the system proceeds to the uncertainty and risk analysis stage. It quantitatively calculates the Spatial Heterogeneous Risk Index (SHRI) for each grid to accurately identify high-risk observation blind spots and pre-calculates the Expected Emission Reduction (EVR) to assess the global information gain of potential node deployments. After obtaining the aforementioned dual-dimensional quantitative indicators of risk and benefit, the system combines the input hardware and energy consumption budget constraints as boundary conditions to trigger and initiate a weighted greedy algorithm for multi-objective combined optimization decision-making.

[0171] Finally, based on the decision results of the weighted greedy algorithm, the system outputs operational instructions for upgrading the capabilities of heterogeneous nodes and deploying encrypted arrays for the preferred target grid. As these instructions are implemented in the physical world, the overall estimation uncertainty of high-risk areas is substantially reduced. Simultaneously, through... Figure 5 The dynamic evolution of the monitoring network at the bottom and the feedback loop of observation updates, along with the reconstructed enhanced heterogeneous monitoring network, will initiate a new round of higher-quality observation data collection and feed it back to the system front end. Thus, the entire carbon sink monitoring system breaks through the limitations of static deployment, forming an intelligent closed loop that continuously adapts and optimizes itself in response to changes in environmental heterogeneity.

[0172] To verify the effectiveness of the proposed method in complex spatiotemporal heterogeneous scenarios, a mesoscale regional carbon sink simulation scenario was constructed. The area of ​​the target simulation region was set to be... Spatial resolution is divided into This means it contains 2500 independent spatial grid cells. In terms of environmental drivers, the system utilizes a high-precision regional atmospheric chemical transport model (WRF-Chem) to generate realistic net ecosystem exchange (NEE, unit: The background evolution field is then superimposed with spatially correlated Gaussian red noise to simulate the real physical disturbances caused by complex terrain undulations and micro-meteorological conditions.

[0173] Under the initial monitoring network configuration, a uniform distribution strategy was adopted to deploy 30 ordinary nodes and 5 core nodes in the target area, and the ensemble Kalman filter (EnKF) algorithm was used to fuse multi-source observation data with an assimilation step size of 1 hour. In order to quantitatively evaluate the performance advantages of the "heterogeneous node encryption strategy" generated in this embodiment, two comparative baselines were set in parallel: the first is the static uniform strategy (baseline A), which maintains the initial physical topology unchanged and continues to perform assimilation and inversion without increasing any hardware investment; the second is the random incremental strategy (baseline B), which, under the same total hardware and total energy consumption budget constraints as in this embodiment, randomly selects grids across the entire domain to perform node additions or capability upgrades.

[0174] In the simulation scenario described above, the system extracts the mean of the global posterior uncertainty, the local maximum posterior variance characterizing the risk of extreme blind zones, and the unit cost efficiency as core evaluation indicators. To eliminate the influence of random errors, the above scheme and its baseline were subjected to multiple independent Monte Carlo experiments under the same initial conditions to verify the robustness and accuracy improvement effect of the method in this embodiment under different perturbation conditions.

[0175] Figure 6 This diagram illustrates a comparative simulation of the evolution trend of posterior uncertainty under different node encryption strategies. The simulation uses a dual Y-axis structure: the left Y-axis represents the mean global posterior uncertainty of the target region (solid line), and the right Y-axis represents the local maximum posterior variance (dashed line), reflecting the risk of extreme observation blind spots. Transparent bands represent the 1-standard-deviation confidence intervals under multiple independent Monte Carlo experiments. The diagram visually compares the uncertainty convergence and extreme value suppression effects of baseline A (static strategy, maintaining the initial network topology), baseline B (random incremental strategy, randomly adding nodes under the same budget), and the proposed scheme (adaptive heterogeneous encryption strategy) in consecutive decision rounds.

[0176] like Figure 6 In the comparison, after the initial few assimilation periods, the global uncertainty of baseline A rapidly enters a convergence bottleneck and stops decreasing, and the local maximum posterior variance, representing the observation blind zone, oscillates at a high level (approximately). This indicates that static networks cannot cope with dynamic physical disturbances caused by complex terrain and micro-weather conditions. Although baseline B has continuously invested hardware resources, its uncertainty convergence efficiency is very low due to the severe dispersion of resources caused by random deployment, and its improvement effect on local high-risk blind areas is limited.

[0177] In contrast, the adaptive heterogeneous encryption scheme proposed in this embodiment exhibits significant convergence capability and a "peak-shaving" effect for high-risk blind spots. After several rounds of decision feedback (e.g., after the 8th round of decision in the figure), the mean posterior uncertainty across the entire domain experiences a significant and precipitous drop, rapidly falling below [a certain threshold]. More importantly, by introducing the Spatial Heterogeneous Risk Index (SHRI) as a risk assessment criterion, the peak of extreme uncertainty (dashed line) was quickly and accurately suppressed. The experimental results objectively demonstrate that this scheme not only maximizes the overall accuracy of carbon sink monitoring within a limited budget, but also possesses excellent engineering robustness in eliminating local "high-risk blind spots" in heterogeneous networks.

[0178] Figure 7 This diagram illustrates an example of the spatial topology evolution and deployment distribution of a heterogeneous node encryption strategy generated using the method of this application within a target area. Part (a) of Figure 7 shows the main diagram of the node deployment distribution. Figure 7 Part (b) is a schematic diagram of multi-layer spatial data overlay, which intuitively shows the final physical network topology distribution driven by the algorithm. The bottom layer of the diagram uses contour lines to show the actual carbon flux gradient distribution caused by topographic relief and ecological type boundaries; the middle layer uses scatter plots of different shapes (squares represent ordinary nodes, and pentagrams represent core nodes) to show the physical deployment coordinates of heterogeneous nodes; and a heat map representing the spatial distribution of residual uncertainty is overlaid in high-risk dense areas.

[0179] Combination Figure 7 The deployment distribution shown in part (a) of this embodiment, driven by the heterogeneous node encryption strategy, does not adopt the traditional grid-based even distribution mode, but exhibits a high degree of "physical boundary clustering." In the southern plains where carbon flux is relatively stable, due to its small spatial gradient and ease of smooth deduction through assimilation algorithms, the system retains only sparse ordinary nodes to maintain background observations. However, in the northern hilly area at the junction with the built-up area, because this region is an ecologically sensitive area with drastic alternation between carbon sources and sinks, the spatial flux gradient is extremely large, and it is affected by the spatial heterogeneity risk index (especially the gradient adjustment index) in this scheme. Driven by the algorithm, the system spontaneously allocates valuable hardware resources to this area, densely deploying and upgrading a large number of high-precision core nodes. This simulation result objectively demonstrates that the method proposed in this application can achieve precise targeted monitoring of high-gradient, high-risk areas under strict resource budget constraints, based on the laws of space physical evolution, thereby effectively optimizing the overall performance and engineering cost-effectiveness of heterogeneous monitoring networks.

[0180] Figure 8This diagram illustrates a comparison of Pareto fronts and a systems engineering boundary analysis of different node encryption decision-making methods in terms of cost-effectiveness. The horizontal axis of the simulation scatter plot represents the normalized computational cost of a single decision, and the vertical axis represents the expected efficiency reduction (EVR). The diagram shows the efficiency distributions of the baseline C (traditional pure variance strategy, i.e., gray data points) and the proposed scheme in this embodiment (weighted greedy decision-making method, i.e., blue data points), and fits their respective Pareto front curves.

[0181] like Figure 8 As shown in the overall performance quadrant distribution, under the same single-decision computational cost, the Pareto front of the method in this embodiment is significantly higher than that of the traditional pure variance strategy. This demonstrates that the proposed scheme has a significant performance advantage in comprehensively suppressing carbon flux uncertainty due to the introduction of a multidimensional risk index and localized expected return assessment.

[0182] at the same time, Figure 8 It also reveals the physical and computational boundary characteristics of large-scale carbon sink monitoring networks under extreme engineering scenarios. Firstly, in scenarios where the system approaches high dimensionality (e.g., the number of grid cells...), At high node densities or with high dimensionality, the computation time of standard algorithms exhibits a non-linear surge due to the curse of dimensionality in the posterior covariance matrix (as shown at the upper end of the curve in the figure). To overcome this computational bottleneck, this application introduces a localization truncation matrix based on the Gaspari-Cohn function in a preferred embodiment. By forcibly eliminating far-end spurious correlations at the mathematical level, the non-linear expansion of computational overhead is effectively curbed.

[0183] Secondly Figure 8 The shaded area in the lower left corner indicates that in practical engineering applications, even when the computational cost of system decision-making approaches a minimum, the "observation blind zone delay before hardware deployment" still objectively exists and is unavoidable in the real physical world. Due to the objective engineering implementation lag caused by sensor capability upgrades or encrypted deployments, if relying solely on static ground nodes, the system will face irreversible gaps in observation data during construction. Based on a profound understanding of this engineering lag boundary, this application further introduces a temporary mobile compensation mechanism for UAV swarms based on airborne carbon flux sensing payloads in the preferred embodiment. Through the rapid high-frequency replacement of the UAV swarm and the real-time assimilation of the dynamic virtual observation operator, this scheme effectively bridges the physical time delay defect represented by the shaded area in the figure, thereby constructing a monitoring closed loop that can effectively optimize both computational efficiency and physical time.

[0184] To address the high-risk observation blind spots and inefficient resource allocation issues caused by the static deployment of existing carbon sink monitoring networks, this application proposes a heterogeneous carbon sink node densification method based on assimilation and inversion feedback. This method breaks down the disconnect between "front-end monitoring" and "back-end inversion," constructing a deep closed-loop mechanism of "data assimilation—error diagnosis—intelligent scheduling." By extracting the posterior covariance within an ensemble Kalman filter framework, this method constructs a Spatial Heterogeneous Risk Index (SHRI) that integrates statistical uncertainty and dynamic spatial gradient, and accurately predicts the expected reduction (EVR) based on the Gaspari-Cohn localization truncation matrix. This mechanism enables the system to break away from the traditional blind equalization strategy. Under strict dual budget constraints of total hardware and total energy consumption, it adaptively targets and allocates heterogeneous hardware resources (including new deployments of core nodes and upgrades of existing ordinary nodes) to the ecological transition zone boundary and key risk areas with significantly abnormally high posterior uncertainty using a weighted greedy algorithm.

[0185] Experimental verification and simulation analysis show that, compared with traditional static uniform deployment and random encryption strategies, this application exhibits a strong global error convergence and local extremum "peak shaving" effect in continuous decision-making rounds, achieving a maximum leap in global carbon flux inversion accuracy with minimal engineering cost. Furthermore, addressing the unavoidable computational bottlenecks and engineering implementation delays in the real physical world, this system not only integrates a local correlation truncation mechanism at the mathematical level to ensure high-dimensional decision-making efficiency, but also innovatively introduces a temporary mobile observation and dynamic virtual operator compensation mechanism based on UAV swarms. Thus, this scheme effectively bridges the spatiotemporal gap between digital commands and physical hardware deployment, providing a robust and cost-effective system-level engineering solution for high-precision, blind-spot-free dynamic monitoring of carbon sinks / emissions under complex terrain conditions.

[0186] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0187] Figure 9 A structural block diagram of an example heterogeneous carbon sink node encryption system based on assimilation inversion feedback according to an embodiment of this application is shown.

[0188] like Figure 9As shown, the heterogeneous carbon sink node encryption system 900 based on assimilation and inversion feedback includes a multi-source data acquisition unit 910, a data assimilation processing unit 920, a spatial heterogeneity risk assessment unit 930, an expected return quantification unit 940, a heterogeneous node optimization decision-making unit 950, and a topology dynamic adjustment unit 960.

[0189] The multi-source data acquisition unit 910 is used to acquire multi-source observation data and corresponding background driving data collected by a heterogeneous monitoring network deployed in the target area; wherein, the target area is divided into multiple spatial grids, and the heterogeneous monitoring network includes ordinary nodes deployed in the spatial grids with an observation accuracy at a first accuracy level, and core nodes with an observation accuracy at a second accuracy level higher than the first accuracy level; the background driving data includes at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture data.

[0190] The data assimilation processing unit 920 is used to input the multi-source observation data and the background driving data into a preset surface carbon cycle model to perform data assimilation processing, so as to obtain the posterior carbon flux estimation results of each spatial grid and the posterior covariance matrix characterizing the estimation uncertainty; wherein, the surface carbon cycle model is used to drive the simulation of the carbon cycle physical process of the target area under the background driving data.

[0191] The spatial heterogeneity risk assessment unit 930 is used to extract the local uncertainty and spatial gradient information characterizing the spatial rate of change of carbon flux for each spatial grid based on the posterior carbon flux estimation results and the posterior covariance matrix, and to calculate the spatial heterogeneity risk index of each spatial grid using the local uncertainty and spatial gradient information of each spatial grid, so as to characterize the monitoring risk of each spatial grid.

[0192] The expected return quantification unit 940 is used to simulate the covariance update process of adding the core node or upgrading the ordinary node deployed in the corresponding candidate grid to the core node for the candidate grid that meets the preset triggering conditions of the spatial heterogeneity risk index, so as to quantify the expected reduction of the global estimation uncertainty corresponding to the candidate grid.

[0193] The heterogeneous node optimization decision unit 950 is used to perform node optimization decisions by comprehensively considering the spatial heterogeneity risk index, the expected reduction amount, and the preset resource budget constraints of each candidate grid, and generate a heterogeneous node encryption strategy; wherein, the heterogeneous node encryption strategy includes core node addition instructions and / or existing ordinary node capability upgrade instructions, and the existing ordinary node capability upgrade instructions are used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level.

[0194] The topology dynamic adjustment unit 960 is used to output the instructions indicated by the heterogeneous node encryption strategy to trigger the core node addition deployment operation and / or ordinary node upgrade operation in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

[0195] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the heterogeneous carbon sink node encryption methods based on assimilation and inversion feedback described above.

[0196] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described heterogeneous carbon sink node encryption methods based on assimilation and inversion feedback.

[0197] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a heterogeneous carbon sink node encryption method based on assimilation inversion feedback.

[0198] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0199] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0200] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0201] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for densifying heterogeneous carbon sink nodes based on assimilation and inversion feedback, characterized in that, The method includes: Acquire multi-source observation data and corresponding background driving data collected by a heterogeneous monitoring network deployed in a target area; wherein, the target area is divided into multiple spatial grids, and the heterogeneous monitoring network includes ordinary nodes deployed in the spatial grids with an observation accuracy at a first accuracy level, and core nodes with an observation accuracy at a second accuracy level higher than the first accuracy level; the background driving data includes at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture data. The multi-source observation data and the background driving data are input into a preset surface carbon cycle model to perform data assimilation processing, so as to obtain the posterior carbon flux estimation results of each spatial grid and the posterior covariance matrix characterizing the estimation uncertainty; wherein, the surface carbon cycle model is used to drive the simulation of the carbon cycle physical process of the target area under the background driving data. Based on the posterior carbon flux estimation results and the posterior covariance matrix, the local uncertainty of each spatial grid and the spatial gradient information characterizing the spatial rate of change of carbon flux are extracted. The spatial heterogeneity risk index of each spatial grid is calculated using the local uncertainty and the spatial gradient information of each spatial grid to characterize the monitoring risk of each spatial grid. For candidate grids whose spatial heterogeneity risk index meets the preset triggering conditions, the covariance update process after adding the core node or upgrading the ordinary node deployed in the corresponding candidate grid to the core node is simulated using the posterior covariance matrix, so as to quantify the expected reduction in the global estimation uncertainty corresponding to the candidate grid. By combining the spatial heterogeneity risk index, the expected reduction amount, and the preset resource budget constraints of each candidate grid, a heterogeneous node encryption strategy is generated. The heterogeneous node encryption strategy includes a core node addition instruction and / or an existing ordinary node capability upgrade instruction. The existing ordinary node capability upgrade instruction is used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level. Output the instructions indicated by the heterogeneous node encryption strategy to trigger the core node new deployment operation and / or ordinary node upgrade operation to be performed in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

2. The method according to claim 1, characterized in that, The step of inputting the multi-source observation data and the background driving data into a preset surface carbon cycle model to perform data assimilation processing, in order to obtain the posterior carbon flux estimation results and the posterior covariance matrix characterizing the estimation uncertainty for each spatial grid, includes: The surface carbon cycle model is used to simulate the net ecosystem exchange process in the target area under the background driving data to generate a priori prediction set. The priori prediction set contains multiple priori set members, and each priori set member is used to characterize the priori state distribution of carbon flux in each spatial grid in the target area. The multi-source observation data is converted into actual observation vectors, and an observation operator is constructed to map the state variables of the surface carbon cycle model on each of the spatial grids to the physical spatial locations of each node. Based on the first or second precision level of each node, an observation error variance matching its hardware physical characteristics is assigned to the corresponding node to construct an observation error covariance matrix with precision heterogeneity. Based on the prior prediction set, the observation operator, and the observation error covariance matrix, the cross covariance between the prior prediction set and the observation space, as well as the prediction error covariance in the observation space, are calculated, and the Kalman gain matrix is ​​solved accordingly. Using the Kalman gain matrix, the observation operator, and the actual observation vector, state updates are performed on each prior set member in the prior prediction set to obtain a posterior set composed of multiple posterior set members; The ensemble mean of the posterior set is determined as the posterior carbon flux estimate for each of the spatial grids, and the sample covariance matrix of the posterior set relative to the ensemble mean is calculated to generate the posterior covariance matrix.

3. The method according to claim 1, characterized in that, Based on the posterior carbon flux estimation results and the posterior covariance matrix, the local uncertainty and spatial gradient information characterizing the spatial rate of change of carbon flux for each spatial grid are extracted, and the spatial heterogeneity risk index for each spatial grid is calculated using the local uncertainty and spatial gradient information of each spatial grid, including: Extract the main diagonal elements of the posterior covariance matrix and determine the main diagonal elements as the local uncertainty corresponding to each spatial grid, so as to characterize the posterior error variance of the corresponding spatial grid; For each of the aforementioned spatial grids, based on the difference between the posterior carbon flux estimation results of the spatial grid and the adjacent spatial grids in each preset orthogonal coordinate direction, an approximate value of the flux partial derivative in each preset orthogonal coordinate direction is determined, and based on the approximate value of the flux partial derivative in each preset orthogonal coordinate direction, the carbon flux gradient intensity of the corresponding spatial grid is determined as the corresponding spatial gradient information. The distance attenuation weight is determined based on the physical spatial distance between the corresponding spatial grid and each grid in its local spatial neighborhood, and the local uncertainty of each grid in the local spatial neighborhood is weighted and aggregated using the distance attenuation weight to obtain the neighborhood distance weighted uncertainty of the corresponding spatial grid. The local uncertainty of the corresponding spatial grid is coupled with the carbon flux gradient intensity adjusted by a preset gradient weight, and the coupling result is calculated as a ratio to the neighborhood distance-weighted uncertainty, which serves as a neighborhood risk benchmark, to determine the spatial heterogeneity risk index of each spatial grid.

4. The method according to claim 1, characterized in that, The process of simulating the covariance update after adding the core node or upgrading an ordinary node deployed in the corresponding candidate grid to the core node using the posterior covariance matrix, in order to quantify the expected reduction in the global estimation uncertainty corresponding to the candidate grid, includes: Based on the physical spatial distance between each of the spatial grids and the preset correlation distance threshold, a localization truncation matrix based on the Gaspari-Cohn function is constructed. The localized truncation matrix is ​​used to perform a localized truncation operation on the posterior covariance matrix to generate a localized posterior covariance matrix. For each of the candidate grids, a virtual observation operator is constructed to map the model state variables when the second precision level observation is set at the candidate grid to the corresponding virtual observation space; By combining the preset observation error variance corresponding to the second accuracy level, the virtual observation operator and the localized posterior covariance matrix, a virtual Kalman gain is generated to characterize the updating effect on the model state estimation after introducing the second accuracy level observation at the candidate grid. The localized posterior covariance matrix is ​​simulated and updated using the virtual Kalman gain to obtain the virtual updated covariance matrix corresponding to the candidate grid. The expected reduction in uncertainty for the candidate grid is determined based on the difference between the localized posterior covariance matrix and the virtual updated covariance matrix in terms of global uncertainty measure; wherein the global uncertainty measure is represented by the trace of the covariance matrix.

5. The method according to claim 1, characterized in that, The decision-making process for generating the existing ordinary node capability upgrade instructions included in the heterogeneous node encryption strategy includes: For each ordinary node deployed in the candidate grid, the assimilation residual sequence of the ordinary node within a consecutive preset number of data assimilation processing cycles is extracted; wherein, the assimilation residual of a single data assimilation processing cycle is the difference between the actual observation data collected by the ordinary node and the forecast data output by the surface carbon cycle model at the corresponding physical spatial location of the ordinary node; Calculate the variance of the assimilation residual sequence for each of the ordinary nodes, and use it as the residual variance characterizing the degree of observation deviation of the corresponding ordinary node; The ecological scene weight of each ordinary node in the spatial grid is obtained. The ecological scene weight is quantitatively assigned according to whether the corresponding spatial grid is located at the boundary between carbon source and carbon sink and whether it belongs to a preset ecological sensitive area. Combining the local uncertainty of the spatial grid where each ordinary node is located, the residual variance of the ordinary node, and the ecological scenario weight of the spatial grid where the ordinary node is located, a node upgrade evaluation index is calculated for each ordinary node using a normalized weighted fusion method. The node upgrade evaluation index is positively correlated with the local uncertainty of the corresponding spatial grid where the ordinary node is located, positively correlated with the residual variance of the corresponding ordinary node, and positively correlated with the ecological scenario weight of the corresponding spatial grid where the ordinary node is located. Furthermore, the local uncertainty and the residual variance are normalized using the maximum value of the corresponding indicator within the target area as the benchmark. When the node upgrade evaluation index of the target ordinary node exceeds the preset evaluation threshold, the target ordinary node is identified as the upgrade target node, and an existing ordinary node capability upgrade instruction is generated for the upgrade target node.

6. The method according to claim 5, characterized in that, The process of combining the spatial heterogeneity risk index of each candidate grid, the expected reduction amount, and the preset resource budget constraints to perform node optimization decisions and generate a heterogeneous node encryption strategy includes: For each candidate grid, the unit execution cost and unit operating energy consumption of the candidate grid are determined according to the candidate instruction type corresponding to the candidate grid; wherein, the candidate instruction type includes core node new instruction and existing ordinary node capability upgrade instruction, the core node new instruction corresponds to equipment purchase cost and full operating energy consumption, and the existing ordinary node capability upgrade instruction corresponds to hardware module replacement cost and energy consumption increment cost. A node selection optimization model is constructed with the goal of maximizing the weighted comprehensive benefit and satisfying the constraints of the total hardware budget limit and the total energy consumption budget limit. The weighted comprehensive benefit is determined by the spatial heterogeneity risk index and the expected reduction amount of each candidate grid after normalization and according to the preset weights. The spatial heterogeneity risk index and the expected reduction amount both contribute positively to the weighted comprehensive benefit. The node selection optimization model is solved using a weighted greedy algorithm, and the unit resource comprehensive benefit score of each candidate grid is calculated based on the weighted comprehensive benefit of each candidate grid, the corresponding unit execution cost, and the unit operating energy consumption. Under the condition of satisfying the total hardware budget limit and the total energy consumption budget limit, the candidate grids are iteratively selected in descending order of the comprehensive benefit score of unit resources until adding the next candidate grid will cause the cumulative cost or cumulative energy consumption to exceed the corresponding limit constraint, at which point the iterative selection stops. Output the set of target grids selected when the iteration stops and the candidate instruction types corresponding to each target grid to generate a heterogeneous node encryption strategy.

7. The method according to claim 1, characterized in that, After outputting the instructions indicated by the heterogeneous node encryption strategy to trigger the execution of core node addition deployment operations and / or ordinary node upgrade operations in the corresponding spatial grid, the method further includes: For each spatial grid that triggers a new deployment operation for a core node or an upgrade operation for a regular node, based on the terrain accessibility and hardware engineering cycle of that spatial grid, calculate the estimated implementation delay from the time the execution command is received until the underlying hardware deployment at that spatial grid is completed and valid data is generated. When the estimated implementation delay time exceeds the preset blind zone safety delay threshold, the three-dimensional spatial coordinates of the spatial grid are determined, and a temporary spatial scheduling instruction containing the three-dimensional spatial coordinates is generated. Based on the temporary space scheduling instruction, a swarm of UAVs carrying airborne carbon flux sensing payloads that match the second accuracy level are scheduled as temporary mobile nodes to fly to the target airspace corresponding to the three-dimensional spatial coordinates to perform temporary high-frequency data acquisition. The temporary high-frequency data collected by the temporary mobile node is input into the preset surface carbon cycle model in real time as newly added multi-source observation data to perform the data assimilation process, and a dynamic virtual observation operator is constructed based on the spatial location of the temporary mobile node in the target airspace. Based on the temporary high-frequency data and the dynamic virtual observation operator, a compensatory data assimilation update is performed to update the posterior carbon flux estimation results and the posterior covariance matrix of each spatial grid until the underlying hardware deployment at that spatial grid is completed and connected to the heterogeneous monitoring network.

8. A heterogeneous carbon sink node encryption system based on assimilation and inversion feedback, characterized in that, The system includes: A multi-source data acquisition unit is used to acquire multi-source observation data and corresponding background driving data collected by a heterogeneous monitoring network deployed in a target area; wherein, the target area is divided into multiple spatial grids, and the heterogeneous monitoring network includes ordinary nodes deployed in the spatial grids with an observation accuracy of a first accuracy level, and core nodes with an observation accuracy of a second accuracy level higher than the first accuracy level; the background driving data includes at least one of the following: meteorological forcing data, satellite remote sensing vegetation index, and soil moisture data. The data assimilation processing unit is used to input the multi-source observation data and the background driving data into a preset surface carbon cycle model to perform data assimilation processing, so as to obtain the posterior carbon flux estimation results of each spatial grid and the posterior covariance matrix characterizing the estimation uncertainty; wherein, the surface carbon cycle model is used to drive the simulation of the carbon cycle physical process of the target area under the background driving data. The spatial heterogeneity risk assessment unit is used to extract the local uncertainty and spatial gradient information characterizing the spatial rate of change of carbon flux for each spatial grid based on the posterior carbon flux estimation results and the posterior covariance matrix, and to calculate the spatial heterogeneity risk index of each spatial grid using the local uncertainty and spatial gradient information of each spatial grid, so as to characterize the monitoring risk of each spatial grid. The expected return quantification unit is used to simulate the covariance update process of adding the core node or upgrading the ordinary node deployed in the corresponding candidate grid to the core node for the candidate grid that meets the preset triggering conditions of the spatial heterogeneity risk index, so as to quantify the expected reduction of the global estimation uncertainty corresponding to the candidate grid. A heterogeneous node optimization decision unit is used to perform node optimization decisions by comprehensively considering the spatial heterogeneity risk index, the expected reduction amount, and the preset resource budget constraints of each candidate grid, and to generate a heterogeneous node encryption strategy; wherein, the heterogeneous node encryption strategy includes core node addition instructions and / or existing ordinary node capability upgrade instructions, and the existing ordinary node capability upgrade instructions are used to improve the observation accuracy of the corresponding ordinary node from the first accuracy level to the second accuracy level. The topology dynamic adjustment unit is used to output the instructions indicated by the heterogeneous node encryption strategy to trigger the core node addition deployment operation and / or ordinary node upgrade operation in the corresponding spatial grid, thereby adjusting the physical deployment topology and / or node capability configuration of the heterogeneous monitoring network.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images