A method for monitoring forest carbon sink

Abstract

The application relates to the technical field of carbon sink monitoring, and discloses a forest carbon sink monitoring method, which comprises the following steps: constructing an air-ground cooperative collection framework; in the process of unmanned aerial vehicle inspection, based on the space-time correlation parameters of identifiers matching the unmanned aerial vehicle and the ground sensor nodes, the space-time fusion of the unmanned aerial vehicle collection data and the ground sensor network collection data is carried out, and an air-ground fusion data set is generated; the air-ground fusion data set is subjected to data preprocessing based on the real-time distance in the space-time correlation parameters, and pretreated data is obtained; the pretreated data is subjected to dynamic transmission according to the communication bandwidth state; the pretreated data transmitted by the overlapping monitoring areas is subjected to cooperative verification, and verification data is obtained; the verification data is processed based on a carbon sink estimation model, and carbon sink estimation results are generated; based on the carbon sink estimation results of the current monitoring period and the previous monitoring period, the carbon sink change rate is determined, and the next unmanned aerial vehicle inspection path is optimized. The application improves the credibility of carbon sink collection data.

Description

Technical Field

[0001] This invention relates to the field of carbon sink monitoring technology, and more specifically, to a method for monitoring forest carbon sinks. Background Technology

[0002] Forest carbon sink monitoring is a fundamental technology for forest resource management, ecological environment assessment, and related work on carbon peaking and carbon neutrality. Current forest carbon sink monitoring mainly employs fixed ground-based sensor monitoring, unmanned aerial vehicle (UAV) patrol monitoring, and a combined air-ground monitoring approach integrating UAVs and ground sensors.

[0003] While fixed ground sensors can continuously acquire monitoring data, their coverage is limited by deployment density, making it difficult to reflect the spatial distribution characteristics of forest carbon sinks. Furthermore, long-term operation of these sensors can lead to baseline drift, affecting monitoring accuracy. Unmanned aerial vehicle (UAV) inspections offer advantages such as wide coverage and high mobility, but limitations in endurance and communication bandwidth result in issues like large data transmission volumes, insufficient real-time performance, and poor adaptability to inspection paths. To balance monitoring range and accuracy, existing technologies propose air-ground collaborative monitoring schemes combining UAVs and ground sensors. However, in this scheme, data acquisition, data calibration, and inspection planning are independent processes, lacking a collaborative mechanism. On one hand, the time reference and spatial scale of data collected by UAVs and ground sensors differ, affecting the accuracy of air-ground data fusion. On the other hand, existing calibration methods primarily use ground sensors to unidirectionally correct UAV data, making it difficult to promptly correct baseline drift caused by long-term operation of ground sensors.

[0004] Therefore, it is necessary to design a method for monitoring forest carbon sinks to address the problems existing in current technologies. Summary of the Invention

[0005] In view of this, the present invention proposes a method for monitoring forest carbon sinks, which aims to solve the problem that the time reference and spatial scale of data collected by UAVs and ground sensors are different, affecting the accuracy of air-ground data fusion; existing calibration methods mainly use ground sensors to perform one-way correction of UAV data, which is difficult to correct the baseline drift caused by long-term operation of ground sensors in a timely manner.

[0006] This invention proposes a method for monitoring forest carbon sinks, comprising: An air-ground collaborative data acquisition architecture is built based on UAVs and ground sensor networks, and identifiers are assigned to ground sensor nodes in each region. During the UAV inspection process, the spatiotemporal correlation parameters between the UAV and ground sensor nodes are matched based on the identifier, and the data collected by the UAV and the data collected by the ground sensor network are spatiotemporally fused to generate an air-ground fusion dataset. Based on the real-time distance in the spatiotemporal correlation parameters, the air-to-ground fusion dataset is preprocessed to obtain preprocessed data; the data preprocessing includes high-precision calibration mode, lightweight noise reduction mode and transition fusion mode. The preprocessed data is dynamically transmitted according to the communication bandwidth status; and the preprocessed data transmitted in the overlapping monitoring area is collaboratively verified to obtain verification data. The verification data is processed based on the carbon sink estimation model to generate carbon sink estimation results; Based on the carbon sink estimation results of the current monitoring cycle and the previous monitoring cycle, the carbon sink change rate is determined, and the carbon sink change rate is used as a constraint to optimize the next UAV inspection path.

[0007] Furthermore, it also includes establishing a related database; the related database includes latitude and longitude coordinates, parameter types, calibration records, and historical baseline data.

[0008] Furthermore, when generating the air-to-ground fusion dataset, the following are included: Using the clock of the ground sensor node as a reference, the onboard clock of the UAV is synchronized through a precise time protocol for time synchronization calibration. Based on the real-time flight altitude of the UAV and the spatial distribution density of the ground sensor node, the spatial resolution of the UAV-collected data is adjusted to match the effective monitoring radius of the ground sensor node for spatial scale matching. Using an ensemble Kalman filter algorithm, the UAV-collected data after the time synchronization calibration and spatial scale matching are assimilated and fused with the data collected by the ground sensor node to generate the air-ground fusion dataset.

[0009] Furthermore, the high-precision calibration mode includes: When the real-time distance is less than or equal to the first distance threshold, the high-precision calibration mode is performed: based on the identifier, historical baseline data of the ground sensor node is retrieved from the associated database; based on the historical baseline data, the error is determined, and the UAV-collected data in the air-ground fusion dataset is corrected; the corrected UAV-collected data is sent back to the ground sensor node for cross-validation; when the deviation is greater than the deviation threshold, the ground sensor node performs self-calibration and updates the historical baseline data in the associated database.

[0010] Furthermore, the lightweight noise reduction mode includes: When the real-time distance is greater than or equal to the second distance threshold, the lightweight noise reduction mode is implemented: the carbon sink monitoring parameters in the air-ground fusion dataset are classified and labeled, and the carbon sink monitoring parameters after classification and labeling are adaptively filtered and compressed using a sliding window.

[0011] Furthermore, the transition fusion mode includes: When the real-time distance is greater than the first distance threshold and less than the second distance threshold, the transition fusion mode is performed: the air-to-ground fusion dataset is subjected to one-way ground calibration and selective compression. Wherein, the first distance threshold is less than the second distance threshold.

[0012] Furthermore, when dynamically transmitting the preprocessed data based on the communication bandwidth status, it includes: When the communication bandwidth is less than the bandwidth threshold, the core parameters in the preprocessed data are transmitted in real time, the auxiliary parameters in the preprocessed data are transmitted in compressed form, and the non-critical variables in the preprocessed data are temporarily stored locally. When the communication bandwidth is greater than or equal to the bandwidth threshold, data augmentation processing is initiated. The data augmentation processing includes performing spatiotemporal correlation verification on multi-sensor data and predicting and marking outliers based on the historical baseline data.

[0013] Furthermore, when performing collaborative verification on the preprocessed data transmitted in the overlapping monitoring area to obtain verification data, the following steps are included: A dynamic communication link is established between the edge computing devices of adjacent UAVs to synchronize the carbon sink parameters of the overlapping monitoring area. When the carbon sink parameter deviation of several UAVs is greater than or equal to the parameter deviation threshold, the UAV closest to the target area is triggered to perform secondary data collection. The area that triggers secondary data collection multiple times in a row is marked as a high uncertainty area, and the access frequency of the next UAV inspection is increased.

[0014] Furthermore, when processing the verification data based on the carbon sink estimation model to generate the carbon sink estimation result, the process includes: Tree characteristics are extracted from the verification data to determine the biomass of a single tree, and the carbon storage of a single tree is obtained based on the carbon conversion coefficient. The carbon storage of a single tree in a preset quadrat is summarized as the quadrat carbon sink value to generate the carbon sink estimation result.

[0015] Furthermore, when optimizing the next UAV inspection path using the carbon sink change rate as a constraint, the following steps are included: A total path cost function is constructed, which includes terrain factor cost, meteorological factor cost, sensor factor cost, and carbon sink change rate driving cost term. When the carbon sink change rate is greater than the change rate threshold, the access weight in the carbon sink change rate driving cost term is increased. The minimum cost path is solved using the particle swarm optimization algorithm, which is used as the next UAV inspection path. When the real-time wind speed is greater than the wind speed threshold or the ground sensor node density deviation is greater than the density deviation threshold, an emergency path reconstruction is performed. The emergency path reconstruction includes calling the backup safe route, correcting the path node spacing, and smoothing the path.

[0016] Compared with existing technologies, the advantages of this invention are as follows: It unifies time and space as dual benchmarks, combined with filtering and fusion, improving the accuracy of carbon sink estimation; it breaks the traditional one-way mode of only calibrating UAVs on the ground, continuously correcting the aging and drift errors of ground sensors, ensuring a stable and reliable long-term monitoring benchmark; it automatically adjusts the inspection frequency according to carbon sink fluctuations, improving monitoring efficiency under the same hardware conditions; it only transmits core data in stages, ensuring stable operation in mountainous and low-bandwidth environments; cross-verification by multiple UAVs reduces accidental measurement errors of single devices, improving the reliability of carbon sink data collection; it features closed-loop linkage across the entire chain, with data exchange and feedback between each link, rather than independent modules, giving the entire monitoring system self-optimization capabilities; structured equipment identification + a four-dimensional database facilitates carbon sink verification and traceability auditing; preprocessing thresholds, path weights, and transmission strategies can all be dynamically adjusted according to forest canopy density, terrain, and weather, adapting to multiple scenarios including dense forests, sparse forests, mountains, and plains; and it automatically switches to safe flight routes in severe weather, reducing the flight risks of UAVs. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for monitoring forest carbon sinks provided in an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] In some embodiments of this application, see Figure 1 As shown, a method for monitoring forest carbon sinks is proposed, including: S100: Based on UAVs and ground sensor networks, an air-ground collaborative data acquisition architecture is built, and identifiers are assigned to ground sensor nodes in various locations; S200: During the UAV inspection process, based on the identifier matching of the spatiotemporal correlation parameters between the UAV and the ground sensor node, the data collected by the UAV and the data collected by the ground sensor network are spatiotemporally fused to generate an air-ground fusion dataset. S300: Based on the real-time distance in the spatiotemporal correlation parameters, the air-to-ground fusion dataset is preprocessed to obtain preprocessed data; the data preprocessing includes high-precision calibration mode, lightweight noise reduction mode and transition fusion mode; S400: Dynamically transmits preprocessed data according to the communication bandwidth status; and performs collaborative verification on the preprocessed data transmitted in the overlapping monitoring area to obtain verification data; S500: Processes verification data based on the carbon sink estimation model to generate carbon sink estimation results; S600: Based on the carbon sink estimation results of the current monitoring cycle and the previous monitoring cycle, determine the carbon sink change rate, use the carbon sink change rate as a constraint, and optimize the next UAV inspection path.

[0020] This embodiment constructs a collaborative data acquisition system between UAVs and ground sensor networks to achieve continuous acquisition, unified management, and subsequent fusion analysis of multi-dimensional monitoring data in forest areas. This addresses the problems of fixed ground monitoring points, limited spatial coverage, and difficulty in correlating UAV remote sensing data with ground-based measured data in traditional forest carbon sequestration monitoring.

[0021] In step S100, an air-ground collaborative data acquisition architecture is first constructed, consisting of a UAV inspection unit, ground sensor nodes, and a correlation data management unit. The UAV serves as a mobile aerial acquisition platform, used to acquire remote sensing information and spatial structure information of a large forest area; the ground sensor nodes serve as fixed acquisition units, used to acquire continuously changing environmental parameters and vegetation status parameters within a local area; and the correlation data management unit is used to establish the correspondence between UAV-acquired data and ground sensor node-acquired data, achieving unified indexing and management of multi-source monitoring data.

[0022] Specifically, within the target forest monitoring area, ground sensor nodes are deployed in a grid pattern based on forest area, topographic relief, vegetation density, and carbon sequestration monitoring needs. Each ground sensor node is located at a preset monitoring position to collect forest ecological status data within its corresponding area. Ground sensor nodes include, but are not limited to, temperature and humidity sensors, soil moisture sensors, light sensors, carbon dioxide concentration sensors, tree trunk growth monitoring sensors, and microclimate monitoring sensors. By collaboratively collecting forest environmental parameters using multiple types of sensors, subsequent carbon sequestration estimation can simultaneously consider vegetation growth environment factors and vegetation structure change factors, thereby improving the accuracy of forest carbon storage assessment.

[0023] To differentiate data between different ground sensor nodes and to correlate data between UAVs and ground sensors, this embodiment assigns a unique identifier to each ground sensor node. The unique identifier is generated using a multi-field combination encoding method, including a region code field, a device number field, and a device time field. The region code field indicates the forest area or administrative region where the sensor node is located; the device number field distinguishes different sensor nodes within the same area; and the device time field records the deployment time or production batch of the sensor node. For example, a sensor node deployed at monitoring location 001 in a certain area with a production date of January 1, 2023, can be assigned the identifier "H N-001-20230101". This structured identification method ensures that each ground sensor node has unique identity information, avoiding data confusion caused by the increasing number of sensors within a forest area.

[0024] After assigning identifiers to ground sensor nodes, a related database is established to store and manage the basic information and historical operational information of ground sensor nodes in various locations. The related database includes node identification information, spatial location information, monitoring parameter type information, calibration record information, and historical baseline data. Specifically, the node identification information uniquely identifies the corresponding ground sensor node; the spatial location information includes the longitude and latitude coordinates of the sensor node, used to determine its spatial location within the forest area; the monitoring parameter type information records the type of data collected by the node, such as vegetation environmental parameters, soil parameters, or air parameters; the calibration record information records the time of each calibration, calibration results, and calibration adjustment parameters; and the historical baseline data stores the range of data changes generated by the sensor node under normal operating conditions over a long period, providing a reference for subsequent anomaly identification and data correction.

[0025] During the database establishment process, the spatial locations of ground sensor nodes are bound to electronic maps of forest areas, creating spatially indexed data records for each sensor node in the database. Specifically, the latitude and longitude coordinates of the sensor node are obtained based on its installation location, and these coordinates are associated with and stored as node identifiers. When a UAV performs an inspection mission and enters the corresponding monitoring area, its onboard computing equipment can quickly query information about nearby ground sensor nodes based on its current location and retrieve historical baseline data from those nodes, enabling rapid matching between UAV-collected data and ground sensor data.

[0026] To ensure the continued validity of the data in the associated database, this embodiment employs a periodic update mechanism for database maintenance. Ground sensor nodes send collected data to the associated database according to a preset upload cycle, which is adjusted based on the accuracy requirements of forest monitoring; in this embodiment, it is set to once every 2 to 10 minutes. After a sensor node completes a data acquisition, it uploads the current acquisition time, node identifier, acquisition parameters, and acquired value to the associated database, and updates the corresponding node's data record based on historical data. When it detects that a sensor node has completed maintenance, equipment replacement, or recalibration, the device status information and calibration records in the database are updated synchronously to ensure that the information stored in the database is consistent with the on-site device status.

[0027] During the drone inspection, the drone performs dynamic coverage data collection of the forest area according to the preset inspection tasks. The drone is equipped with multispectral imaging equipment, lidar equipment, and airborne edge computing equipment. The multispectral imaging equipment is used to collect forest vegetation reflectance characteristic information and obtain vegetation spectral response data in different bands; the lidar equipment is used to obtain information on forest canopy height, tree spatial structure, and stand density; the airborne edge computing equipment is used to perform preliminary processing of the collected data and complete data association based on the ground sensor node identification information.

[0028] Specifically, during flight, the UAV acquires its own position coordinates, flight altitude, and flight time information in real time, and searches for ground sensor nodes within a preset range based on its current location. The onboard edge computing device accesses a related database using the ground sensor node identifiers to obtain the spatial coordinates, monitoring parameter types, and historical baseline data of the corresponding nodes. For example, when the UAV flies over a forest area, it determines multiple ground sensor nodes within its coverage area based on its current latitude and longitude coordinates, and retrieves corresponding historical monitoring records using the node identifiers, enabling the remote sensing data collected by the UAV to establish a correlation with the environmental data collected by the ground sensor nodes.

[0029] Ground sensor nodes upload data according to a fixed sampling period, while the UAV performs dynamic inspection tasks based on a preset inspection strategy. Specifically, the UAV flies along a generated inspection path, covering approximately 5 square kilometers of forest area every 30 minutes, and continuously collects multispectral image data and laser point cloud data of the target area during flight. Simultaneously, the UAV uses onboard edge computing equipment to spatially locate the collected area and queries the corresponding associated database using ground sensor node identifiers to achieve matching between the UAV's collected area, the location of the ground sensor nodes, and historical monitoring data.

[0030] In step S200, a unified spatiotemporal correlation relationship is constructed between UAV-collected data and ground sensor network-collected data through identifier-based air-to-ground data association, time synchronization calibration, spatial scale matching, and multi-source data assimilation and fusion, generating an air-to-ground fused dataset for forest carbon sink calculation. Because UAV remote sensing data has the characteristics of wide coverage and strong spatial continuity, while ground sensor data has the characteristics of high sampling accuracy and strong local temporal continuity, the two types of data suffer from problems such as different time bases, inconsistent spatial scales, and differences in observation accuracy during the acquisition process. Therefore, this embodiment establishes a spatiotemporal synchronization calibration mechanism to fuse aerial remote sensing observation data and ground sensor monitoring data under a unified spatial scale and a unified time base, thereby improving the accuracy and continuity of forest carbon sink monitoring data.

[0031] Specifically, during the inspection mission, the onboard edge computing device acquires real-time information on the drone's current flight status and ground sensor nodes within its coverage area. The drone's flight status information includes its current latitude and longitude coordinates, flight altitude, flight speed, attitude angle, and acquisition time. The ground sensor node information includes node identifiers, installation location coordinates, monitoring parameter types, effective monitoring radius, and historical data. Based on the drone's current spatial location, the onboard edge computing device retrieves ground sensor nodes within a preset range from a related database and establishes a data mapping relationship between the drone and the corresponding ground sensor node based on the unique identifier of each node.

[0032] After establishing the data mapping relationship, the spatiotemporal correlation parameters between the UAV and the ground sensor nodes are determined. These parameters include spatial distance parameters, time synchronization deviation parameters, coverage parameters, and data correspondence parameters. Specifically, the spatial distance parameter describes the distance between the current data acquisition location of the UAV and the actual installation location of the ground sensor node; the time synchronization deviation parameter describes the difference between the UAV's data acquisition time and the ground sensor's data acquisition time; the coverage parameter determines the ground sensor monitoring area corresponding to the UAV's remote sensing data; and the data correspondence parameter determines the correspondence between spatial grid cells in the UAV's data acquisition and the ground sensor nodes. These spatiotemporal correlation parameters enable accurate matching of UAV-acquired data to the corresponding ground sensor nodes, facilitating subsequent multi-source data fusion processing.

[0033] To reduce fusion errors caused by asynchronous acquisition times of air-to-ground data, this embodiment first performs time synchronization calibration. Specifically, the GPS clock of the ground sensor node is used as the air-to-ground data synchronization reference, and the UAV's onboard clock is synchronized and calibrated using the IEEE 1588 Precision Time Protocol (PTP). During the synchronization process, the ground sensor node acts as the PTP master clock device, and the UAV's onboard edge computing device acts as the slave clock device. The ground sensor node sends synchronization messages to the UAV at a preset period.

[0034] When the UAV receives a Sync message from the ground sensor node, it records the message reception time and returns a Delay_Req request message to the ground sensor node. Upon receiving the request message, the ground sensor node generates a Delay_Resp response message and sends back the corresponding time information. The UAV's onboard edge computing device calculates the time offset between the UAV's onboard clock and the ground sensor node's clock based on the Sync message's sending and receiving times, as well as network transmission delay information. It then corrects the onboard clock according to the time offset to ensure that the timestamps of the data collected by the UAV are consistent with those of the data collected by the ground sensors.

[0035] In this embodiment, the time synchronization method described above controls the data timestamp deviation between the UAV and the ground sensor nodes to within 10ms. For example, during UAV inspection, the UAV collects multispectral image data of a forest area, while ground sensor nodes in the corresponding area collect carbon dioxide concentration data. After PTP time synchronization calibration, both types of data correspond to the same time reference, avoiding misinterpretation of changes in carbon dioxide concentration caused by changes in vegetation photosynthesis due to time differences as spatial differences. By reducing time deviation, the time consistency in the subsequent carbon sink parameter fusion process is improved.

[0036] After completing time synchronization calibration, this embodiment further performs spatial scale matching processing. Since the data collected by UAV remote sensing equipment is typically continuous area data, while the data collected by ground sensor nodes is fixed-location point data, there is a problem of inconsistent spatial resolution. Therefore, this embodiment adaptively adjusts the spatial scale of the UAV-collected data based on the UAV's real-time flight altitude, the imaging parameters of the remote sensing equipment, and the spatial distribution density of the ground sensor nodes, so that the spatial units corresponding to the UAV remote sensing data match the effective monitoring area of ​​the ground sensor nodes.

[0037] Specifically, the spatial resolution of the current remote sensing data is first calculated based on the UAV's current flight altitude and the imaging parameters of the multispectral sensor. For example, when the UAV is flying at an altitude of 100m, the spatial resolution of the image acquired by the multispectral sensor can reach 0.05m / pixel. At this time, the data collected by the UAV has high spatial accuracy, while the effective monitoring range of ground sensor nodes, such as carbon dioxide concentration sensors, may be an area with a radius of 10m. Therefore, there is a significant difference in spatial scale between the two.

[0038] To address the aforementioned differences, a spatial matching grid is established based on the effective monitoring radius of the ground sensor nodes. When the spatial resolution of the UAV imagery is detected to be higher than the effective monitoring scale of the ground sensor, spatial aggregation processing is performed on the UAV imagery, fusing multiple high-resolution pixel units to form a spatial grid corresponding to the monitoring range of the ground sensor. For example, multispectral imagery data with a resolution of 0.05 m / pixel is resampled according to a spatial range of 10 m × 10 m to obtain spatial data units corresponding to the monitoring range of the ground sensor nodes.

[0039] When changes in UAV flight altitude reduce the spatial resolution of remote sensing imagery, making it impossible to directly match the effective monitoring range of ground sensors, spatial interpolation methods are used to scale the remote sensing data. This involves estimating the remote sensing feature values ​​of the target's spatial location using neighboring pixel information, ensuring that the UAV remote sensing data covers the corresponding spatial area of ​​the ground sensor nodes. This spatial scale matching process ensures that the forest canopy structure data and multispectral index data acquired by the UAV and the environmental parameter data acquired by the ground sensors have a unified spatial representation, reducing data fusion errors caused by differences in spatial scale.

[0040] After completing time synchronization calibration and spatial scale matching, this embodiment further employs the Ensemble Kalman Filter (EnKF) algorithm to assimilate and fuse UAV data and ground sensor data, generating an air-ground fused dataset. Specifically, firstly, the UAV remote sensing data after spatial scale matching is used as prior state information to describe the spatial continuous distribution of carbon sink-related parameters within the target forest area; then, high-precision monitoring data collected by ground sensor nodes is used as observation update information to correct uncertain areas in the UAV remote sensing data.

[0041] During data fusion, a state vector representing the state of forest carbon sinks is constructed. This state vector includes, but is not limited to, the carbon dioxide concentration parameter, the Normalized Difference Vegetation Index (NDVI) parameter, the vegetation cover parameter, and the canopy structure parameter corresponding to each spatial grid cell. In this embodiment, multiple state set members are generated based on the target forest area, with each set member representing a possible forest state distribution. In this embodiment, the number of set members is set to 50, using multiple state sets to describe the uncertainty of the UAV remote sensing data.

[0042] In the specific fusion process, an initial state set is first generated based on data collected by UAVs to obtain prior estimates of carbon sink parameters in the forest area. Subsequently, after ground sensor nodes upload actual monitoring data for their corresponding locations, the corresponding spatial grid positions are determined based on the sensor node identifiers, and the prior estimates are updated using ground sensor observations. Ensemble Kalman filtering calculates state correction weights based on prior estimation errors and observation errors, and adjusts the carbon sink parameter values ​​in each spatial grid cell according to these correction weights to obtain the fused posterior estimate.

[0043] For example, for a target grid cell within a forest area, UAV multispectral data calculates the NDVI index and vegetation cover status of that area, while corresponding ground sensor nodes collect data on carbon dioxide concentration changes in that area. Ensemble Kalman filtering is used to fuse the spatially continuous information provided by the UAV remote sensing data with the local high-precision information provided by the ground sensors. This results in a fusion that maintains both the wide-area coverage capability of the UAV remote sensing data and the local measurement accuracy of the ground sensor data.

[0044] After the above processing, an air-ground fusion dataset is generated. The air-ground fusion dataset includes spatial location information, temporal information, forest carbon sink monitoring parameters, and corresponding data confidence information. Specifically, spatial location information indicates the location of the forest area corresponding to the fusion data; temporal information indicates the corresponding data collection time; forest carbon sink monitoring parameters include carbon dioxide concentration, NDVI index, vegetation cover, and biomass-related parameters; and data confidence information indicates the reliability of the fusion results.

[0045] In step S300, the generated air-to-ground fusion dataset undergoes adaptive data preprocessing based on the real-time distance between the UAV and the ground sensor nodes to obtain data processing results for subsequent carbon sink estimation. Since the spatial distance between the UAV and different ground sensor nodes constantly changes when the UAV performs inspection tasks in forest areas, when the UAV is close to the ground sensor nodes, it has better data calibration conditions and can use high-precision monitoring data from the ground sensors to finely correct the UAV's remote sensing data. When the UAV is far from the ground sensor nodes, the data communication distance increases, and the data volume increases. Continuing to perform high-precision calibration at this time would increase the edge computing load and reduce inspection efficiency. Therefore, this embodiment dynamically switches between different data preprocessing modes based on the real-time distance between the UAV and the ground sensor nodes, including a high-precision calibration mode, a lightweight noise reduction mode, and a transitional fusion mode, enabling the data processing process to achieve a dynamic balance between accuracy and efficiency according to changes in spatial location.

[0046] Specifically, during the UAV inspection process, the onboard edge computing device acquires the UAV's current spatial position in real time and determines the real-time distance between the UAV and various ground sensor nodes based on the established spatiotemporal correlation parameters. The real-time distance is the spatial distance between the UAV's current position coordinates and the target ground sensor node's position coordinates, used to determine whether the UAV is currently within the effective calibration range of the ground sensor node.

[0047] In determining the real-time distance, the airborne edge computing device first obtains the UAV's current latitude and longitude coordinates and flight altitude information from the UAV navigation system, and simultaneously reads the latitude and longitude coordinates corresponding to the target ground sensor node from the associated database. Based on the UAV's three-dimensional spatial position and the ground sensor node's spatial position, the spatial distance between them is calculated, and the calculated real-time distance is used as the basis for switching data preprocessing modes.

[0048] This embodiment presets a first distance threshold and a second distance threshold. The first distance threshold represents the short-range range within which the UAV can perform high-precision data calibration with the ground sensor node, and the second distance threshold represents the range within which the UAV enters a long-range inspection state. Specifically, the first distance threshold can be set to 50 meters, and the second distance threshold can be set to 200 meters. In practical applications, the thresholds can be dynamically adjusted according to the forest area type, vegetation density, terrain undulation, and communication environment. For example, in densely canopied forest areas with severe obstruction, the first distance threshold can be adjusted to 30 meters to ensure high-precision calibration results, as the distance between the UAV and the ground sensor node is easily affected by vegetation obstruction. In sparsely vegetated areas with open terrain, the first distance threshold can be increased to 80 meters to improve the coverage of high-precision calibration.

[0049] When the real-time distance between the UAV and the target ground sensor node is detected to be less than or equal to a first distance threshold, the high-precision calibration mode is entered. At this time, the UAV is located in the vicinity of the ground sensor node, and the UAV's remote sensing acquisition area has a high degree of spatial overlap with the effective monitoring range of the ground sensor, which provides the conditions for finely correcting the UAV's acquired data using historical data from the ground sensor.

[0050] In high-precision calibration mode, the airborne edge computing device first accesses the associated database based on the unique identifier of the target ground sensor node and retrieves the historical baseline data corresponding to that node. The historical baseline data includes continuous monitoring records collected by the ground sensor node under normal operating conditions, such as carbon dioxide concentrations, vegetation environmental parameters, and corresponding time information collected over the past 30 days. By accessing the historical baseline data, the data variation patterns under the long-term stable state of the current sensor node can be obtained, providing a reference for the calibration of the data collected by the UAV.

[0051] In the specific calibration process, the historical average state of the ground sensor nodes is determined based on historical baseline data, and the current data collected by the UAV is compared with the historical baseline state to obtain the current data deviation. Here, the historical baseline data set is set as B_ID={b1,b2,...,bn}, where bi represents the i-th historical sample value and n represents the number of valid historical samples. In this embodiment, valid sample data from the most recent 30 days can be selected as the calculation basis. The error value E_t is determined based on the difference between the historical baseline mean and the current UAV data, and the UAV data is corrected according to the error value.

[0052] For example, within a forest area, a drone collects CO2 concentration data (X_t) within a spatial grid. The historical baseline average of the corresponding ground sensor nodes indicates that the CO2 concentration in this area should remain within a relatively stable range under normal conditions. When a deviation is detected between the drone-collected value and the historical baseline average, this deviation is used as an error compensation amount to correct the drone-collected data, making the corrected drone data closer to the actual forest environment.

[0053] After completing the calibration of the UAV-collected data, this embodiment further performs a two-way closed-loop cross-validation process. Specifically, the airborne edge computing device sends the calibrated UAV-collected data to the corresponding ground sensor node via an air-to-ground communication link. After receiving the calibrated data, the ground sensor node compares it with its own currently real-time collected data to determine whether there is a significant deviation between the two.

[0054] Specifically, the ground sensor node calculates the relative deviation between the UAV calibration data and its own real-time monitoring data. When the calculated deviation exceeds a preset deviation threshold, it indicates that the ground sensor node may have measurement drift, temperature influence, sensor aging, or baseline changes caused by long-term operation. For example, if the difference between the CO2 concentration data collected by the ground sensor node and the UAV calibration data exceeds 5%, the ground sensor node is determined to be in an abnormal state.

[0055] When a deviation exceeds a preset threshold, the ground sensor node automatically triggers a self-calibration process. Specifically, the ground sensor node retrieves its stored historical baseline data, internally corrects the current sensor acquisition parameters, and generates new calibration parameters. Simultaneously, it uploads the updated calibration results, calibration time, and corrected historical data to the associated database, updating the original historical baseline data. This method enables the ground sensor node to continuously utilize UAV high-spatial coverage data for reverse verification, achieving a closed-loop mechanism for long-term maintenance of ground sensors using UAV data.

[0056] The aforementioned high-precision calibration mode achieves a shift from the traditional one-way "ground sensor calibration of UAV data" to "two-way closed-loop calibration." On one hand, it uses long-term stable data from ground sensors to correct errors in UAV remote sensing acquisition; on the other hand, it uses continuous UAV spatial observation results to detect the drift status of ground sensors, thereby improving the stability of the entire forest monitoring network during long-term operation.

[0057] When the real-time distance between the UAV and the target ground sensor node is detected to be greater than or equal to the second distance threshold, the system enters a lightweight noise reduction mode. At this time, the UAV is conducting a large-scale forest patrol, and the distance between it and the ground sensor node is relatively far, making effective spatial correspondence calibration impossible. Therefore, the system prioritizes reducing data processing complexity to improve the efficiency of the UAV patrol.

[0058] In the lightweight noise reduction mode, the airborne edge computing device first classifies and labels the carbon sink monitoring parameters in the air-to-ground fusion dataset according to their importance. Specifically, based on the degree of influence of different parameters on the forest carbon sink calculation results, the monitoring parameters are divided into core parameters, auxiliary parameters, and non-critical variables.

[0059] The core parameters directly reflect the changing state of forest carbon sinks, including CO2 concentration gradients, vegetation spectral characteristics, and the NDVI index. These parameters have a significant impact on carbon sink estimation results and therefore have high importance weights. For example, the weight of the CO2 concentration parameter can be set to 0.9, and the weight of the vegetation spectral characteristics parameter can be set in the range of 0.8 to 1.0. Auxiliary parameters include indirect influencing factors such as air humidity and ambient temperature, with weights set in the range of 0.3 to 0.5. Non-critical variables include parameters that change in a short period of time but have a small impact on long-term carbon sink trends, such as instantaneous wind speed abrupt changes, with weights set below 0.1.

[0060] After parameter classification, different data filtering strategies are adopted for parameters of different levels. For core parameters, a weighted mean filtering method based on a sliding window is used. In this embodiment, the sliding window length N=10, that is, 10 consecutive sets of historical sampling data are used for analysis each time, and different weights are assigned according to the data time distance, so that data more recent than the current time has a higher influence, thereby reducing the impact of random noise while preserving the trend of forest carbon sink changes.

[0061] For auxiliary parameters, a standard mean filtering method is used to reduce minor fluctuations caused by environmental disturbances by calculating the average value of the data within a sliding window. For example, for air humidity data, since its variation frequency is high but its impact on carbon sink estimation is relatively small, only smoothing processing is needed to meet the fusion requirements.

[0062] For non-critical variables, no complex filtering is performed; only records of abnormal changes are retained. When a sudden change exceeding a preset range is detected in a non-critical variable, only the abnormal time point and abnormal amplitude are recorded for subsequent environmental state analysis, without continuously consuming communication and computing resources.

[0063] When the lightweight noise reduction mode detects that the data volume exceeds the preset transmission capacity, data compression processing is initiated. For core parameters, low-loss compression is used to ensure that the data accuracy remains within the preset range; for auxiliary parameters, the data storage accuracy is reduced, for example, by converting to an 8-16 bit data format; for non-critical variables, only abnormal event records are saved. Through hierarchical compression, the integrity of core carbon sequestration monitoring parameters is ensured while reducing the amount of data transmitted.

[0064] When the real-time distance between the UAV and the ground sensor node is detected to be greater than a first distance threshold and less than a second distance threshold, the system enters a transitional fusion mode. This mode is used to handle the data processing needs generated during the transition of the UAV from the near-range calibration area to the far-range inspection area.

[0065] In the transitional fusion mode, the airborne edge computing device simultaneously performs unidirectional ground calibration and selective compression processing. Specifically, it corrects the basic errors in the UAV-collected data based on historical baseline data from ground sensor nodes, but does not perform high-precision bidirectional cross-validation to reduce communication and computational burden. At the same time, it selectively compresses the data according to the importance level of the parameters.

[0066] Specifically, for core carbon sink parameters such as CO2 concentration and NDVI index, high data accuracy is maintained, with only necessary data format optimizations performed; for auxiliary parameters such as air humidity and ambient temperature, lightweight compression is implemented; and for non-critical variables, only essential state information is retained. This processing method achieves a smooth transition between accuracy and efficiency.

[0067] In step S400, the preprocessed data is dynamically transmitted based on the current communication bandwidth status, and collaborative verification is performed on the data from multiple overlapping UAV monitoring areas to generate verification data. Because forest areas typically have complex terrain, unstable communication conditions, and wide UAV inspection ranges, especially in mountainous and dense forest areas, wireless communication links are easily affected by factors such as terrain obstruction, vegetation attenuation, and changes in flight distance, causing UAV-collected data to not always maintain stable transmission. Therefore, this embodiment constructs a dynamic data transmission mechanism based on communication bandwidth status, adjusts the transmission priority of different types of monitoring data according to real-time communication resources, and utilizes data interaction between multiple UAV edge computing nodes to perform collaborative verification of overlapping monitoring areas, thereby improving the reliability of forest carbon sequestration monitoring data.

[0068] Specifically, during data transmission, the airborne edge computing device monitors the current air-to-ground communication link status in real time, acquiring communication status parameters such as available communication bandwidth, data transmission rate, link latency, and packet loss rate. These parameters determine whether the current communication environment meets the real-time transmission requirements of all monitoring data. When communication bandwidth decreases, the transmission priority of non-critical data is reduced, prioritizing the transmission of core parameters that significantly impact carbon sink estimation results. When communication bandwidth is sufficient, the data processing depth is increased, further enhancing multi-source monitoring data to improve data quality.

[0069] Before dynamic transmission, the airborne edge computing device first receives the generated preprocessed data and classifies and manages it according to data attributes. The preprocessed data includes core parameters, auxiliary parameters, and non-critical variables. Among them, the core parameters are data that directly affect the calculation results of forest carbon sink, including vegetation height, vegetation cover, soil organic carbon spectral characteristics, NDVI index, and CO2 concentration change characteristics; the auxiliary parameters are data used to assist in correcting the carbon sink estimation process, including air humidity, soil temperature, and ambient temperature; the non-critical variables are data that have a relatively small impact on carbon sink calculation and are mainly used for environmental status recording, including instantaneous air pressure, instantaneous solar radiation intensity, and short-term random disturbance parameters.

[0070] A dynamic transmission strategy is set based on the current communication bandwidth status. When the current communication bandwidth is detected to be less than a preset bandwidth threshold, a low-bandwidth transmission mode is entered. In this embodiment, considering the characteristics of the communication environment in forest areas, the bandwidth threshold in mountainous environments can be set to 2Mbps. When the actual communication bandwidth is lower than 2Mbps, the airborne edge computing device prioritizes the real-time uploading of core parameters.

[0071] Specifically, for core parameters, a real-time transmission mode is adopted, whereby the core carbon sink monitoring data collected by the UAV and preprocessed is directly uploaded to the cloud server or ground control center in chronological order of sampling time. Since core parameters directly participate in the subsequent carbon storage estimation process, if a significant time delay or data loss occurs, it will be impossible to fully recover the data through subsequent algorithms; therefore, they are given the highest transmission priority.

[0072] For example, during forest patrols, when drones collect data on vegetation height, canopy structure, and soil organic carbon spectral data for a certain area, the data is uploaded in real time even if the current communication bandwidth is insufficient, so as to ensure that the subsequent carbon sink estimation model can obtain complete input.

[0073] For auxiliary parameters, a compressed transmission mode is adopted. Specifically, the compression ratio is dynamically adjusted according to the current communication bandwidth. When the communication bandwidth decreases, the compression ratio is increased to reduce the amount of data transmitted; when the communication bandwidth approaches a threshold, the compression ratio is decreased to improve data accuracy. In this embodiment, the compression ratio of auxiliary parameters can be set within the range of 30% to 70%. For example, when the communication bandwidth is at a low level, air humidity and soil temperature data are transmitted using a 50% compression ratio, retaining only the main characteristic information affecting the carbon sink change trend, while reducing the communication load.

[0074] For non-critical variables, a local temporary storage mode is adopted. Specifically, the airborne edge computing device temporarily stores data such as instantaneous air pressure and instantaneous irradiance in the local storage module, and records the corresponding collection time, spatial location, and UAV status information. Once the communication link is restored to normal, the temporarily stored data is retransmitted according to a preset priority. Since these parameters are mainly used for environmental auxiliary analysis and do not directly participate in the core calculations for carbon sequestration estimation, the use of delayed transmission will not affect the overall monitoring results.

[0075] When the current communication bandwidth is detected to be greater than or equal to a preset bandwidth threshold, a high-bandwidth data enhancement mode is entered. In this embodiment, in areas with good communication environments such as plains and open areas, the bandwidth threshold can be set to 5 Mbps. When the actual communication bandwidth reaches 5 Mbps or more, the edge computing device no longer simply performs data transmission, but initiates data enhancement processing.

[0076] In data augmentation processing, the first step is to perform spatiotemporal correlation verification of multi-sensor data. Specifically, using an established unified spatiotemporal benchmark, data collected by different UAVs and different ground sensor nodes are matched in time and space to determine whether the data collected by multiple sensors in the same area and within the same time range meet the preset correlation relationship.

[0077] For example, when adjacent drones simultaneously cover a forest area, the NDVI index, vegetation cover, and CO2 concentration gradient data collected by the two drones are acquired separately and matched according to their spatial location and collection time. If the data collected by different drones conforms to the preset trend, the monitoring data for that area is considered reliable; if there is a significant deviation, further investigation is conducted to determine whether there are factors such as sensor anomalies, attitude changes, lighting changes, or communication anomalies.

[0078] Furthermore, based on the updated historical baseline data from ground sensors, anomalies are predicted and flagged in the current transmitted data. Specifically, the edge computing device accesses historical baseline data from the associated database and compares the currently acquired parameters with their historical normal range of variation. When a parameter is detected to exceed its historical range, the data is flagged as an anomaly, and the location, time, and magnitude of the parameter change are recorded.

[0079] For example, when the CO2 concentration in a certain area suddenly increases significantly, while historical data for the same period do not show similar changes, it is determined that the data may be affected by environmental disturbances or equipment malfunctions. The data is marked as abnormal, and its reliability is reduced in subsequent fusion calculations.

[0080] The bandwidth-adaptive transmission mechanism described above enables dynamic adjustments to data transmission strategies based on different communication environments. In low-bandwidth environments, a core parameter priority transmission mechanism ensures the integrity of key data for carbon sink estimation; in high-bandwidth environments, data augmentation processing improves the quality of monitoring data, achieving a dynamic balance between communication resources and monitoring accuracy.

[0081] To address situations where the monitoring ranges of multiple drones overlap, distributed collaborative verification is implemented to reduce the impact of single drone acquisition errors on forest carbon sink estimation results. When drones perform large-scale inspection tasks, adjacent drones may overlap in some areas, providing multiple independent observation results. Therefore, cross-validation can be performed using the consistency of multi-drone data.

[0082] Specifically, when the distance between adjacent drones is less than a preset communication range, a dynamic communication link is established between the edge computing devices of each drone. In this embodiment, when the distance between adjacent drones is less than or equal to 1 km, a temporary data exchange channel is established via millimeter-wave communication. Millimeter-wave communication features high transmission rate and low latency, which can meet the needs of real-time synchronous monitoring of parameters between drones.

[0083] After establishing a communication link, the edge computing devices of each UAV synchronously exchange carbon sink monitoring parameters for overlapping monitoring areas. These parameters include the NDVI index, estimated vegetation carbon storage, CO2 concentration trends, and other key ecological parameters. Based on the UAV location coordinates and collection time, data collected by multiple UAVs are mapped to the same spatial grid and compared for consistency.

[0084] During the specific verification process, parameter deviation thresholds are set to determine whether there are abnormal differences between the data collected by multiple drones. In this embodiment, the vegetation carbon storage deviation threshold can be set to 5%, and the CO2 concentration deviation threshold can be set to 3%. When the deviation of carbon sink parameters collected by multiple drones from the same area is less than the corresponding threshold, the data consistency of that area is considered to meet the requirements, and the fused data is used as valid verification data.

[0085] When the deviation between multiple drone acquisition results is greater than or equal to the parameter deviation threshold, it is determined that there is high measurement uncertainty in the overlapping area. The causes of this deviation may include changes in drone flight attitude, local airflow disturbances, multispectral sensor obstruction by tree canopies, changes in lighting conditions, or momentary equipment malfunctions.

[0086] At this point, a secondary data acquisition command is triggered, and the drone closest to the target area is selected based on the current location of each drone to perform the supplementary measurement task. Specifically, the drone closest to the target area adjusts its current flight path and enters the target area to perform secondary data acquisition. In this embodiment, the supplementary measurement range is controlled within 0.5 square kilometers to ensure the efficiency of the supplementary measurement while avoiding a significant impact on the overall inspection task.

[0087] After the secondary data collection, the carbon sink monitoring parameters for the target area are reacquired and compared with the initial data collection results. If the secondary data collection results are consistent with other UAV data, the monitoring results for that area are updated; if significant deviations still exist, the uncertainty level for that area is further increased.

[0088] Statistical analysis was performed on areas where secondary data acquisition was repeatedly triggered. When a region repeatedly triggers supplementary measurement tasks, it indicates that there may be long-term data instability factors in that region, such as severe terrain obstruction, complex vegetation structure, drastic environmental changes, or insufficient sensor observation conditions. Therefore, this region was marked as a high-uncertainty area, and its corresponding spatial location, number of anomalies, and deviation were recorded.

[0089] Information on high-uncertainty areas is synchronized to the subsequent inspection path optimization module to adjust the drone inspection strategy for the next cycle. During the next inspection, the frequency of visits to high-uncertainty areas is increased, along with the number of drone coverages and data collection density, thereby gradually reducing the data uncertainty in that area.

[0090] The aforementioned distributed collaborative verification mechanism enables data mutual verification among multiple drones. On one hand, by comparing data from overlapping areas, the reliability of forest carbon sink monitoring results is improved; on the other hand, by identifying high-uncertainty areas, a basis is provided for the dynamic optimization of subsequent drone inspection paths, enabling drones to prioritize access to areas with lower data reliability and improving overall monitoring efficiency.

[0091] In step S500, the output verification data is received. The verification data includes forest monitoring data after air-to-ground fusion, multi-UAV collaborative verification, and anomaly correction. The verification data includes multispectral image data, laser point cloud data, vegetation structure parameter data, and environmental parameter data collected by ground sensors. Among them, multispectral image data is used to obtain the spectral response characteristics of forest vegetation, laser point cloud data is used to obtain the spatial structure information of trees, and ground sensor data is used to provide high-precision environmental monitoring results for local areas.

[0092] This embodiment deploys a lightweight carbon sink estimation model in the edge computing device of an unmanned aerial vehicle (UAV). This allows the UAV to perform preliminary carbon sink quantification calculations at the edge after data collection and verification, without needing to upload all raw data to a remote server for processing. This reduces communication overhead and improves the efficiency of generating carbon sink monitoring results. The lightweight carbon sink estimation model includes a feature extraction module, a biomass calculation module, a carbon storage conversion module, and a spatial aggregation module.

[0093] First, the edge computing device performs feature extraction processing on the verification data, extracting key feature parameters to characterize the tree growth status from multi-source monitoring data. Specifically, it determines the spatial structure information of individual trees based on UAV laser point cloud data, including tree height, canopy height, canopy area, and tree spatial location; it extracts tree species spectral response characteristics and vegetation index characteristics, including NDVI index, red edge index, and reflectance information in different bands, based on multispectral image data; and it obtains environmental auxiliary parameters for the corresponding area based on ground sensor data, including soil temperature and humidity, air humidity, and carbon dioxide concentration changes.

[0094] The biomass of individual trees is determined based on the extracted tree characteristics. Specifically, the edge computing device calls pre-stored allometric growth equation parameters based on the tree species classification results. The allometric growth equation describes the relationship between tree size characteristics and biomass, and it is pre-calibrated according to different tree species, including diameter at breast height (DBH) parameters, tree height parameters, and corresponding correction coefficients.

[0095] In this embodiment, the aboveground biomass of a single tree is determined by a preset biomass calculation model based on the tree's diameter at breast height (D), tree height (H), and tree species-specific parameters a, b, and c. Specifically, during the calculation process, the edge computing device obtains the allometric growth model parameters corresponding to the tree species from the parameter table corresponding to the tree species' spectral classification label index, and calculates the biomass of a single tree using the tree's structural characteristics.

[0096] To adapt to the limited computing power of edge computing devices on drones, this embodiment performs a lightweighting of the traditional carbon sink estimation model. Specifically, lookup tables are pre-generated for the parameters of the allometric growth equations corresponding to different tree species and the results of nonlinear calculations involved in the process, and stored in the internal storage unit of the edge computing device. When the drone performs inspection tasks, it no longer performs complex power function calculations in real time. Instead, it directly queries the corresponding calculation parameters based on the input tree species label, diameter at breast height (DBH) range, and tree height range, and then obtains the biomass result of a single tree through simple multiplication and addition operations.

[0097] For example, for a specific tree species, a lookup table is pre-established containing biomass results corresponding to different diameter at breast height (DBH) and tree height ranges. When an edge computing device acquires the DBH and tree height parameters of a tree, it queries the lookup table based on the corresponding tree species label to obtain the corresponding biomass estimate. This method reduces the computational complexity of the edge computing process while ensuring the accuracy of carbon sink estimation, enabling drones to output carbon sink estimation results in real time during flight.

[0098] After obtaining the biomass of individual trees, the biomass is converted into carbon storage per tree according to a preset carbon conversion coefficient. Specifically, a corresponding carbon conversion coefficient is set according to different tree species, and the biomass of an individual tree is multiplied by the carbon conversion coefficient to obtain the fixed carbon content of an individual tree. The carbon conversion coefficient is preset according to the forest type and tree species characteristics, for example, it can be set in the range of 0.45 to 0.5, and adjusted according to different tree species.

[0099] The target monitoring area is divided into multiple spatial quadrats. The carbon storage of all individual trees within each quadrat is summed to obtain the corresponding quadrat-scale carbon sink value. In this embodiment, the forest area is divided into 20m × 20m spatial grids, with each grid corresponding to a carbon sink estimation unit. Edge computing devices statistically analyze the carbon storage of all individual trees within each grid and sum them to obtain the quadrat carbon sink value corresponding to that spatial grid.

[0100] For example, within a 20m × 20m quadrat area, multiple individual trees are identified, and the carbon storage of each tree is calculated. Then, the carbon storage of all individual trees is summed to obtain an estimated forest carbon sink for the current monitoring period in that quadrat area. This spatial quadrat aggregation method transforms the individual tree-scale estimation results into regional-scale carbon sink monitoring results, providing fundamental data for subsequent trend analysis.

[0101] After completing the carbon sink estimation, the carbon sink estimation results corresponding to the current monitoring period are generated. The carbon sink estimation results include spatial location information, monitoring time information, quadrat number, single-tree carbon storage information, quadrat carbon sink value, and data confidence information. Among them, spatial location information is used to determine the forest area corresponding to the carbon sink estimation results; monitoring time information is used to record the time when the results are generated; and quadrat carbon sink value is used to represent the current carbon storage level of the corresponding area.

[0102] The edge computing device locally caches the estimated carbon sink values ​​for each spatial location generated in the current monitoring period for subsequent calculations of carbon sink change rates. Specifically, it saves the carbon sink estimate C_current for each spatial grid at the current moment and retrieves the corresponding carbon sink estimate C_previous from the local cache for the previous monitoring period. The previous monitoring period can be set according to the monitoring task requirements, for example, to the monitoring results from 24 hours ago.

[0103] In step S600, the carbon sink change rate is calculated based on the carbon sink estimation results of the current monitoring period and the previous monitoring period. Specifically, the carbon sink growth or decline trend of the corresponding region is determined based on the changes in the carbon sink estimation values ​​of the two monitoring periods at the same spatial location.

[0104] When the current carbon sink estimate for a region increases significantly compared to the previous period, it indicates that the region may be experiencing rapid vegetation growth, improved environmental conditions, or enhanced carbon absorption capacity. Conversely, when the carbon sink estimate decreases significantly, it indicates that the region may be experiencing vegetation degradation, environmental anomalies, or ecological disturbances. Therefore, using the carbon sink change rate as an important reference factor for optimizing UAV inspection paths can allow UAVs to prioritize areas with significant carbon sink changes.

[0105] Furthermore, this embodiment constructs a total path cost function for UAV inspection path optimization. The total path cost function includes terrain factor cost, meteorological factor cost, sensor factor cost, and carbon sink change rate driving cost term.

[0106] Specifically, the terrain factor cost is used to describe the path cost caused by terrain undulations during UAV flight. When the terrain slope is large and the altitude change is significant, the difficulty of UAV flight increases, and the corresponding terrain factor cost increases.

[0107] Meteorological factor costs are used to describe the impact of current weather conditions on the flight safety of drones. Meteorological factors include real-time wind speed, precipitation probability, temperature, and meteorological stability indicators. When wind speed increases or the precipitation probability rises, the flight risk of drones increases, and the corresponding meteorological factor costs increase.

[0108] Sensor factor cost is used to describe the coverage of a ground sensor network. When the density of ground sensor nodes in a certain area is low or the data reliability is insufficient, the access priority of that area is increased to obtain more monitoring data.

[0109] The carbon sink change rate-driven cost term is used to reflect the dynamic changes in forest carbon sinks. Unlike traditional path planning methods that only consider flight safety and environmental factors, this embodiment directly incorporates the carbon sink change trend into the path optimization process, enabling UAVs to proactively go to areas with significant carbon sink changes for focused monitoring.

[0110] Specifically, the carbon sink change rate is calculated based on the current monitoring period and the previous monitoring period: when the carbon sink change rate in a certain area exceeds the preset change rate threshold, the area is determined to be a dynamic carbon sink change area, and the access weight of the area in the total path cost function is increased.

[0111] For example, when the carbon sink change rate in a certain area reaches more than 20%, the area is marked as a key monitoring area, and the weight of the carbon sink driving cost item in that area is increased to twice that of ordinary areas, so that subsequent path optimization algorithms are more inclined to generate inspection routes that pass through that area.

[0112] Through this method, drone inspection routes no longer follow fixed cycles and routes, but are dynamically adjusted according to changes in the forest ecological state. When carbon sink conditions in a certain area change drastically, the frequency of drone visits is increased; when the carbon sink conditions in a certain area are stable, the access priority is reduced, thereby achieving a rational allocation of inspection resources.

[0113] This embodiment employs a particle swarm optimization algorithm to solve the total path cost function, obtaining the optimal inspection path for the UAV in the next monitoring cycle. Specifically, each candidate inspection route is treated as a particle, and the path node position is treated as the particle state. The path scheme is continuously updated through information interaction between particles.

[0114] In the particle swarm optimization process, preset population size, number of iterations, and learning factors are set to enable the particle swarm to search for the path scheme with the minimum total cost within a finite time. In this embodiment, the particle swarm size can be set to 50, the maximum number of iterations to 30, the inertia weight to 0.7, and the acceleration factors c1 and c2 to 1.5.

[0115] In each iteration, the total cost corresponding to each candidate path is calculated, and the particle motion direction is updated based on the current optimal path and the global optimal path. Finally, the inspection path with the lowest total cost is obtained, and this path is used as the execution path for the next UAV inspection mission.

[0116] To ensure the flight safety of UAVs in abnormal environments, this embodiment sets up a dynamic path reconstruction mechanism. When the real-time wind speed exceeds a preset safety threshold, or the ground sensor node density deviation exceeds a preset range, an emergency path reconstruction is triggered.

[0117] Specifically, when the real-time wind speed exceeds 10 m / s, it is determined that the current environment does not meet the conditions for safe flight of the UAV, and a preset backup safe route is invoked to guide the UAV to avoid dangerous areas; when the ground sensor node density deviation is detected to exceed 20%, it is determined that some sensor nodes may be malfunctioning or have communication abnormalities, and the spacing between path nodes is readjusted to improve the effective monitoring coverage.

[0118] During the emergency path reconstruction process, the generated path is further smoothed to reduce energy consumption and attitude change risks caused by the UAV’s sharp turns, enabling the UAV to maintain stable flight in complex environments.

[0119] Furthermore, to achieve dynamic adjustment of path planning weights, this embodiment adjusts the weight parameters in the total path cost function based on the real-time meteorological risk level. When meteorological conditions are favorable, the weight of meteorological factors is reduced, and the weight of carbon sink change-driven factors is increased, causing the UAV to focus more on areas of ecological change; when there are risky environments such as strong winds or heavy rain, the weight of meteorological factors is increased to prioritize the flight safety of the UAV.

[0120] For example, in clear and light wind conditions, the weight of meteorological factors is set relatively low; when the wind speed exceeds 8 m / s, the weight of meteorological factors is increased; when the wind speed exceeds 10 m / s or the probability of precipitation exceeds 80%, the weight of meteorological factors is increased to 0.6, and the weights of other factors are readjusted according to the normalization rule so that the sum of the weights of all influencing factors remains at 1.

[0121] In one specific embodiment, a subtropical evergreen broad-leaved forest area of ​​approximately 50 square kilometers was selected as the monitoring target. The area has complex terrain, including continuous mountains, low-lying valleys, and gentle slopes. The forest vegetation is predominantly evergreen broad-leaved forest, with some areas containing trees of different ages. Due to the significant topographic relief, uneven vegetation distribution, and insufficient coverage by traditional fixed monitoring points, the monitoring method proposed in this invention, based on air-ground collaborative data acquisition, data fusion, dynamic preprocessing, and optimization of carbon sink change-driven paths, is employed to achieve continuous and refined monitoring of forest carbon storage changes.

[0122] During the deployment phase, based on the forest area, terrain features, and vegetation distribution, an air-ground collaborative monitoring system was constructed, consisting of a ground sensor network, a drone inspection platform, edge computing devices, and a dynamic path planning platform. One hundred ground sensor nodes were deployed within the target forest area, with an average spacing of approximately 500 meters between adjacent nodes, ensuring that each sensor node could cover the main forest ecological areas. Each ground sensor node was equipped with a carbon dioxide concentration sensor, an air humidity sensor, and a vegetation spectral acquisition module to continuously acquire forest environmental status data.

[0123] To achieve unified data management across different sensor nodes, a unique identifier is assigned to each ground sensor node. In this embodiment, the identifier adopts the structure of "regional code + device number + deployment date," such as "H N-001-20230101," where "HN" represents the Hunan regional code, "001" represents the 001st monitoring node, and "20230101" represents the device deployment date. Based on the above identifier system, a four-dimensional relational database is established. The database records the spatial coordinates, acquisition parameter types, historical operational data, device calibration records, and historical baseline information corresponding to each node, providing a data foundation for subsequent UAV data matching, error correction, and anomaly detection.

[0124] Simultaneously, three multi-rotor drones were deployed within the forest area as an aerial mobile monitoring platform. The drones have a flight time of 60 minutes, a maximum payload of approximately 5 kg, and are equipped with multispectral cameras, lidar devices, and edge computing modules. The multispectral cameras are used to acquire the spectral response characteristics of forest vegetation, the lidar is used to acquire information on tree height, canopy structure, and forest stand spatial distribution, and the edge computing device is used to perform data fusion, carbon sink estimation, and path optimization calculations. In this embodiment, the edge computing device uses an ARM Cortex-A76 processor and is configured with 512GB of storage space, which can meet the real-time data processing needs during drone inspections.

[0125] A lightweight carbon sink estimation model, an ensemble Kalman filter fusion module, an IEEE 1588 precise time synchronization module, and a parameter weight management module are pre-deployed within the edge computing device. The lightweight carbon sink estimation model calculates forest carbon storage based on tree structure parameters and vegetation spectral characteristics; the ensemble Kalman filter module fuses UAV remote sensing data with ground sensor data; the time synchronization module ensures consistent data acquisition time between air and ground; and the parameter weight management module adjusts path planning parameters based on communication status, meteorological conditions, and carbon sink changes.

[0126] Furthermore, a distributed collaborative verification mechanism for unmanned aerial vehicles (UAVs) is constructed. In this embodiment, the maximum collaborative communication distance between UAVs is set to 1 km. When multiple UAVs simultaneously cover the same area, a temporary data exchange link is established via millimeter-wave communication to synchronously compare the NDVI index, vegetation carbon storage estimation results, and CO2 concentration change data within the overlapping area. A vegetation carbon storage deviation threshold θ = 5% is set. When the deviation of the carbon storage estimation results obtained by different UAVs for the same area reaches or exceeds 5%, a supplementary data collection mechanism is triggered.

[0127] Furthermore, a dynamic path planning model was established, using meteorological data, terrain data, sensor distribution data, and carbon sink change data as the basis for UAV path optimization. Local meteorological station data in the forest area was accessed to obtain meteorological parameters such as wind speed and precipitation probability in real time, while a digital terrain database was used to obtain terrain information such as regional slope and altitude. The path planning model was solved using a particle swarm optimization algorithm, with 50 particles and a maximum number of iterations of 30. The initial weights of the total path cost function were set as follows: terrain factor weight α = 0.3, meteorological factor weight β = 0.3, sensor factor weight γ = 0.2, and carbon sink change driving factor weight δ = 0.2. A carbon sink change rate threshold of 20% was set; when the carbon sink change rate in a certain area exceeded this threshold, the inspection priority of that area was increased.

[0128] In actual monitoring, the first step is to implement air-ground collaborative data acquisition and identification management. Ground sensor nodes continuously collect forest environmental data at a sampling cycle of 2 to 10 minutes and upload the results to a four-dimensional correlation database. For example, if a ground sensor node collects data showing a CO2 concentration of 420 ppm, an air humidity of 68%, and a vegetation spectral index of 0.82 in the current area, this data is then bound to the corresponding node identifier, collection time, and spatial coordinates and stored in the database.

[0129] The drone performs inspection tasks based on dynamic path planning. During the initial monitoring period, the drone scans the area according to a preset coverage strategy, covering approximately 5 square kilometers of forest area every 30 minutes. During flight, multispectral cameras continuously acquire forest canopy image data, while lidar simultaneously obtains spatial structure information of trees. Onboard edge computing devices retrieve historical data from nearby ground sensor nodes based on the current location. For example, when the drone flies near the node “H N-015-20230101”, it quickly retrieves 30 days of historical monitoring data for that node using its node identifier, providing a reference for subsequent data calibration.

[0130] Subsequently, spatiotemporal synchronization calibration is performed. Before takeoff and during inspection, the UAV is synchronized with the GPS clock of the ground sensor node via the IEEE 1588 precise time protocol to keep the timestamp deviation between the UAV-collected data and the ground sensor data within 10ms.

[0131] For example, when a drone flies at an altitude of 100m, the spatial resolution of the images acquired by the multispectral camera reaches 0.05m / pixel, while the effective monitoring radius of the ground CO2 sensor is about 10m. Due to the significant difference in spatial scale between the two, the drone images are automatically resampled to convert the high-resolution images into a 10m×10m spatial grid, so that the drone's area remote sensing data and the ground sensor's point monitoring data have a consistent spatial scale.

[0132] After completing temporal and spatial matching, an ensemble Kalman filter algorithm is used for data assimilation and fusion. Specifically, UAV remote sensing data is used as a priori estimates, and ground sensor monitoring data is used as observation updates. A state vector containing CO2 concentration and NDVI index is constructed, with the ensemble set to 50 members. The fused forest state data is obtained through a grid-by-grid update method. For example, for a 10m×10m grid area, the UAV remote sensing results show an NDVI index of 0.76, while the ground sensor monitoring results show that the vegetation state in this area is slightly higher than the remote sensing estimate. The fusion result is adjusted by ensemble Kalman filtering to generate an air-ground fused dataset that is closer to the actual state.

[0133] Subsequently, a distance-driven data preprocessing mode switch is executed. When the UAV is less than 50 meters away from the ground sensor node, it enters a high-precision calibration mode. At this time, the data collected by the UAV highly overlaps with the monitoring area of ​​the ground sensor, and the edge computing device calls the historical baseline data of the corresponding node over the past 30 days to correct the errors in the UAV's data collection results.

[0134] For example, a ground sensor node might have a historical average CO2 concentration of 415 ppm, while a drone's current reading is 425 ppm. The deviation between the two is calculated, and the drone's data is corrected based on the historical baseline. The corrected data is then sent back to the ground sensor node, which further cross-validates it with its own real-time readings. When the deviation exceeds 5%, the ground sensor's self-calibration process is triggered, and the updated calibration results are written to the four-dimensional correlation database.

[0135] When the drone is more than 200 meters away from the ground sensor node, it enters a lightweight noise reduction mode. At this time, the drone is in a large-scale inspection state, prioritizing data processing efficiency. The CO2 concentration parameter is assigned a weight of 0.9 and a weighted mean filter with a window length of N=10 is used; the air humidity parameter is assigned a weight of 0.4 and a normal mean filter is used; for low-impact parameters such as instantaneous wind speed, only records of abnormal changes are retained.

[0136] When the drone is within a distance of 50 to 200 meters, it enters a transitional fusion mode, performs unidirectional ground calibration on the core carbon sink parameters, and compresses the auxiliary parameters to achieve a balance between data accuracy and computational efficiency.

[0137] After data preprocessing is completed, bandwidth-adapted transmission is performed. When the drone is in a mountainous area and the communication bandwidth drops below 2Mbps, the edge computing device initiates a low-bandwidth transmission strategy.

[0138] Among them, vegetation height and soil organic carbon spectral data are used as primary parameters and are uploaded in real time; air humidity and soil temperature are used as secondary parameters and are transmitted with a 50% compression rate; instantaneous air pressure and light intensity are used as tertiary parameters and are temporarily stored in the drone's local storage, to be retransmitted after communication is restored.

[0139] When the drone enters an open area and the communication bandwidth recovers to above 5Mbps, data augmentation processing is initiated to perform spatiotemporal correlation analysis on data from multiple sensors and to detect abnormal changes using historical baseline data. For example, if the CO2 concentration in a certain area suddenly increases, but there is no similar trend in the same period in history, the data is automatically marked as an abnormal candidate value.

[0140] Distributed collaborative verification is then performed. The three drones synchronize the NDVI index of the overlapping area every minute via millimeter-wave communication. When it is detected that the NDVI indices calculated by two drones for the same area are 0.81 and 0.87 respectively, with a deviation of 7.4%, exceeding the preset 5% threshold, it is determined that there is a high degree of uncertainty in that area.

[0141] At this point, the distance between the current positions of the three drones and the target area is calculated. The drone with the closest distance is selected to perform secondary data collection, with the supplementary measurement area controlled within 0.5 square kilometers. If a certain area triggers supplementary measurement for multiple consecutive monitoring cycles, the area is marked as a high uncertainty area, and this information is sent to the path planning module to increase the frequency of subsequent inspections.

[0142] After data verification is completed, edge carbon sequestration is performed. The edge computing device extracts the diameter at breast height (DBH), tree height, and tree species spectral classification information of individual trees from the verification data, and calls the corresponding allometric growth parameters to calculate the biomass of individual trees.

[0143] For example, for a certain evergreen broad-leaved tree, its diameter at breast height is identified as 25cm and its height as 18m. The biomass of a single tree is calculated according to the corresponding allometric growth model for this tree species, and then multiplied by the corresponding carbon conversion coefficient to obtain the carbon storage of the tree.

[0144] Subsequently, the carbon storage of all individual trees within a 20m×20m quadrat area is summed to obtain the carbon sink value for that quadrat area. For example, if a total of 38 trees are identified in a quadrat area, and the cumulative carbon storage is 12.6 tons, then the carbon sink estimation result for that spatial grid is recorded as 12.6 tons.

[0145] The carbon sink value at each spatial location in the current monitoring period is recorded as C_current, and the corresponding result C_previous from the monitoring period 24 hours ago is called to calculate the carbon sink change rate R_carbon.

[0146] During the path optimization process, when a carbon sink change rate in a certain area reaches or exceeds 20%, that area is determined to be a region with dynamic carbon sink changes. For example, if the carbon storage in a valley area was 100 tons 24 hours ago and the current monitoring result is 125 tons, then the change rate reaches 25%, exceeding the 20% threshold.

[0147] At this point, the path planning engine increases the weight of the carbon sink-driven cost item in this area, raises the access weight of this area to twice that of ordinary areas, and calls the particle swarm optimization algorithm again to search for the optimal inspection path, so that the drone can fly to this area first for further monitoring.

[0148] Under normal weather conditions, particle swarm optimization calculations are performed again every 20 seconds to dynamically update the inspection path. When the weather station predicts that the wind speed will reach 11 m / s in the next 2 hours, which is higher than the safe flight threshold of 10 m / s for drones, the weight of the meteorological factor β is increased to 0.6, and the weights of other factors are adjusted according to the weight normalization rule, while triggering emergency path reconstruction.

[0149] During the emergency path reconstruction process, backup safe routes are invoked, the distance between drone waypoints is adjusted, and the path curve is smoothed to enable drones to avoid high-risk areas and safely complete monitoring tasks.

[0150] In summary, the unified time and space benchmarks, combined with filtering and fusion, improve the accuracy of carbon sink estimation; it breaks away from the traditional one-way mode of calibrating UAVs solely on the ground, continuously correcting for aging and drift errors of ground sensors, ensuring a stable and reliable long-term monitoring benchmark; the path automatically adjusts the inspection frequency according to carbon sink fluctuations, improving monitoring efficiency under the same hardware conditions; it only transmits core data in stages, ensuring stable operation in mountainous and low-bandwidth environments; cross-verification by multiple UAVs reduces accidental measurement errors from single devices, improving the reliability of carbon sink data collection; the entire link is closed-loop, with data exchange and feedback between each link, rather than independent modules, giving the entire monitoring system self-optimization capabilities; structured equipment identification and a four-dimensional database facilitate carbon sink verification and source tracing audits; preprocessing thresholds, path weights, and transmission strategies can all be dynamically adjusted according to forest canopy density, terrain, and weather, adapting to multiple scenarios including dense forests, sparse forests, mountains, and plains; and it automatically switches to safe flight routes in severe weather, reducing the flight risks of UAVs.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for monitoring forest carbon sinks, characterized in that, include: An air-ground collaborative data acquisition architecture is built based on UAVs and ground sensor networks, and identifiers are assigned to ground sensor nodes in each region. During the UAV inspection process, the spatiotemporal correlation parameters between the UAV and ground sensor nodes are matched based on the identifier, and the data collected by the UAV and the data collected by the ground sensor network are spatiotemporally fused to generate an air-ground fusion dataset. Based on the real-time distance in the spatiotemporal correlation parameters, the air-to-ground fusion dataset is preprocessed to obtain preprocessed data. The data preprocessing includes a high-precision calibration mode, a lightweight noise reduction mode, and a transition fusion mode. The preprocessed data is dynamically transmitted according to the communication bandwidth status. The preprocessed data transmitted in the overlapping monitoring area is then collaboratively verified to obtain verification data. The verification data is processed based on the carbon sink estimation model to generate carbon sink estimation results; Based on the carbon sink estimation results of the current monitoring cycle and the previous monitoring cycle, the carbon sink change rate is determined, and the carbon sink change rate is used as a constraint to optimize the next UAV inspection path. The high-precision calibration modes include: When the real-time distance is less than or equal to the first distance threshold, the high-precision calibration mode is performed: based on the identifier, historical baseline data of the ground sensor node is retrieved from the associated database; based on the historical baseline data, the error is determined, and the UAV-collected data in the air-ground fusion dataset is corrected; the corrected UAV-collected data is sent back to the ground sensor node for cross-validation; when the deviation is greater than the deviation threshold, the ground sensor node performs self-calibration and updates the historical baseline data in the associated database. The lightweight noise reduction mode includes: When the real-time distance is greater than or equal to the second distance threshold, the lightweight noise reduction mode is implemented: the carbon sink monitoring parameters in the air-ground fusion dataset are classified and labeled, and the carbon sink monitoring parameters after classification and labeling are adaptively filtered and compressed using a sliding window. The transition fusion mode includes: When the real-time distance is greater than the first distance threshold and less than the second distance threshold, the transition fusion mode is performed: the air-to-ground fusion dataset is subjected to one-way ground calibration and selective compression processing.

2. The method for monitoring forest carbon sinks according to claim 1, characterized in that, It also includes establishing a related database; the related database includes latitude and longitude coordinates, parameter types, calibration records, and historical baseline data.

3. The method for monitoring forest carbon sinks according to claim 2, characterized in that, When generating the air-to-ground fusion dataset, the following are included: Using the clock of the ground sensor node as a reference, the onboard clock of the UAV is synchronized through a precise time protocol for time synchronization calibration. Based on the real-time flight altitude of the UAV and the spatial distribution density of the ground sensor node, the spatial resolution of the UAV-collected data is adjusted to match the effective monitoring radius of the ground sensor node for spatial scale matching. Using an ensemble Kalman filter algorithm, the UAV-collected data after the time synchronization calibration and spatial scale matching are assimilated and fused with the data collected by the ground sensor node to generate the air-ground fusion dataset.

4. The method for monitoring forest carbon sinks according to claim 3, characterized in that, When dynamically transmitting the preprocessed data based on the communication bandwidth status, the following is included: When the communication bandwidth is less than the bandwidth threshold, the core parameters in the preprocessed data are transmitted in real time, the auxiliary parameters in the preprocessed data are transmitted in compressed form, and the non-critical variables in the preprocessed data are temporarily stored locally. When the communication bandwidth is greater than or equal to the bandwidth threshold, data augmentation processing is initiated. The data augmentation processing includes performing spatiotemporal correlation verification on multi-sensor data and predicting and marking outliers based on the historical baseline data.

5. The method for monitoring forest carbon sinks according to claim 4, characterized in that, When performing collaborative verification on the preprocessed data transmitted in the overlapping monitoring area to obtain verification data, the following steps are included: A dynamic communication link is established between the edge computing devices of adjacent UAVs to synchronize the carbon sink parameters of the overlapping monitoring area. When the carbon sink parameter deviation of several UAVs is greater than or equal to the parameter deviation threshold, the UAV closest to the target area is triggered to perform secondary data collection. The area that triggers secondary data collection multiple times in a row is marked as a high uncertainty area, and the access frequency of the next UAV inspection is increased.

6. The method for monitoring forest carbon sinks according to claim 5, characterized in that, When processing the verification data based on the carbon sink estimation model to generate carbon sink estimation results, the following steps are included: Tree characteristics are extracted from the verification data to determine the biomass of a single tree, and the carbon storage of a single tree is obtained based on the carbon conversion coefficient. The carbon storage of a single tree in a preset quadrat is summarized as the quadrat carbon sink value to generate the carbon sink estimation result.

7. The method for monitoring forest carbon sinks according to claim 6, characterized in that, When optimizing the next UAV inspection path using the carbon sink change rate as a constraint, the following are included: A total path cost function is constructed, which includes terrain factor cost, meteorological factor cost, sensor factor cost, and carbon sink change rate driving cost term. When the carbon sink change rate is greater than the change rate threshold, the access weight in the carbon sink change rate driving cost term is increased. The minimum cost path is solved using the particle swarm optimization algorithm, which is used as the next UAV inspection path. When the real-time wind speed is greater than the wind speed threshold or the ground sensor node density deviation is greater than the density deviation threshold, an emergency path reconstruction is performed. The emergency path reconstruction includes calling the backup safe route, correcting the path node spacing, and smoothing the path.

Images