A Blockchain-Based Method and System for Monitoring Agricultural Carbon Sequestration

Abstract

This invention discloses a blockchain-based method and system for monitoring agricultural carbon sinks, comprising: collecting and preprocessing multi-source monitoring data on agricultural carbon sinks; combining plot-period carbon sink contribution feature vectors based on plot-period carbon sink basic data; conducting multi-source data quality assessment to generate plot-period data credibility coefficients; inputting the carbon sink contribution branch coding layer to generate carbon sink contribution branch representations; inputting the branch evidence parameter generation layer to generate branch evidence parameters; inputting the credible evidence synthesis layer to synthesize carbon sink evidence parameters; inputting the dual modulation correction layer to generate credible carbon sink contribution feature vectors; inputting the net carbon sink output layer to generate plot-period net carbon sink amounts; and forming on-chain evidence records of plot-period carbon sinks. This invention employs deep evidence regression and blockchain evidence storage to achieve credible accounting of agricultural carbon sinks, possessing the advantages of stable results, traceable processes, and tamper-proof nature.

Description

Technical Field

[0001] This invention relates to the field of agricultural carbon sequestration monitoring, and in particular to a blockchain-based method and system for agricultural carbon sequestration monitoring. Background Technology

[0002] Currently, in the field of agricultural carbon sequestration monitoring, existing technologies typically collect agricultural carbon sequestration-related data through soil sensors, remote sensing images, meteorological data, and agricultural records, and then calculate the carbon sequestration of a plot based on carbon emission factors or empirical accounting models. However, multi-source monitoring data suffers from differences in sources, inconsistent collection completeness, temporal and spatial matching errors, and interference from outlier data. Existing methods often directly involve data in the calculation, lacking a reliability assessment of data quality, resulting in insufficient stability of net carbon sequestration calculation results.

[0003] Meanwhile, while existing technologies can upload carbon sequestration results to the blockchain for evidence storage, most only record the final carbon sequestration result at a single point, lacking layered evidence storage of basic data on land parcel-cycle carbon sequestration, credible carbon sequestration contribution characteristics, and the formation process of net carbon sequestration. Furthermore, existing carbon sequestration accounting models typically only output prediction results, failing to combine model evidence uncertainty with data credibility for feature modulation, making it difficult to achieve credible generation and on-chain traceable verification of the carbon sequestration accounting process. Summary of the Invention

[0004] One objective of this invention is to propose a blockchain-based method and system for monitoring agricultural carbon sinks. This invention employs deep evidence regression and blockchain notarization to achieve reliable accounting of agricultural carbon sinks, and has the advantages of stable results, traceable process, and tamper-proof performance.

[0005] According to an embodiment of the present invention, a blockchain-based method for monitoring agricultural carbon sinks includes: Collect multi-source monitoring data on agricultural carbon sequestration and preprocess it to form basic data on the periodic carbon sequestration of land plots; Carbon sink source characteristics are extracted from the basic data of land parcel periodic carbon sinks and combined to form a feature vector of land parcel periodic carbon sink contribution. Based on the basic data of carbon sequestration in the land parcel cycle, a multi-source data quality assessment is conducted to generate a reliability coefficient for the land parcel cycle data. The feature vector of carbon sink contribution of the land parcel cycle is input into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation is generated through the branch evidence parameter generation layer. The branch evidence parameters are then corrected in the credible evidence synthesis layer according to the credibility coefficient of the land parcel cycle data to synthesize the carbon sink evidence parameters. Carbon sink evidence parameters, land parcel cycle data credibility coefficient, and land parcel cycle carbon sink contribution feature vector are input into the dual modulation correction layer. An uncertainty gate coefficient is generated based on the carbon sink evidence parameters. The land parcel cycle data credibility coefficient and uncertainty gate coefficient are fused to form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel cycle carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. Input the credible carbon sink contribution feature vector and carbon sink evidence parameters into the net carbon sink output layer to generate the plot's periodic net carbon sink amount. Write the basic data of land parcel cycle carbon sink, the credible carbon sink contribution feature vector, and the evidence information corresponding to the net carbon sink of land parcel cycle into the blockchain to form the on-chain evidence record of land parcel cycle carbon sink.

[0006] Optionally, the agricultural carbon sink multi-source monitoring data includes soil monitoring status data, crop growth status data, meteorological environment status data, agricultural operation status data, remote sensing monitoring status data, and data quality record data. The preprocessing includes field unification, time alignment, spatial matching, missing data completion, and anomaly removal.

[0007] Optionally, the formation of the cyclic carbon sink contribution feature vector of the land parcel includes: Read crop growth status data, perform stage correction on biomass change status, and generate crop carbon sequestration characteristics; Based on soil monitoring status data, soil carbon sequestration characteristics are generated. Read agricultural operation status data, unify the direction of each agricultural operation status's effect on carbon fixation increase and carbon emission reduction, and generate agricultural emission reduction characteristics. Read meteorological and environmental status data and remote sensing status data, perform statistical coding of monitoring cycles, and generate environmental regulation characteristics; Based on crop carbon sequestration characteristics, soil carbon sequestration characteristics, agricultural emission reduction characteristics, and environmental regulation characteristics, a plot-based periodic carbon sink contribution characteristic vector is formed.

[0008] Optionally, the generation of the reliability coefficient of the land parcel periodic data includes: Read the data quality records from the basic data of the periodic carbon sequestration of the land parcel and generate the data quality status; The credibility of the data quality status is fused to generate the credibility coefficient of the land parcel periodic data.

[0009] Optionally, the synthesis of the carbon sink evidence parameters includes: Input the periodic carbon sink contribution feature vector of the land parcel into the carbon sink contribution branch encoding layer of the deep evidence regression carbon sink accounting model to generate the corresponding carbon sink contribution branch representation. The deep evidence regression carbon sink accounting model includes a carbon sink contribution branch coding layer, a branch evidence parameter generation layer, a credible evidence synthesis layer, a dual modulation correction layer, and a net carbon sink output layer. The branch evidence parameter generation layer performs evidence regression mapping on the carbon sink contribution branch representation to generate branch evidence parameters; The credible evidence synthesis layer reads the credibility coefficient of the land parcel periodic data, generates the branch credibility correction coefficient based on the land parcel periodic data credibility coefficient, corrects the branch evidence parameters based on the branch credibility correction coefficient, and generates the corrected branch evidence parameters. Evidence is synthesized based on the revised branch evidence parameters to generate carbon sink evidence parameters.

[0010] Optionally, the generation of the credible carbon sink contribution feature vector includes: Carbon sink evidence parameters, land parcel period data credibility coefficient, and land parcel period carbon sink contribution feature vector are input into a dual modulation correction layer, which includes an uncertainty analysis unit, a modulation vector generation unit, a dual modulation parameter generation unit, and a feature correction unit. The uncertainty analysis unit reads the carbon sink evidence parameters, performs evidence strength normalization and uncertainty mapping on the carbon sink evidence parameters, and generates uncertainty gating coefficients. The reliability coefficient and uncertainty gating coefficient of the land parcel periodic data are input into the modulation vector generation unit to generate a reliability modulation vector and an uncertainty modulation vector; In the dual modulation parameter generation unit, a scaling modulation vector and an offset modulation vector are generated based on the confidence modulation vector and the uncertainty modulation vector. The feature correction unit reads the plot periodic carbon sink contribution feature vector, scaling modulation vector, and offset modulation vector to generate a reliable carbon sink contribution feature vector.

[0011] Optionally, the generation of the site's periodic net carbon sink includes: The credible carbon sink contribution feature vector and carbon sink evidence parameters are input into the net carbon sink output layer, which includes a feature evidence fusion unit, a carbon sink mapping unit, an evidence mean correction unit, and a net carbon sink quantity output unit. The feature evidence fusion unit reads the credible carbon sink contribution feature vector and carbon sink evidence parameters, and generates a feature evidence fusion representation. The feature evidence fusion representation is input into the carbon sink mapping unit to generate the initial net carbon sink mapping amount; The evidence mean correction unit calculates the evidence-corrected net carbon sink mapping based on the initial net carbon sink mapping and the characteristic evidence fusion representation. The evidence-corrected net carbon sink mapping is input into the net carbon sink output unit, and scale restoration is performed based on the monitoring cycle length and agricultural plot area corresponding to the plot cycle carbon sink basic data to generate the plot cycle net carbon sink.

[0012] Optionally, the formation of the evidence record on the land parcel's periodic carbon sink chain includes: Read the basic data of carbon sequestration in the periodic land parcel, the credible carbon sequestration contribution feature vector, and the net carbon sequestration in the periodic land parcel; The basic data of the periodic carbon sequestration of land parcels are serialized and hashed to obtain the basic data hash. The trusted carbon sink contribution feature vector is subjected to vector serialization and hashing to obtain the trusted feature hash. Agricultural plot identifiers and monitoring cycle identifiers are read from the basic data of plot period carbon sequestration. The results of plot period net carbon sequestration, agricultural plot identifiers and monitoring cycle identifiers are serialized and hashed to obtain the result hash. The basic data hash, trusted feature hash, and result hash are combined to form the carbon sink process hash input sequence, and the carbon sink process hash input sequence is hashed to obtain the carbon sink process root hash. The corresponding basic data generation time is read from the basic data of the land plot cycle carbon sink, the corresponding credible feature generation time is read from the credible carbon sink contribution feature vector, and the corresponding net carbon sink generation time is read from the net carbon sink amount of the land plot cycle. The agricultural land plot identifier, monitoring cycle identifier, basic data hash, credible feature hash, result hash, carbon sink process root hash, data version number and the above generation time are written into the blockchain to form the on-chain evidence record of the land plot cycle carbon sink.

[0013] An agricultural carbon sequestration monitoring system based on blockchain according to an embodiment of the present invention includes: The data processing module is used to collect multi-source monitoring data of agricultural carbon sequestration and preprocess it to form basic data of plot periodic carbon sequestration. The feature construction module is used to extract carbon sink source features based on the basic data of land parcel periodic carbon sinks and combine them to form a feature vector of land parcel periodic carbon sink contribution. The quality assessment module is used to conduct multi-source data quality assessment based on the basic data of land parcel periodic carbon sequestration and generate the reliability coefficient of land parcel periodic data. The evidence parameter generation module is used to input the feature vector of the carbon sink contribution of the land parcel period into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation generates branch evidence parameters through the branch evidence parameter generation layer. The branch evidence parameters are corrected according to the credibility coefficient of the land parcel period data in the credible evidence synthesis layer to synthesize the carbon sink evidence parameters. The dual modulation correction module is used to input carbon sink evidence parameters, land parcel period data credibility coefficient and land parcel period carbon sink contribution feature vector into the dual modulation correction layer and form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel period carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. The output module is used to generate the periodic net carbon sink of the land parcel based on the credible carbon sink contribution feature vector and carbon sink evidence parameters. The on-chain evidence storage module is used to generate on-chain evidence storage records for the periodic carbon sink of land parcels.

[0014] The beneficial effects of this invention are: This invention proposes a blockchain-based method and system for monitoring agricultural carbon sinks. It performs multi-source data quality assessment based on plot-cycle carbon sink baseline data, generates a plot-cycle data credibility coefficient, and introduces this coefficient into the credible evidence synthesis layer of a deep evidence regression carbon sink accounting model. This ensures that data credibility is not simply used as an additional weight in the final result calculation, but rather to correct branch evidence parameters and synthesize carbon sink evidence parameters. Through this processing, the evidentiary support of data from unstable sources, incomplete collections, with low temporal and spatial matching quality, or with numerous abnormal records is reduced. This avoids low-quality data participating with the same influence intensity in plot-cycle net carbon sink accounting, improving the stability and credibility of the net carbon sink calculation results.

[0015] Meanwhile, this invention also generates uncertainty gating coefficients based on carbon sink evidence parameters through a dual modulation correction layer, and integrates the reliability coefficients of plot periodic data to form dual modulation parameters. Then, scaling and offset corrections are applied to the plot periodic carbon sink contribution feature vector to generate a reliable carbon sink contribution feature vector. Thus, model uncertainty is no longer merely used for result interpretation, but participates in the feature correction process, enabling carbon sink accounting to simultaneously consider external data quality and model evidence uncertainty, reducing the interference of multi-source data fluctuations, abnormal collections, and insufficient evidence on net carbon sink output. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a blockchain-based agricultural carbon sequestration monitoring method and system proposed in this invention; Figure 2 This is a schematic diagram of the internal structure of a deep evidence regression carbon sequestration accounting model for an agricultural carbon sequestration monitoring method and system based on blockchain proposed in this invention. Figure 3This is a schematic diagram illustrating the blockchain-based hierarchical hash storage and on-chain record formation process of an agricultural carbon sink monitoring method and system proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A blockchain-based method for monitoring agricultural carbon sequestration includes: Collect multi-source monitoring data on agricultural carbon sequestration and preprocess it to form basic data on the periodic carbon sequestration of land plots; Carbon sink source characteristics are extracted from the basic data of land parcel periodic carbon sinks and combined to form a feature vector of land parcel periodic carbon sink contribution. Based on the basic data of carbon sequestration in the land parcel cycle, a multi-source data quality assessment is conducted to generate a reliability coefficient for the land parcel cycle data. The feature vector of carbon sink contribution of the land parcel cycle is input into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation is generated through the branch evidence parameter generation layer. The branch evidence parameters are then corrected in the credible evidence synthesis layer according to the credibility coefficient of the land parcel cycle data to synthesize the carbon sink evidence parameters. Carbon sink evidence parameters, land parcel cycle data credibility coefficient, and land parcel cycle carbon sink contribution feature vector are input into the dual modulation correction layer. An uncertainty gate coefficient is generated based on the carbon sink evidence parameters. The land parcel cycle data credibility coefficient and uncertainty gate coefficient are fused to form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel cycle carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. Input the credible carbon sink contribution feature vector and carbon sink evidence parameters into the net carbon sink output layer to generate the plot's periodic net carbon sink amount. Write the basic data of land parcel cycle carbon sink, the credible carbon sink contribution feature vector, and the evidence information corresponding to the net carbon sink of land parcel cycle into the blockchain to form the on-chain evidence record of land parcel cycle carbon sink.

[0019] In this embodiment, the multi-source monitoring data for agricultural carbon sequestration includes soil monitoring status data, crop growth status data, meteorological environment status data, agricultural operation status data, remote sensing monitoring status data, and data quality record data. Soil monitoring status data includes soil organic carbon status, soil temperature and humidity status, soil nutrient status, and soil disturbance status. Crop growth status data includes vegetation cover status, leaf area status, biomass change status, and crop growth stage status. Meteorological environment status data includes suitable temperature status, precipitation supply status, and light supply status. Agricultural operation status data includes fertilization operation status, irrigation operation status, straw return to the field status, and tillage disturbance status. Remote sensing monitoring status data includes remote sensing growth response status. Data quality record data includes data source records, valid acquisition records, spatial matching records, temporal acquisition records, and anomaly removal records. The preprocessing includes field unification, temporal alignment, spatial matching, missing data completion, and anomaly removal.

[0020] In this embodiment, the formation of the cyclic carbon sink contribution feature vector of the land parcel includes: Read crop growth status data, perform stage correction on biomass change status, and generate crop carbon sequestration characteristics; The generation of crop carbon sequestration features specifically includes: reading crop growth status data, extracting vegetation cover status, leaf area status, biomass change status, and crop growth stage status; normalizing the vegetation cover status, leaf area status, and biomass change status; determining the growth stage identifier corresponding to the current monitoring period based on the crop growth stage status; calling the biomass accumulation coefficient corresponding to the growth stage identifier; scaling the normalized biomass change status using the biomass accumulation coefficient to obtain the stage-corrected biomass change status; and performing vector concatenation and linear mapping on the normalized vegetation cover status, normalized leaf area status, and stage-corrected biomass change status to obtain the crop carbon sequestration mapping vector; and normalizing the crop carbon sequestration mapping vector again to generate crop carbon sequestration features. Based on soil monitoring status data, soil carbon sequestration characteristics are generated. The generation of soil carbon sequestration characteristics specifically includes: reading soil monitoring status data, extracting soil organic carbon status, soil temperature and humidity status, soil nutrient status, and soil disturbance status; calculating the difference between the soil organic carbon status in the current monitoring period and the soil organic carbon status in the previous monitoring period to obtain the periodic change of soil organic carbon; reading soil organic carbon status in multiple consecutive monitoring periods, calculating the average of the periodic change of soil organic carbon in adjacent monitoring periods to obtain the periodic change trend of soil organic carbon; concatenating the soil organic carbon status, soil organic carbon periodic change, and soil organic carbon periodic change trend in the current monitoring period to obtain the soil organic carbon differential coding result; reading soil temperature and soil humidity status in soil temperature and humidity status; reading nitrogen content, phosphorus content, and potassium content status in soil nutrient status; and normalizing each status according to the maximum and minimum values ​​in historical monitoring periods to obtain the scale-unified soil temperature, humidity, and nutrient status; and concatenating the soil organic carbon differential coding result, the scale-unified soil temperature, humidity, and nutrient status, and the soil disturbance status into vectors, and performing linear mapping and normalization to generate soil carbon sequestration characteristics. Read agricultural operation status data, unify the direction of each agricultural operation status's effect on carbon fixation increase and carbon emission reduction, and generate agricultural emission reduction characteristics. The generation of agricultural emission reduction features specifically includes: reading agricultural operation status data, extracting fertilization operation status, irrigation operation status, straw return to field status, and tillage disturbance status, and performing normalization processing; configuring the direction of action label according to the direction of the effect of each agricultural operation status on the increase of carbon fixation and the reduction of carbon emissions; retaining the normalized value of agricultural operation status with positive contribution label; and performing reverse transformation by subtracting the normalized value from 1 for agricultural operation status with negative disturbance label to obtain the agricultural operation status after direction unification; and performing vector concatenation of the agricultural operation status after direction unification, followed by linear mapping and normalization processing to generate agricultural emission reduction features. Read meteorological and environmental status data and remote sensing status data, perform statistical coding of monitoring cycles, and generate environmental regulation characteristics; The generation of environmental regulation characteristics specifically includes: reading meteorological and environmental status data and remote sensing status data, extracting temperature suitability status, precipitation supply status, light supply status and remote sensing growth response status, calculating the periodic mean, periodic maximum, periodic minimum, periodic variation amplitude and periodic effective collection ratio for the above status in the same agricultural plot and the same monitoring period, concatenating the obtained statistics into vectors to form an initial environmental regulation vector, performing linear mapping and normalization on the initial environmental regulation vector to generate environmental regulation characteristics. Environmental regulation characteristics, crop carbon sequestration characteristics, soil carbon sequestration characteristics and agricultural emission reduction characteristics together constitute carbon sink source characteristics. Based on crop carbon sequestration characteristics, soil carbon sequestration characteristics, agricultural emission reduction characteristics, and environmental regulation characteristics, a plot-based periodic carbon sink contribution characteristic vector is formed. The generation of the plot-based periodic carbon sink contribution feature vector specifically includes: concatenating the vectors of crop carbon sequestration characteristics, soil carbon sequestration characteristics, agricultural emission reduction characteristics, and environmental regulation characteristics to form the plot-based periodic carbon sink contribution feature vector.

[0021] In this embodiment, the generation of the reliability coefficient of the land parcel periodic data includes: Read the data quality records from the basic data of the periodic carbon sequestration of the land parcel and generate the data quality status; The generation of data quality status specifically includes: reading data source records, valid collection records, spatial matching records, temporal collection records, and anomaly removal records from the data quality record data, and generating data source status, collection integrity status, spatial matching status, temporal continuity status, and anomaly marking status respectively. First, the data source type, source identifier, and verification result are read from the data source record. Then, the source trust level mapping table is queried based on the data source type, source identifier, and verification result. The obtained source trust level value is used as the data source status. The source trust level mapping table is generated by reading the registered data source list and historical on-chain verification records. The actual data source information in the valid collection records is also read. The effective number of collections and the required number of collections are calculated. The number of effective collections is divided by the number of required collections to obtain the collection completion status. The collection location coverage area and agricultural plot boundary area are read from the spatial matching record. The collection location coverage area is divided by the agricultural plot boundary area to obtain the spatial matching status. The interval between adjacent effective collection times is read from the time collection record. The interval fluctuation ratio is generated based on the ratio of the standard deviation to the average value of the interval between adjacent effective collection times. The interval fluctuation ratio is subtracted from 1 to obtain the time continuity status. The number of abnormal collection values ​​and the total number of collections are read from the anomaly removal record. The number of abnormal collection values ​​is divided by the total number of collections to obtain the anomaly ratio. The anomaly marking status is subtracted from 1. The generation of the source trust level mapping table specifically includes: reading the list of registered data sources, determining the basic trust level corresponding to equipment acquisition sources, platform interface sources, remote sensing data sources, and manual input sources. Specifically, the basic trust level for equipment acquisition sources with passed equipment signature verification is 1; for platform interface sources with passed interface signature verification, the basic trust level is 0.9; for remote sensing data sources with passed image metadata verification, the basic trust level is 0.85; for manual input sources with passed review and verification, the basic trust level is 0.7; and for manually supplemented sources with passed review and verification, the basic trust level is 0. 0.6. The basic trust level of data sources that are manually supplemented but fail the review and verification is 0.4. The basic trust level of data sources that are not registered or have missing verification results is 0.2. Read the historical on-chain verification records, determine the number of times the on-chain evidence is consistent and the total number of on-chain verifications for the same source identifier within the historical monitoring period, divide the number of times the on-chain evidence is consistent by the total number of on-chain verifications to obtain the historical consistency coefficient, multiply the basic trust level by the historical consistency coefficient to obtain the source trust level value, and store the data source type, source identifier, verification result and source trust level value accordingly to form a source trust level mapping table. The data quality status is fused in a consistent direction to generate a reliability coefficient for the periodic data of land parcels. The generation of the land parcel periodic data reliability coefficient specifically includes: taking the data source status, complete collection status, and spatial matching status as positive state variables and retaining their normalized values; taking the fluctuation ratio of adjacent valid collection time intervals in the time collection records as negative state variables, and subtracting the fluctuation ratio of adjacent valid collection time intervals from 1 to obtain the time continuity status; taking the anomaly ratio in the anomaly removal records as negative state variables, and subtracting the anomaly ratio from 1 to obtain the anomaly labeling status; concatenating the data source status, complete collection status, spatial matching status, time continuity status, and anomaly labeling status in a unified direction to form a data quality status vector; and performing normalization processing and linear mapping on the data quality status vector to generate the land parcel periodic data reliability coefficient.

[0022] In this embodiment, the synthesis of carbon sink evidence parameters includes: Input the periodic carbon sink contribution feature vector of the land parcel into the carbon sink contribution branch encoding layer of the deep evidence regression carbon sink accounting model to generate the corresponding carbon sink contribution branch representation. The deep evidence regression carbon sink accounting model includes a carbon sink contribution branch coding layer, a branch evidence parameter generation layer, a credible evidence synthesis layer, a dual modulation correction layer, and a net carbon sink output layer. The training process of the deep evidence regression carbon sink accounting model includes: constructing a training sample set, which includes multi-source monitoring data of sample agricultural carbon sinks and periodic net carbon sink verification data of sample plots; preprocessing the multi-source monitoring data of sample agricultural carbon sinks to form basic data of periodic carbon sinks of sample plots; generating periodic carbon sink contribution feature vectors and periodic data credibility coefficients of sample plots based on the basic data of periodic carbon sinks of sample plots; inputting the periodic carbon sink contribution feature vectors of sample plots into the carbon sink contribution branch encoding layer to obtain the sample carbon sink contribution branch representation; inputting the sample carbon sink contribution branch representations into the branch evidence parameter generation layer to obtain the sample branch evidence parameters; inputting the sample branch evidence parameters and the periodic data credibility coefficients of sample plots into the credible evidence synthesis layer to obtain the sample carbon sink evidence parameters; inputting the sample carbon sink evidence parameters, the periodic data credibility coefficients of sample plots, and the periodic carbon sink contribution feature vectors of sample plots into the dual modulation correction layer to obtain the sample credible carbon sink contribution feature vectors; and finally, inputting the sample carbon sink evidence parameters, the periodic data credibility coefficients of sample plots, and the periodic carbon sink contribution feature vectors of sample plots into the dual modulation correction layer to obtain the sample credible carbon sink contribution feature vectors. The carbon sink contribution feature vector and sample carbon sink evidence parameters are input into the net carbon sink output layer to obtain the predicted net carbon sink of the sample plots for the period. The net carbon sink accounting loss is calculated based on the difference between the predicted net carbon sink of the sample plots for the period and the verified net carbon sink of the sample plots for the period. The evidence constraint loss is calculated based on the correspondence between the sample evidence strength parameter in the sample carbon sink evidence parameters and the difference. The credible evidence consistency loss is calculated based on the correspondence between the credibility coefficient of the sample plot period data and the sample branch evidence strength parameter in the sample branch evidence parameters. A comprehensive training loss is generated based on the net carbon sink accounting loss, the evidence constraint loss, and the credible evidence consistency loss. The trainable parameters in the carbon sink contribution branch encoding layer, the branch evidence parameter generation layer, the credible evidence synthesis layer, the dual modulation correction layer, and the net carbon sink output layer are updated in reverse based on the comprehensive training loss until the comprehensive training loss reaches the convergence condition or the number of training rounds reaches the training limit, resulting in the trained deep evidence regression carbon sink accounting model. The improvements over existing deep evidence regression models include: First, transforming the overall regression input structure into a carbon sink contribution branch encoding layer, generating carbon sink contribution branch representations according to carbon sink source characteristics, thus matching the model input structure with the agricultural carbon sink formation mechanism; Second, transforming the single evidence parameter output structure into a branch evidence parameter generation layer, generating branch evidence parameters corresponding to different carbon sink sources; Third, setting a credible evidence synthesis layer, introducing the land parcel period data credibility coefficient to correct the branch evidence strength parameter in the branch evidence parameters, so that data credibility acts on the degree of evidence support, rather than simply weighting the net carbon sink amount of the land parcel period; Fourth, transforming the uncertainty output in the ordinary deep evidence regression model into an uncertainty gating mechanism, and fusing the uncertainty gating coefficient and the land parcel period data credibility coefficient into dual modulation parameters through a dual modulation correction layer; Fifth, using the dual modulation parameters to perform scaling and offset correction on the land parcel period carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector; Sixth, inputting the credible carbon sink contribution feature vector and carbon sink evidence parameters together into the net carbon sink output layer, so that the net carbon sink amount of the land parcel period is simultaneously constrained by credible feature expression and carbon sink evidence. The generation of carbon sink contribution branch representations specifically includes: reading the plot periodic carbon sink contribution feature vector, branching according to the arrangement position of carbon sink source features in the plot periodic carbon sink contribution feature vector, obtaining crop carbon sequestration input segment, soil carbon sequestration input segment, agricultural emission reduction input segment, and environmental regulation input segment, and performing linear mapping, nonlinear activation, and layer normalization processing on the crop carbon sequestration input segment, soil carbon sequestration input segment, agricultural emission reduction input segment, and environmental regulation input segment respectively to generate crop carbon sequestration branch representation, soil carbon sequestration branch representation, agricultural emission reduction branch representation, and environmental regulation branch representation; The branch evidence parameter generation layer performs evidence regression mapping on the carbon sink contribution branch representation to generate branch evidence parameters; The generation of branch evidence parameters specifically includes: reading the branch representations of crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation; performing linear mapping and nonlinear activation processing on each carbon sink contribution branch representation to obtain the branch regression implicit representation; inputting the branch regression implicit representation into the mean parameter output channel, scale parameter output channel, shape parameter output channel, and evidence strength parameter output channel, respectively; generating the branch net carbon sink mean parameter from the mean parameter output channel, the branch net carbon sink scale parameter from the scale parameter output channel, the branch evidence shape parameter from the shape parameter output channel, and the branch evidence strength parameter from the evidence strength parameter output channel. The mean parameter output channel, scale parameter output channel, and shape parameter output channel... The output channels for the evidence strength parameter have independent mapping parameters, receive the same branch regression implicit representation in parallel, and perform linear mapping processing. The branch net carbon sink scale parameter, branch evidence shape parameter, and branch evidence strength parameter are subjected to positive value constraint processing. Specifically, the above output values ​​are input into a smoothing positive value function and then superimposed with a positive lower limit to make their values ​​positive. The branch net carbon sink mean parameter, the positively constrained branch net carbon sink scale parameter, the positively constrained branch evidence shape parameter, and the positively constrained branch evidence strength parameter are stored as evidence parameter groups for the same carbon sink source, forming branch evidence parameters for the corresponding carbon sink source. The branch evidence parameters are bound to the corresponding parameter groups according to crop carbon sequestration label, soil carbon sequestration label, agricultural emission reduction label, and environmental regulation label. The credible evidence synthesis layer reads the credibility coefficient of the land parcel periodic data, generates the branch credibility correction coefficient based on the land parcel periodic data credibility coefficient, corrects the branch evidence parameters based on the branch credibility correction coefficient, and generates the corrected branch evidence parameters. The generation of the revised branch evidence parameters specifically includes: reading the credibility coefficient of the plot periodic data, and reading the branch evidence parameters of crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation; using the credibility coefficient of the plot periodic data as the branch credibility correction coefficient for each branch evidence parameter; reading the branch net carbon sink mean parameter, branch net carbon sink scale parameter, branch evidence shape parameter, and branch evidence strength parameter from each branch evidence parameter; multiplying each branch evidence strength parameter by the corresponding branch credibility correction coefficient to obtain the revised branch evidence strength parameter; keeping the branch net carbon sink mean parameter, branch net carbon sink scale parameter, and branch evidence shape parameter unchanged; and repackaging the branch net carbon sink mean parameter, branch net carbon sink scale parameter, branch evidence shape parameter, and revised branch evidence strength parameter to generate the revised branch evidence parameters. Evidence is synthesized based on the revised branch evidence parameters to generate carbon sink evidence parameters; The generation of carbon sink evidence parameters specifically includes: reading the branch net carbon sink mean parameter, branch net carbon sink scale parameter, branch evidence shape parameter, and modified branch evidence strength parameter from the modified branch evidence parameters; performing evidence strength normalization and fusion on the branch net carbon sink mean parameter according to the modified branch evidence strength parameter to generate the net carbon sink mean parameter; performing evidence strength normalization and fusion on the branch net carbon sink scale parameter according to the modified branch evidence strength parameter to generate the net carbon sink scale parameter; performing evidence strength normalization and fusion on the branch evidence shape parameter according to the modified branch evidence strength parameter to generate the evidence shape parameter; adding the modified branch evidence strength parameters to generate the evidence strength parameter; and sequentially encapsulating the net carbon sink mean parameter, net carbon sink scale parameter, evidence shape parameter, and evidence strength parameter to generate the carbon sink evidence parameters. The generation of the net carbon sink mean parameter, net carbon sink scale parameter, and evidence shape parameter specifically includes: reading the branch net carbon sink mean parameter, branch net carbon sink scale parameter, branch evidence shape parameter, and corrected branch evidence strength parameter from the corrected branch evidence parameters; adding the corrected branch evidence strength parameters to obtain the total branch evidence strength; dividing each corrected branch evidence strength parameter by the total branch evidence strength to obtain the corresponding evidence fusion coefficient; the credible evidence synthesis layer multiplies and sums the net carbon sink mean parameters of each branch based on the evidence fusion coefficient to generate the net carbon sink mean parameter; multiplies and sums the net carbon sink scale parameters of each branch based on the evidence fusion coefficient to generate the net carbon sink scale parameter; and multiplies and sums the evidence shape parameters of each branch based on the evidence fusion coefficient to generate the evidence shape parameter.

[0023] In this embodiment, the generation of the credible carbon sink contribution feature vector includes: Carbon sink evidence parameters, land parcel period data credibility coefficient, and land parcel period carbon sink contribution feature vector are input into a dual modulation correction layer, which includes an uncertainty analysis unit, a modulation vector generation unit, a dual modulation parameter generation unit, and a feature correction unit. The uncertainty analysis unit reads the carbon sink evidence parameters, performs evidence strength normalization and uncertainty mapping on the carbon sink evidence parameters, and generates uncertainty gating coefficients. The generation of uncertainty gating coefficients specifically includes: reading the net carbon sink mean parameter, net carbon sink scale parameter, evidence shape parameter, and evidence strength parameter from the carbon sink evidence parameters; performing evidence strength normalization on the net carbon sink scale parameter and evidence shape parameter according to the evidence strength parameter to generate data fluctuation uncertainty and evidence support uncertainty; sequentially concatenating the data fluctuation uncertainty and evidence support uncertainty to form an uncertainty state vector; performing linear mapping and normalization on the uncertainty state vector to generate uncertainty mapping quantity; and performing gating compression on the uncertainty mapping quantity to generate uncertainty gating coefficients with values ​​ranging from 0 to 1. The generation of data fluctuation uncertainty specifically includes: comparing the evidence strength parameter with the first evidence threshold and the second evidence threshold, determining the first evidence calibration factor based on the comparison result; when the evidence strength parameter is less than the first evidence threshold, using the low evidence scale calibration factor as the first evidence calibration factor and the low evidence shape calibration factor as the second evidence calibration factor; when the evidence strength parameter is greater than or equal to the first evidence threshold and less than the second evidence threshold, using the medium evidence scale calibration factor as the first evidence calibration factor and the medium evidence shape calibration factor as the second evidence calibration factor; when the evidence strength parameter is greater than or equal to the second evidence threshold, using the high evidence scale calibration factor as the first evidence calibration factor and the high evidence shape calibration factor as the second evidence calibration factor. The low evidence scale calibration factor, medium evidence scale calibration factor, high evidence scale calibration factor, low evidence shape calibration factor, medium evidence shape calibration factor, and high evidence shape calibration factor are arranged in descending order of value; multiplying the net carbon sink scale parameter by the first evidence calibration factor to generate data fluctuation uncertainty; and multiplying the evidence shape parameter by the second evidence calibration factor to generate evidence support uncertainty. The determination of the first evidence threshold and the second evidence threshold specifically includes: reading historical carbon sink evidence parameters generated within the historical monitoring period, extracting historical evidence strength parameters from the historical carbon sink evidence parameters, sorting the historical evidence strength parameters in ascending order of value to obtain a historical evidence strength sequence, reading the historical evidence strength parameter located at the first quantile position in the historical evidence strength sequence as the first evidence threshold, and reading the historical evidence strength parameter located at the second quantile position in the historical evidence strength sequence as the second evidence threshold, wherein the first quantile position is located at one-third of the historical evidence strength sequence, the second quantile position is located at two-thirds of the historical evidence strength sequence, and the first evidence threshold is less than the second evidence threshold; The determination of low-evidence-scale, medium-evidence-scale, and high-evidence-scale calibration factors specifically includes: reading historical net carbon sinks and historical verified net carbon sinks of land parcels, and calculating the absolute difference between the two as the historical accounting deviation. For the low-evidence, medium-evidence, and high-evidence sample groups, the mean historical accounting deviation and the mean historical net carbon sink scale parameter are calculated for each evidence sample group. The mean historical accounting deviation in the low-evidence sample group is divided by the mean historical net carbon sink scale parameter to obtain the low-evidence-scale calibration factor; the mean historical accounting deviation in the medium-evidence sample group is divided by the mean historical net carbon sink scale parameter to obtain the medium-evidence-scale calibration factor; and the mean historical accounting deviation in the high-evidence sample group is divided by the mean historical net carbon sink scale parameter to obtain the high-evidence-scale calibration factor. The determination of low evidence shape calibration factor, medium evidence shape calibration factor, and high evidence shape calibration factor specifically includes: For low evidence sample group, medium evidence sample group, and high evidence sample group, calculate the historical average accounting deviation and the historical average evidence shape parameter in each evidence sample group respectively, and divide the historical average accounting deviation in the low evidence sample group by the historical average evidence shape parameter in the low evidence sample group to obtain the low evidence shape calibration factor; divide the historical average accounting deviation in the medium evidence sample group by the historical average evidence shape parameter in the medium evidence sample group to obtain the medium evidence shape calibration factor; divide the historical average accounting deviation in the high evidence sample group by the historical average evidence shape parameter in the high evidence sample group to obtain the high evidence shape calibration factor. The specific process of gated compression is as follows: Read the uncertainty mapping records within the historical monitoring period, sort the uncertainty mapping records in ascending order of value to form a historical uncertainty mapping sequence, take the uncertainty mapping corresponding to one-third of the positions in the historical uncertainty mapping sequence as the first uncertainty threshold, take the uncertainty mapping corresponding to two-thirds of the positions in the historical uncertainty mapping sequence as the second uncertainty threshold, when the current uncertainty mapping is less than the first uncertainty threshold, assign the uncertainty gate coefficient to 1, when the current uncertainty mapping is greater than or equal to the second uncertainty threshold, assign the uncertainty gate coefficient to 0.2, when the current uncertainty mapping is greater than or equal to the first uncertainty threshold and less than the second uncertainty threshold, subtract the first uncertainty threshold from the current uncertainty mapping to obtain the first difference, subtract the first uncertainty threshold from the second uncertainty threshold to obtain the second difference, divide the first difference by the second difference to obtain the interval position ratio, multiply the interval position ratio by 0.8 to obtain the gate attenuation amount, and subtract the gate attenuation amount from 1 to obtain the uncertainty gate coefficient; The reliability coefficient and uncertainty gating coefficient of the land parcel periodic data are input into the modulation vector generation unit to generate a reliability modulation vector and an uncertainty modulation vector; The generation of the credibility modulation vector and the uncertainty modulation vector specifically includes linear mapping processing of the credibility coefficient and uncertainty gating coefficient of the plot periodic data respectively; In the dual modulation parameter generation unit, a scaling modulation vector and an offset modulation vector are generated based on the confidence modulation vector and the uncertainty modulation vector. The generation of scaling modulation vectors and offset modulation vectors specifically includes: reading the confidence modulation vector and the uncertainty modulation vector, aligning them according to the same dimension; when the dimensions of the confidence modulation vector and the uncertainty modulation vector are inconsistent, repeatedly expanding the lower-dimensional vector according to the dimension of the plot period carbon sink contribution feature vector to obtain a confidence modulation vector and an uncertainty modulation vector with consistent dimensions; calculating the product of the corresponding position components in the confidence modulation vector and the uncertainty modulation vector to obtain the collaborative modulation component; calculating the absolute value of the difference between the corresponding position components in the confidence modulation vector and the uncertainty modulation vector to obtain the conflict modulation component; sequentially concatenating the confidence modulation vector, the uncertainty modulation vector, the collaborative modulation component, and the conflict modulation component to form a modulation fusion representation; averaging the confidence modulation component and the uncertainty modulation component to obtain the basic retained component; multiplying the basic retained component with the collaborative modulation component to obtain the collaborative enhancement component; adding the basic retained component with the collaborative enhancement component to obtain the scaling candidate component; pruning the scaling candidate component to form the scaling modulation component; and assembling the scaling modulation components into a scaling modulation vector according to the arrangement order of each feature component in the plot period carbon sink contribution feature vector. The generation of the offset modulation vector specifically includes: reading the conflict modulation component and the plot periodic carbon sink contribution feature vector in the modulation fusion representation; calculating the mean of each feature component in the plot periodic carbon sink contribution feature vector within the current monitoring period to obtain the feature period mean; calculating the difference between each feature component and the feature period mean to obtain the feature deviation component; multiplying the conflict modulation component and the feature deviation component to obtain the offset candidate component; limiting the amplitude of the offset candidate component to obtain the offset modulation component; and assembling the offset modulation components into an offset modulation vector according to the arrangement order of each feature component in the plot periodic carbon sink contribution feature vector. The feature correction unit reads the plot periodic carbon sink contribution feature vector, scaling modulation vector, and offset modulation vector to generate a reliable carbon sink contribution feature vector. The generation of the credible carbon sink contribution feature vector specifically includes: reading the plot periodic carbon sink contribution feature vector, scaling modulation vector, and offset modulation vector, aligning their positions, multiplying each feature component in the plot periodic carbon sink contribution feature vector with the corresponding scaling modulation component in the scaling modulation vector to obtain the scaling-corrected feature component, adding the scaling-corrected feature component with the corresponding offset modulation component in the offset modulation vector to obtain the double-modulation-corrected feature component, combining all the double-modulation-corrected feature components in their original order to form the double-modulation-corrected carbon sink contribution feature vector, and then normalizing the double-modulation-corrected carbon sink contribution feature vector to generate the credible carbon sink contribution feature vector.

[0024] In this embodiment, the generation of the periodic net carbon sink of the land parcel includes: The credible carbon sink contribution feature vector and carbon sink evidence parameters are input into the net carbon sink output layer, which includes a feature evidence fusion unit, a carbon sink mapping unit, an evidence mean correction unit, and a net carbon sink quantity output unit. The feature evidence fusion unit reads the credible carbon sink contribution feature vector and carbon sink evidence parameters, and generates a feature evidence fusion representation. The generation of the feature evidence fusion representation specifically includes: reading the net carbon sink mean parameter, net carbon sink scale parameter, evidence shape parameter and evidence strength parameter from the credible carbon sink contribution feature vector and carbon sink evidence parameters; normalizing the net carbon sink scale parameter, evidence shape parameter and evidence strength parameter and then concatenating them sequentially with the credible carbon sink contribution feature vector to generate the feature evidence fusion representation. The feature evidence fusion representation is input into the carbon sink mapping unit to generate the initial net carbon sink mapping amount; The generation of the initial net carbon sink mapping specifically includes: reading the feature evidence fusion representation, extracting credible carbon sink contribution feature components according to their arrangement in the feature evidence fusion representation, and dividing them into credible components for crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation; calculating the mean of each credible component; reading the normalized result of the evidence strength parameter in the feature evidence fusion representation; using the normalized result of the evidence strength parameter as the evidence constraint ratio; and multiplying and summing the mean values ​​of each credible component to generate the basic carbon sink mapping. The product coefficients corresponding to the mean values ​​of each component are jointly determined by the evidence constraint ratio and the length proportion of the corresponding carbon sink source characteristics in the periodic carbon sink contribution feature vector of the land parcel. The normalized result of the net carbon sink scale parameter in the feature evidence fusion representation is read, and the basic carbon sink mapping amount is multiplied by the normalized result of the net carbon sink scale parameter to obtain the scale response amount. The normalized result of the evidence shape parameter in the feature evidence fusion representation is read, and the basic carbon sink mapping amount is multiplied by the normalized result of the evidence shape parameter to obtain the shape response amount. The basic carbon sink mapping amount, the scale response amount, and the shape response amount are added together to generate the initial net carbon sink mapping amount. The generation of the evidence constraint ratio specifically includes: reading the evidence strength parameter in the carbon sink evidence parameters and reading the historical evidence strength parameter in the historical carbon sink evidence parameters; sorting the historical evidence strength parameters in ascending order of value to form a historical evidence strength sequence; reading the minimum evidence strength value and the maximum evidence strength value in the historical evidence strength sequence; subtracting the minimum evidence strength value from the evidence strength parameter to obtain the evidence strength difference; subtracting the minimum evidence strength value from the maximum evidence strength value to obtain the evidence strength range value; dividing the evidence strength difference by the evidence strength range value to obtain the evidence strength normalization result; when the evidence strength normalization result is less than 0, correcting the evidence strength normalization result to 0; when the evidence strength normalization result is greater than 1, correcting the evidence strength normalization result to 1; and using the corrected evidence strength normalization result as the evidence constraint ratio. The generation of the product coefficients corresponding to the means of each component specifically includes: reading the credible components of crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation; counting the number of each credible component; summing the number of components of the four types of credible components to obtain the total number of credible components; dividing the number of credible components of crop carbon sequestration by the total number of credible components to obtain the proportion of crop carbon sequestration length; dividing the number of credible components of soil carbon sequestration by the total number of credible components to obtain the proportion of soil carbon sequestration length; dividing the number of credible components of agricultural emission reduction by the total number of credible components to obtain the proportion of agricultural emission reduction length; and dividing the number of credible components of environmental regulation by the total number of credible components to obtain the proportion of environmental regulation length. The evidence constraint ratio is multiplied by the proportions of crop carbon sequestration length, soil carbon sequestration length, agricultural emission reduction length, and environmental regulation length, respectively, to obtain the initial product coefficients for crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation. The four initial product coefficients are added together to obtain the initial coefficient sum. The initial product coefficient of crop carbon sequestration is divided by the initial coefficient sum to obtain the crop carbon sequestration product coefficient. The initial product coefficient of soil carbon sequestration is divided by the initial coefficient sum to obtain the soil carbon sequestration product coefficient. The initial product coefficient of agricultural emission reduction is divided by the initial coefficient sum to obtain the agricultural emission reduction product coefficient. The initial product coefficient of environmental regulation is divided by the initial coefficient sum to obtain the environmental regulation product coefficient. The evidence mean correction unit calculates the evidence-corrected net carbon sink mapping based on the initial net carbon sink mapping and the characteristic evidence fusion representation. The calculation of the evidence-corrected net carbon sink mapping amount specifically includes: reading the normalized results of the initial net carbon sink mapping amount, the net carbon sink mean parameter and the evidence strength parameter in the feature evidence fusion representation; calculating the difference between the initial net carbon sink mapping amount and the net carbon sink mean parameter to obtain the net carbon sink mapping deviation; multiplying the net carbon sink mapping deviation amount by the normalized result of the evidence strength parameter to obtain the evidence correction amount; and subtracting the evidence correction amount from the initial net carbon sink mapping amount to obtain the evidence-corrected net carbon sink mapping amount. Specifically, when the normalized result of the evidence strength parameter is 1, the evidence-corrected net carbon sink mapping amount is equal to the net carbon sink mean parameter; when the normalized result of the evidence strength parameter is 0, the evidence-corrected net carbon sink mapping amount is equal to the initial net carbon sink mapping amount; and when the normalized result of the evidence strength parameter is between 0 and 1, the evidence-corrected net carbon sink mapping amount is between the initial net carbon sink mapping amount and the net carbon sink mean parameter. The evidence-corrected net carbon sink mapping is input into the net carbon sink output unit, and scale restoration is performed based on the monitoring cycle length and agricultural plot area corresponding to the plot cycle carbon sink basic data to generate the plot cycle net carbon sink. The generation of net carbon sinks for a given period specifically includes: first, reading the start and end times of the monitoring period from the basic data on the periodic carbon sinks of the land parcel; subtracting the start time from the end time to obtain the length of the monitoring period; converting the length of the monitoring period into the number of days; multiplying the evidence-corrected net carbon sink mapping amount by the number of days to obtain the periodic-scale net carbon sink mapping amount; obtaining the area of ​​the agricultural land parcel; multiplying the periodic-scale net carbon sink mapping amount by the area of ​​the agricultural land parcel to obtain the land parcel-scale net carbon sink mapping amount; if the unit of carbon sink recorded in the basic data on the periodic carbon sinks of the land parcels is inconsistent with the unit required for output, the net carbon sink output unit calls the unit conversion factor according to the carbon sink unit identifier, and multiplies the land parcel-scale net carbon sink mapping amount by the unit conversion factor to obtain the periodic net carbon sink of the land parcel under a unified unit.

[0025] In this embodiment, the formation of evidence records on the land parcel's periodic carbon sink chain includes: Read the basic data of carbon sequestration in the periodic land parcel, the credible carbon sequestration contribution feature vector, and the net carbon sequestration in the periodic land parcel; The basic data of the periodic carbon sequestration of land parcels are serialized and hashed to obtain the basic data hash. The generation of basic data hash specifically includes: reading the field names, field values, and field sources from the basic data of the land parcel's periodic carbon sequestration; sorting each field according to the dictionary order of the field names; concatenating the sorted field names, field values, and field sources in sequence to form a basic data sequence; and performing a hash operation on the basic data sequence to obtain the basic data hash. The trusted carbon sink contribution feature vector is subjected to vector serialization and hashing to obtain the trusted feature hash. The generation of trusted feature hash specifically includes: reading each feature component and its sequence number in the trusted carbon sink contribution feature vector, converting each feature component into a fixed-length numerical string in ascending order of sequence number, concatenating the fixed-length numerical strings to form a trusted feature sequence, and performing a hash operation on the trusted feature sequence to obtain the trusted feature hash; Agricultural plot identifiers and monitoring cycle identifiers are read from the basic data of plot period carbon sequestration. The results of plot period net carbon sequestration, agricultural plot identifiers and monitoring cycle identifiers are serialized and hashed to obtain the result hash. The generation of the result hash specifically includes: concatenating the agricultural plot identifier, monitoring period identifier, and net carbon sink of the plot period in the order of agricultural plot identifier first, monitoring period identifier in the middle, and net carbon sink of the plot period last to form a net carbon sink result sequence; performing a hash operation on the net carbon sink result sequence to obtain the result hash; The basic data hash, trusted feature hash, and result hash are combined to form the carbon sink process hash input sequence, and the carbon sink process hash input sequence is hashed to obtain the carbon sink process root hash. The generation of the carbon sink process root hash specifically includes: concatenating the basic data hash, trusted feature hash, and result hash in the order of basic data hash first, trusted feature hash in the middle, and result hash last to form the carbon sink process hash input sequence; performing a hash operation on the carbon sink process hash input sequence to obtain the carbon sink process root hash; The specific process of hashing is as follows: First, the input sequence to be hashed is uniformly converted into a UTF-8 encoded string. For numerical fields, the system retains a fixed number of decimal places and converts them into fixed-length numerical strings. For time fields, the system converts them into time strings in a fixed order of year, month, day, hour, minute, and second. For identifier fields, the system retains their standard encoding form. After encoding, the system adds a fixed separator between adjacent fields and adds the agricultural plot identifier, monitoring cycle identifier, and data version number to the standard hash input string, so that the hash result is bound to the plot, cycle, and data version simultaneously. At the same time, a cryptographic hash operation is performed on the standard hash input string to obtain a 256-bit hash digest, and the hash digest is converted into a hexadecimal string as the final hash value. The above basic data sequence generates a basic data hash after this processing, the trusted feature sequence generates a trusted feature hash after this processing, the net carbon sink result sequence generates a result hash after this processing, and the carbon sink process hash input sequence, which combines the basic data hash, trusted feature hash, and result hash, generates a carbon sink process root hash after this processing. The corresponding basic data generation time is read from the basic data of the land plot cycle carbon sink, the corresponding credible feature generation time is read from the credible carbon sink contribution feature vector, and the corresponding net carbon sink generation time is read from the net carbon sink amount of the land plot cycle. The agricultural land plot identifier, monitoring cycle identifier, basic data hash, credible feature hash, result hash, carbon sink process root hash, data version number and the above generation time are written into the blockchain to form the on-chain evidence record of the land plot cycle carbon sink.

[0026] A blockchain-based agricultural carbon sequestration monitoring system includes: The data processing module is used to collect multi-source monitoring data of agricultural carbon sequestration and preprocess it to form basic data of plot periodic carbon sequestration. The feature construction module is used to extract carbon sink source features based on the basic data of land parcel periodic carbon sinks and combine them to form a feature vector of land parcel periodic carbon sink contribution. The quality assessment module is used to conduct multi-source data quality assessment based on the basic data of land parcel periodic carbon sequestration and generate the reliability coefficient of land parcel periodic data. The evidence parameter generation module is used to input the feature vector of the carbon sink contribution of the land parcel period into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation generates branch evidence parameters through the branch evidence parameter generation layer. The branch evidence parameters are corrected according to the credibility coefficient of the land parcel period data in the credible evidence synthesis layer to synthesize the carbon sink evidence parameters. The dual modulation correction module is used to input carbon sink evidence parameters, land parcel period data credibility coefficient and land parcel period carbon sink contribution feature vector into the dual modulation correction layer and form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel period carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. The output module is used to generate the periodic net carbon sink of the land parcel based on the credible carbon sink contribution feature vector and carbon sink evidence parameters. The on-chain evidence storage module is used to generate on-chain evidence storage records for the periodic carbon sink of land parcels.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a high-standard farmland agricultural carbon sequestration monitoring scenario in a certain county. This area is primarily planted with grain crops, with clear plot boundaries and readily available data sources including soil monitoring, remote sensing interpretation, meteorological observations, and agricultural operation records. Previous accounting methods mainly relied on manual recording, periodic soil testing, and empirical factor estimation. When remote sensing images are affected by cloud cover, soil sensors experience short-term data gaps, or agricultural records are supplemented, the net carbon sequestration of the same plot within adjacent monitoring periods is prone to fluctuations caused by non-production factors. Furthermore, verification personnel find it difficult to trace the underlying data and processing procedures that led to the final result.

[0028] In this scenario, multi-source monitoring data on agricultural carbon sequestration is first collected. Soil monitoring status data, crop growth status data, meteorological environment status data, agricultural operation status data, remote sensing monitoring status data, and data quality records are then processed through field unification, time alignment, spatial matching, missing data completion, and anomaly removal to form basic data for plot-based periodic carbon sequestration. Subsequently, crop carbon sequestration characteristics, soil carbon sequestration characteristics, agricultural emission reduction characteristics, and environmental regulation characteristics are extracted from this basic data and combined to form a plot-based periodic carbon sequestration contribution feature vector. Simultaneously, data quality records are read to generate a plot-based periodic data reliability coefficient, enabling data from different sources, with varying degrees of completeness, and with different matching quality to enter subsequent calculations with differentiated reliability levels.

[0029] The deep evidence regression carbon sink accounting model branches and encodes the plot-cycle carbon sink contribution feature vector to form branch representations of carbon sink contributions for crop carbon sequestration, soil carbon sequestration, agricultural emission reduction, and environmental regulation. Branch evidence parameters are generated through a branch evidence parameter generation layer. A credible evidence synthesis layer corrects the branch evidence parameters based on the credibility coefficient of the plot-cycle data, ensuring that low-quality data is no longer included in the accounting with equal evidence strength. A dual modulation correction layer further generates uncertainty gating coefficients based on the carbon sink evidence parameters and merges them with the plot-cycle data credibility coefficients to form dual modulation parameters. These parameters are then used to perform scaling and offset corrections on the plot-cycle carbon sink contribution feature vector to obtain a credible carbon sink contribution feature vector. Finally, the net carbon sink output layer combines the credible carbon sink contribution feature vector and the carbon sink evidence parameters to generate the plot-cycle net carbon sink.

[0030] After the calculation is completed, the basic data of the land parcel cycle carbon sink, the trusted carbon sink contribution feature vector, and the net carbon sink amount of the land parcel cycle are serialized and hashed respectively to form the basic data hash, trusted feature hash, and result hash, which are then combined to generate the carbon sink process root hash. The above hashes, agricultural land parcel identifiers, monitoring cycle identifiers, data version numbers, and generation times are written into the blockchain to form an on-chain evidence record of the land parcel cycle carbon sink.

[0031] To illustrate the implementation effect, agricultural carbon sequestration records of contiguous high-standard farmland in a certain county were compiled within the same monitoring period. Traditional methods estimate agricultural carbon sequestration based on manual agricultural records, periodic soil monitoring data, remote sensing vegetation indices, and conventional carbon emission factors. The method of this invention employs multi-source data quality assessment, a deep evidence regression carbon sequestration calculation model, credible evidence synthesis, dual modulation correction, and a hierarchical hash chain evidence storage mechanism to generate the periodic net carbon sequestration of each plot and its on-chain evidence records. The comparison includes net carbon sequestration calculation errors, the impact of outliers, verification consistency, evidence completeness, and result traceability efficiency.

[0032] Table 1. Comparison of the overall performance of agricultural carbon sequestration monitoring, accounting, and evidence storage.

[0033] As can be seen from Table 1, the ordinary machine learning regression method reduces the error to 0.47tCO2e / ha·cycle, but it still mainly relies on overall feature regression and does not explicitly handle different data quality states and model evidence uncertainty. Therefore, the error increase is still 31.4% after the influence of abnormal data.

[0034] The on-chain verification pass rate of conventional blockchain result storage methods has increased to 91.3%, indicating that storing the final result on-chain can improve the tamper-proof capability of the result record. However, its average verification time is still 10.8 minutes per block cycle, and the number of data traceability levels is only 1. This indicates that it mainly proves whether the final result is consistent, but cannot further locate whether the difference originates from the basic data, trusted features, or the result generation process. In contrast, the method of this invention achieves an on-chain verification pass rate of 98.1%, reduces the average verification time to 4.3 minutes per block cycle, and achieves 3 traceability levels, enabling layered verification of basic data hash, trusted feature hash, and result hash.

[0035] The method of this invention achieves a mean absolute error of 0.29 tCO2e / ha·cycle for net carbon sink, significantly lower than other algorithms. The coefficient of variation for accounting stability is reduced to 8.4%, and the error increase after the influence of abnormal data is reduced to 15.7%. This performance improvement is due to the following: This invention first generates a reliability coefficient for plot periodic data through multi-source data quality assessment, then corrects the branch evidence parameters in the reliable evidence synthesis layer, reducing the evidence support level of low-quality data. Simultaneously, the dual modulation correction layer applies data reliability and uncertainty gating to the plot periodic carbon sink contribution feature vector, forming a reliable carbon sink contribution feature vector, thus preventing a single abnormal collection or supplementary record from directly dominating the net carbon sink output.

[0036] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A blockchain-based agricultural carbon sink monitoring method, characterized in that, include: Collect multi-source monitoring data on agricultural carbon sequestration and preprocess it to form basic data on the periodic carbon sequestration of land plots; Carbon sink source characteristics are extracted from the basic data of land parcel periodic carbon sinks and combined to form a feature vector of land parcel periodic carbon sink contribution. Based on the basic data of carbon sequestration in the land parcel cycle, a multi-source data quality assessment is conducted to generate a reliability coefficient for the land parcel cycle data. The feature vector of carbon sink contribution of the land parcel cycle is input into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation is generated through the branch evidence parameter generation layer. The branch evidence parameters are then corrected in the credible evidence synthesis layer according to the credibility coefficient of the land parcel cycle data to synthesize the carbon sink evidence parameters. Carbon sink evidence parameters, land parcel cycle data credibility coefficient, and land parcel cycle carbon sink contribution feature vector are input into the dual modulation correction layer. An uncertainty gate coefficient is generated based on the carbon sink evidence parameters. The land parcel cycle data credibility coefficient and uncertainty gate coefficient are fused to form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel cycle carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. Input the credible carbon sink contribution feature vector and carbon sink evidence parameters into the net carbon sink output layer to generate the plot's periodic net carbon sink amount. Write the basic data of land parcel cycle carbon sink, the credible carbon sink contribution feature vector, and the evidence information corresponding to the net carbon sink of land parcel cycle into the blockchain to form the on-chain evidence record of land parcel cycle carbon sink. 2.The blockchain-based agricultural carbon sink monitoring method of claim 1, wherein, The agricultural carbon sink multi-source monitoring data includes soil monitoring status data, crop growth status data, meteorological environment status data, agricultural operation status data, remote sensing monitoring status data, and data quality record data. The preprocessing includes field unification, time alignment, spatial matching, missing data completion, and anomaly removal. 3.The blockchain-based agricultural carbon sink monitoring method of claim 1, wherein, The formation of the cyclic carbon sink contribution feature vector of the land parcel includes: Read crop growth status data, perform stage correction on biomass change status, and generate crop carbon sequestration characteristics; Based on soil monitoring status data, soil carbon sequestration characteristics are generated. Read agricultural operation status data, unify the direction of each agricultural operation status's effect on carbon fixation increase and carbon emission reduction, and generate agricultural emission reduction characteristics. Read meteorological and environmental status data and remote sensing status data, perform statistical coding of monitoring cycles, and generate environmental regulation characteristics; Based on crop carbon sequestration characteristics, soil carbon sequestration characteristics, agricultural emission reduction characteristics, and environmental regulation characteristics, a plot-based periodic carbon sink contribution characteristic vector is formed. 4.The blockchain-based agricultural carbon sink monitoring method of claim 1, wherein, The generation of the reliability coefficient of the land parcel periodic data includes: Read the data quality records from the basic data of the periodic carbon sequestration of the land parcel and generate the data quality status; The credibility of the data quality status is fused to generate the credibility coefficient of the land parcel periodic data.

5. The method for monitoring agricultural carbon sinks based on blockchain according to claim 1, characterized in that, The synthesis of the carbon sink evidence parameters includes: Input the periodic carbon sink contribution feature vector of the land parcel into the carbon sink contribution branch encoding layer of the deep evidence regression carbon sink accounting model to generate the corresponding carbon sink contribution branch representation. The deep evidence regression carbon sink accounting model includes a carbon sink contribution branch coding layer, a branch evidence parameter generation layer, a credible evidence synthesis layer, a dual modulation correction layer, and a net carbon sink output layer. The branch evidence parameter generation layer performs evidence regression mapping on the carbon sink contribution branch representation to generate branch evidence parameters; The credible evidence synthesis layer reads the credibility coefficient of the land parcel periodic data, generates the branch credibility correction coefficient based on the land parcel periodic data credibility coefficient, corrects the branch evidence parameters based on the branch credibility correction coefficient, and generates the corrected branch evidence parameters. Evidence is synthesized based on the revised branch evidence parameters to generate carbon sink evidence parameters.

6. The method for monitoring agricultural carbon sinks based on blockchain according to claim 1, characterized in that, The generation of the credible carbon sink contribution feature vector includes: Carbon sink evidence parameters, land parcel period data credibility coefficient, and land parcel period carbon sink contribution feature vector are input into a dual modulation correction layer, which includes an uncertainty analysis unit, a modulation vector generation unit, a dual modulation parameter generation unit, and a feature correction unit. The uncertainty analysis unit reads the carbon sink evidence parameters, performs evidence strength normalization and uncertainty mapping on the carbon sink evidence parameters, and generates uncertainty gating coefficients. The reliability coefficient and uncertainty gating coefficient of the land parcel periodic data are input into the modulation vector generation unit to generate a reliability modulation vector and an uncertainty modulation vector; In the dual modulation parameter generation unit, a scaling modulation vector and an offset modulation vector are generated based on the confidence modulation vector and the uncertainty modulation vector. The feature correction unit reads the plot periodic carbon sink contribution feature vector, scaling modulation vector, and offset modulation vector to generate a reliable carbon sink contribution feature vector.

7. The method for monitoring agricultural carbon sinks based on blockchain according to claim 1, characterized in that, The generation of the periodic net carbon sink of the land parcel includes: The credible carbon sink contribution feature vector and carbon sink evidence parameters are input into the net carbon sink output layer, which includes a feature evidence fusion unit, a carbon sink mapping unit, an evidence mean correction unit, and a net carbon sink quantity output unit. The feature evidence fusion unit reads the credible carbon sink contribution feature vector and carbon sink evidence parameters, and generates a feature evidence fusion representation. The feature evidence fusion representation is input into the carbon sink mapping unit to generate the initial net carbon sink mapping amount; The evidence mean correction unit calculates the evidence-corrected net carbon sink mapping based on the initial net carbon sink mapping and the characteristic evidence fusion representation. The evidence-corrected net carbon sink mapping is input into the net carbon sink output unit, and scale restoration is performed based on the monitoring cycle length and agricultural plot area corresponding to the plot cycle carbon sink basic data to generate the plot cycle net carbon sink.

8. The method for monitoring agricultural carbon sinks based on blockchain according to claim 1, characterized in that, The formation of the evidence records on the periodic carbon sink chain of the land parcel includes: Read the basic data of carbon sequestration in the periodic land parcel, the credible carbon sequestration contribution feature vector, and the net carbon sequestration in the periodic land parcel; The basic data of the periodic carbon sequestration of land parcels are serialized and hashed to obtain the basic data hash. The trusted carbon sink contribution feature vector is subjected to vector serialization and hashing to obtain the trusted feature hash. Agricultural plot identifiers and monitoring cycle identifiers are read from the basic data of plot period carbon sequestration. The results of plot period net carbon sequestration, agricultural plot identifiers and monitoring cycle identifiers are serialized and hashed to obtain the result hash. The basic data hash, trusted feature hash, and result hash are combined to form the carbon sink process hash input sequence, and the carbon sink process hash input sequence is hashed to obtain the carbon sink process root hash. The corresponding basic data generation time is read from the basic data of the land plot cycle carbon sink, the corresponding credible feature generation time is read from the credible carbon sink contribution feature vector, and the corresponding net carbon sink generation time is read from the net carbon sink amount of the land plot cycle. The agricultural land plot identifier, monitoring cycle identifier, basic data hash, credible feature hash, result hash, carbon sink process root hash, data version number and the above generation time are written into the blockchain to form the on-chain evidence record of the land plot cycle carbon sink.

9. A blockchain-based agricultural carbon sequestration monitoring system, implementing the blockchain-based agricultural carbon sequestration monitoring method according to any one of claims 1 to 8, characterized in that, include: The data processing module is used to collect multi-source monitoring data of agricultural carbon sequestration and preprocess it to form basic data of plot periodic carbon sequestration. The feature construction module is used to extract carbon sink source features based on the basic data of land parcel periodic carbon sinks and combine them to form a feature vector of land parcel periodic carbon sink contribution. The quality assessment module is used to conduct multi-source data quality assessment based on the basic data of land parcel periodic carbon sequestration and generate the reliability coefficient of land parcel periodic data. The evidence parameter generation module is used to input the feature vector of the carbon sink contribution of the land parcel period into the carbon sink contribution branch coding layer of the deep evidence regression carbon sink accounting model. The generated carbon sink contribution branch representation generates branch evidence parameters through the branch evidence parameter generation layer. The branch evidence parameters are corrected according to the credibility coefficient of the land parcel period data in the credible evidence synthesis layer to synthesize the carbon sink evidence parameters. The dual modulation correction module is used to input carbon sink evidence parameters, land parcel period data credibility coefficient and land parcel period carbon sink contribution feature vector into the dual modulation correction layer and form dual modulation parameters. The dual modulation parameters are used to perform scaling correction and offset correction on the land parcel period carbon sink contribution feature vector to generate a credible carbon sink contribution feature vector. The output module is used to generate the periodic net carbon sink of the land parcel based on the credible carbon sink contribution feature vector and carbon sink evidence parameters. The on-chain evidence storage module is used to generate on-chain evidence storage records for the periodic carbon sink of land parcels.

Images