Grassland carbon sink calculation model, grassland carbon credit calculation method and storage medium

Abstract

The invention relates to the technical field of carbon sink calculation, in particular to a grassland carbon sink calculation model, a grassland carbon credit calculation method and a storage medium. And calculating overground biological carbon reserve, underground root carbon reserve and soil organic carbon content by using a multi-task learning model based on a Transform architecture and a bidirectional long-short-term memory network. According to the method, multi-dimensional calculation of the carbon sink amount is achieved, the model generalization ability and the calculation precision are ensured, meanwhile, the manual dependency degree is reduced, and the calculation efficiency is further improved.

Description

Technical Field

[0001] The present invention relates to the field of carbon sink calculation technology, in particular to a grassland carbon sink calculation model, a grassland carbon credit calculation method, and a storage medium. Background Art

[0002] Currently, grassland carbon credit calculations primarily rely on manual plot surveys or single remote sensing data. However, manual field surveys require significant manpower, take weeks or even months to complete, and lack coverage over large areas of grassland. Carbon sink calculations based on remote sensing data ignore the dynamics of underground root biomass and soil organic carbon, leading to high error rates. Furthermore, remote sensing data-based carbon sink calculations rely heavily on spectral data and fail to consider factors that significantly influence carbon sinks, such as meteorological data and grazing activity, severely impacting their accuracy. Summary of the Invention

[0003] The purpose of this application is to provide a grassland carbon sink calculation model, a grassland carbon credit calculation method, and a storage medium. Through a bidirectional long short-term memory network and a multi-task learning model based on the Transformer architecture, combined with an aboveground biomass carbon storage module, an underground root carbon storage module, and a soil organic carbon content module; a multi-dimensional grassland carbon sink calculation model is established to solve the high dependence on manual labor and the inability to adapt to large-scale grassland carbon monitoring needs; at the same time, each module covers all major aspects that affect grassland carbon storage changes, ensuring coverage and improving the accuracy of carbon sink calculation.

[0004] The present application provides a grassland carbon sink calculation model, including a bidirectional long short-term memory network and a multi-task learning model based on the Transformer architecture. The multi-task learning model includes: an aboveground biomass carbon storage module, an underground root carbon storage module, and a soil organic carbon content module; the training data of the model includes grassland ecological knowledge and historical carbon flux data.

[0005] Furthermore, an attention mechanism module for optimizing calculation weights is also included.

[0006] On the other hand, the present application also provides a grassland carbon credit calculation method based on a grassland carbon sink calculation model, which uses any of the grassland carbon sink calculation models described above to calculate the carbon sink amount of the target area.

[0007] Furthermore, the grassland carbon credit calculation method based on the grassland carbon sink calculation model of the present application includes the following steps:

[0008] Use the multimodal data acquisition module to collect basic data of the target area;

[0009] Performing spatiotemporal alignment and / or noise filtering processing on the basic data using a data fusion module;

[0010] The basic data that has undergone spatiotemporal alignment and / or noise filtering processing is input into the grassland carbon sink calculation model to calculate the carbon sink amount.

[0011] Furthermore, the collecting of basic data of the target area using the multimodal data collection module includes:

[0012] Acquiring remote sensing information of a target area using a hyperspectral satellite remote sensing unit, wherein the remote sensing information includes above-ground plant mass and vegetation coverage;

[0013] Use the multispectral scanning unit of the UAV to obtain the surface degradation information of the target area;

[0014] An underground carbon monitoring network is used to obtain underground information, including soil temperature, humidity, and root biomass.

[0015] Furthermore, the data fusion module is used to perform spatiotemporal alignment and / or noise filtering on the basic data, including:

[0016] unifying the temporal and spatial references of the remote sensing information, the surface degradation information, and the subsurface information using a spatiotemporal alignment engine;

[0017] The mapping relationship between vegetation index and soil carbon content was established using multi-source data association units based on the basic data after unified benchmarking;

[0018] Use variational autoencoders to repair abnormal and / or missing data.

[0019] Furthermore, it also includes:

[0020] Use the dynamic carbon sink correction module to adjust carbon sinks based on precipitation, temperature, and grazing data;

[0021] Use online incremental learning module to predict future carbon sequestration trends.

[0022] Furthermore, it also includes accessing livestock GPS positioning collar data to analyze grazing paths and densities; combined with the grassland rotational grazing management system, it evaluates livestock feeding intensity and vegetation recovery.

[0023] Furthermore, the carbon sink amount is output and displayed using the result output and evidence storage module.

[0024] In another aspect of the present application, a computer-readable storage medium is provided, which stores a computer program, and when the program is executed by a processor, the method as described in any one of the above items is implemented.

[0025] The grassland carbon sink calculation model, grassland carbon credit calculation method, and storage medium provided in this application adopt a bidirectional long short-term memory network and a multi-task learning model based on the Transformer architecture, wherein the multi-task learning model includes: an aboveground biomass carbon storage module, an underground root carbon storage module, and a soil organic carbon content module; the training data of the model includes the design of grassland ecology knowledge and historical carbon flux data. Through a multi-task learning model based on the Transformer architecture and a bidirectional long short-term memory network, combined with the aboveground biomass carbon storage module, the underground root carbon storage module, and the soil organic carbon content module, a grassland carbon sink calculation model based on multiple dimensions is established to solve the problem of high artificial dependence and inability to adapt to large-scale grassland carbon monitoring needs; at the same time, each module covers all major aspects that affect grassland carbon storage changes, ensuring coverage and improving the accuracy of carbon sink calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0027] Figure 1 This is a flowchart of an embodiment of a grassland carbon credit calculation method based on a grassland carbon sink calculation model provided in this application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0029] Carbon sinks refer to changes in carbon produced by activities, processes, or mechanisms that absorb atmospheric carbon dioxide and store it in organisms and soils through natural processes such as plant photosynthesis. For example, forest carbon sinks can be represented by the sum of changes in forest carbon storage over a given period. Carbon credits are greenhouse gas emission reduction or sequestration enhancement projects developed in accordance with the rules, procedures, and methodologies established by relevant standards or mechanisms, and are issued after verification, verification, and certification by an independent third-party organization. Each carbon credit represents the reduction or removal of one ton of carbon dioxide equivalent from the air.

[0030] This application aims to address the following issues with existing grassland carbon sink calculation methods: 1) low efficiency, as manual sample surveys require a large amount of manpower and a long cycle (weeks to months), making it difficult to cover wide areas of grassland; 2) insufficient precision, as a single model ignores the dynamic changes in belowground biomass (root carbon) and soil organic carbon, with an error rate generally exceeding 25%; 3) single data dimension, relying on spectral data and failing to integrate key factors such as meteorological data (precipitation, temperature) and grazing activity records; 4) lack of dynamics, making it impossible to respond in real time to the carbon sink impacts of grassland degradation, overgrazing, or extreme climate events (such as drought); and 5) poor interpretability: black-box models lead to a lack of transparency in carbon credit certification, making it difficult to meet the verification requirements of the international carbon market. A multi-dimensional grassland carbon sink calculation model is established, covering modules for aboveground biomass carbon storage, belowground root carbon storage, and soil organic carbon content, so that the model covers all aspects of the main carbon sources and main carbon consumption in grasslands. Moreover, the model-based grassland carbon credit calculation method provided in this application collects multi-modal basic data, and the basic data is aligned and processed before being analyzed by the model, ensuring the comprehensiveness and validity of the input data, and thus achieving the accuracy and comprehensiveness of the output results.

[0031] An embodiment of the present application provides a grassland carbon sink calculation model, including a bidirectional long short-term memory network and a multi-task learning model based on the Transformer architecture. The multi-task learning model includes: an aboveground biomass carbon storage module, an underground root carbon storage module, and a soil organic carbon content module; the training data of the model includes grassland ecological knowledge and historical carbon flux data.

[0032] In this embodiment, the grassland carbon sink model based on the Transformer architecture uses multi-dimensional data established by information collected by satellite remote sensing data collection devices, information collected by unmanned aerial vehicle multispectral scanning devices, information collected by underground carbon monitoring devices, and information collected by grazing activity data interfaces as input. The aboveground biomass carbon storage module analyzes the vegetation index (NDVI, EVI), leaf area index (LAI), vegetation coverage, degraded patches, etc. from the input, and calculates the carbon sink of aboveground plants from them; at the same time, the underground root carbon storage module analyzes the aboveground vegetation index, leaf area index, soil moisture, etc. from the input, and calculates the carbon sink of underground roots from them; and the soil organic carbon content module combines the input near-infrared spectral corresponding data and temperature and humidity data to calculate the carbon sink of soil, etc.

[0033] In this embodiment, the Transformer architecture is a deep learning model based on the self-attention mechanism. This model is a prior art and will not be described in detail here. In order to more effectively model the spatiotemporal features in grassland remote sensing data and perform task decomposition, the bidirectional long short-term memory network (BiLSTM) is integrated with the Transformer-based multi-task learning model. BiLSTM is good at capturing the long-term dependency features of the forward and backward directions in time series, and is suitable for characterizing dynamic trends in grassland changes; while the Transformer architecture, through the self-attention mechanism, has the ability to model long-distance dependencies and multi-scale information fusion, and is particularly suitable for parallel processing and multi-task modeling.

[0034] In the integrated architecture, the BiLSTM layer, as a temporal feature extraction module, prioritizes bidirectional encoding of input remote sensing time series data to extract potential evolutionary patterns. Its output serves as the input feature embedding for the Transformer backbone model, which further leverages a multi-head attention mechanism to achieve feature sharing and differential modeling across tasks. Ultimately, the multiple task branches complete subtasks such as grassland classification, degradation monitoring, and biomass estimation. During training, the model is optimized using a joint loss function, enabling multi-dimensional intelligent perception of grasslands.

[0035] The grassland carbon sink calculation module provided in this embodiment also includes a digital elevation module for calculating the impact of terrain on carbon sinks, and a grazing carbon sink adjustment module for calculating the impact of grazing activities on carbon sinks. The grazing carbon sink adjustment module extracts GPS positioning data of livestock during grazing, analyzes grazing paths and density, and combines grassland rotational grazing data to assess livestock grazing intensity and vegetation recovery. When using the grassland carbon sink calculation module to estimate future carbon sinks, the input data corresponding to grassland surface images, aboveground biomass, etc. are adjusted to adjust the prediction results.

[0036] The grassland carbon sink calculation module provided in this embodiment also includes an attention mechanism module for optimizing calculation weights.

[0037] (Attention Mechanism), adjust the weights of each module to improve prediction accuracy.

[0038] The grassland carbon sink calculation module provided in this embodiment also includes an interpretable module, which uses the SHAP (Shapley Additive Explanations) method to calculate the contribution of each input factor to the carbon sink estimation, generate a carbon sink contribution factor weight distribution map, and improve carbon credit transparency.

[0039] The training data in this example covers temperate grasslands and alpine meadows. The output is the carbon storage and uncertainty range for each reservoir. The training data in this example is output in the following table format. Other output formats are also possible. Those skilled in the art can configure this format based on their preferences and circumstances. This is not intended to limit the output format of the training data.

[0040] Another embodiment of the present application further provides a grassland carbon credit calculation method based on a grassland carbon sink calculation model, which applies the grassland carbon sink calculation model described above to calculate the carbon sink amount of a target area.

[0041] like Figure 1 As shown, the grassland carbon credit calculation method provided in this embodiment includes the following steps:

[0042] S1: Use the multimodal data acquisition module to collect basic data of the target area;

[0043] S2: Using a data fusion module to perform spatiotemporal alignment and / or noise filtering on the basic data;

[0044] S3: Inputting the basic data processed by spatiotemporal alignment and / or noise filtering into the grassland carbon sink calculation model to calculate the carbon sink amount.

[0045] In the grassland carbon credit calculation method provided in this embodiment, the multimodal data acquisition module is used to collect basic data of the target area, including:

[0046] Acquiring remote sensing information of a target area using a hyperspectral satellite remote sensing unit, wherein the remote sensing information includes above-ground plant mass and vegetation coverage;

[0047] Use the multispectral scanning unit of the UAV to obtain the surface degradation information of the target area;

[0048] An underground carbon monitoring network is used to obtain underground information, including soil temperature, humidity, and root biomass.

[0049] In the grassland carbon credit calculation method provided in this embodiment, the data fusion module is used to perform spatiotemporal alignment and / or noise filtering on the basic data, including:

[0050] unifying the temporal and spatial references of the remote sensing information, the surface degradation information, and the subsurface information using a spatiotemporal alignment engine;

[0051] The mapping relationship between vegetation index and soil carbon content was established using multi-source data association units based on the basic data after unified benchmarking;

[0052] Use variational autoencoders to repair abnormal and / or missing data.

[0053] In this embodiment, the spatiotemporal alignment engine includes a Kalman filter algorithm. The spatiotemporal alignment engine unifies the temporal and spatial references of the remote sensing information, the surface degradation information, and the subsurface information. This engine uses the Kalman filter algorithm to align multiple spatiotemporal data, unifying timestamps and geographic coordinates, and then correcting the geometric distortion of satellite imagery using a high-precision DEM (full name).

[0054] In this embodiment, the multi-source data association unit is used to establish a mapping relationship between vegetation index and soil carbon content based on the basic data after a unified benchmark, including using a deep neural network (DNN) to establish a nonlinear mapping relationship between NDVI, soil moisture and soil organic carbon, and combining a random forest model to analyze the correlation between vegetation index and underground root carbon.

[0055] In this embodiment, the use of a variational autoencoder to repair abnormal data and / or missing data includes using a variational autoencoder (VAE) to repair missing data to improve data integrity, and removing sensor noise through wavelet transform to improve data accuracy.

[0056] The grassland carbon credit calculation method provided in this embodiment also includes using a dynamic carbon sink correction module to adjust carbon sink amounts based on precipitation, temperature, and grazing data; and using an online incremental learning module to predict future carbon sink trends. If the dynamic carbon sink correction module detects a continuous drought warning, the online incremental learning module reassesses the carbon sink decay rate.

[0057] The dynamic carbon sink correction module described in this embodiment uses an adaptive neural network to model sudden events (such as drought, heavy rain, and increased grazing pressure) in real time. Its input data includes access to national meteorological data to monitor key climate parameters such as precipitation and temperature that affect carbon storage. It also includes a grazing overload alarm model to detect whether livestock density exceeds ecological thresholds in real time. The online incremental learning module described in this embodiment uses a recurrent neural network (RNN) to achieve long-term trend prediction, improving the model's dynamic adaptability.

[0058] The grassland carbon credit calculation method provided in this embodiment also includes accessing livestock GPS positioning collar data to analyze grazing paths and densities; and combining grassland rotational grazing management systems to evaluate livestock feeding intensity and vegetation recovery.

[0059] The grassland carbon credit calculation method provided in this embodiment also includes outputting and displaying the carbon sink amount using a result output and evidence storage module.

[0060] In the grassland carbon credit calculation method provided in this embodiment, the output and evidence storage module outputs carbon sink amounts to generate a timestamped carbon credit certificate and displays grassland carbon sink spatial gradients and degradation risk warning maps via a multi-dimensional visualization platform. The degradation risk warning map can be a heat map, but other types of maps are also possible. Those skilled in the art can choose the appropriate one based on their needs, and this will not be elaborated upon here.

[0061] The output and evidence storage module uploads the calculated carbon sink data to the blockchain after hash processing, and generates a carbon credit certificate at the same time. The carbon sink information can be verified in the blockchain through the carbon credit certificate, making the stored information immutable.

[0062] In this embodiment, the hyperspectral satellite remote sensing unit can use the Sentinel-2 satellite to obtain images with a resolution of 10 meters to 50 meters for analyzing the vegetation index (NDVI). In this embodiment, the vegetation index NDVI can be calculated using NDVI = (near-infrared reflectivity - red light reflectivity) / (near-infrared reflectivity + red light reflectivity). Value range: 0.4 to 0.6 represents high-density grassland (strong carbon absorption capacity); 0.1-0.4 usually represents sparse grassland or short grassland; and ≤0 usually represents water bodies, sandy land, urban areas and other areas without vegetation. NDVI can be used to monitor the changing trend of carbon sequestration capacity in the long term.

[0063] In this embodiment, the multispectral scanning unit of the drone uses a multispectral or hyperspectral camera carried by the drone to obtain degraded grassland information with a resolution of 0.5 to 1 meter. The multispectral spectrum here includes multiple bands, each of which is used to obtain different information: blue light is used to obtain information related to water and soil characteristics, green light is used to obtain information related to vegetation health and chlorophyll concentration, red light (620-750nm) is used to obtain vegetation type and NDVI related information, red edge (700-740nm) is used to monitor plant stress and growth status information, and near infrared is used to obtain vegetation activity and NDVI related information. Thermal infrared and other methods can also be used to obtain soil moisture and temperature change information.

[0064] In this embodiment, each spectrum can be captured using a multi-lens multispectral camera, with each band using a separate lens to capture images, ensuring high spectral resolution. Naturally, the multiple lenses require time synchronization of the images captured, with the spatial alignment between the lenses controlled by a registration system. Spatial registration of images from multiple lenses is well known in the art and will not be discussed further here. Alternatively, an integrated multispectral camera can be used, with a single lens capturing images by switching between wavelengths using a beam splitter prism or filter wheel. While this type of camera may have slightly lower wavelength capture accuracy, it offers advantages such as cost savings and ease of operation. Those skilled in the art can select the camera type as needed, and this will not be discussed further here.

[0065] In order to reduce manual intervention in this embodiment, a route can be set for the drone, and each time the drone patrols, it patrols according to a highly overlapping route.

[0066] In this embodiment, the images acquired by the multispectral unit of the drone can be directly input into the data fusion module without being processed. Of course, some simple processing can also be performed to eliminate related effects, such as repairing missing images due to dust interference.

[0067] The remote sensing information, surface degradation information, underground information, etc. obtained in this embodiment can be transmitted to the data fusion module via a low-power wide area network (LoRa) to ensure real-time data transmission and low power consumption.

[0068] In this embodiment, the spatiotemporal alignment engine is used to align the drone data with the remote sensing imagery, downsample the drone camera data to the same resolution as the remote sensing imagery, and unify the temporal and spatial resolutions of different data sources, making them comparable and fusible at the same scale. Establishing a mapping relationship between satellite imagery and ground sensor data involves modeling and matching the remote sensing imagery with the real-time ground sensor data in terms of space and attributes. Possible mapping models include linear / nonlinear regression models, or machine learning models such as decision trees (CART, XGBoost), random forests, support vector machines (SVM), and neural networks (MLP, CNN). Those skilled in the art can select these models as needed, and this will not be elaborated on here.

[0069] This application first uses a spatiotemporal alignment engine to unify data of varying resolutions and temporal dimensions. Machine learning and modeling algorithms then establish a mapping relationship between remote sensing data and storefront data. Based on this comprehensive spatial and temporal mapping, the application estimates uninhabited areas, improving the accuracy of data on uninhabited areas. This estimation of uninhabited areas ensures the integrity of the entire forest data, which helps improve the accuracy of carbon sink calculations.

[0070] A computer-readable storage medium stores a computer program, which implements the method described above when executed by a processor.

[0071] The grassland carbon sink calculation model, grassland carbon credit calculation method, and storage medium provided in this embodiment achieve full automation of the entire process from data acquisition to carbon credit certification through multimodal data collection, data fusion and quality control, large-scale forestry model calculation, dynamic correction, and result storage. This achieves:

[0072] 1. Full process automation: No manual intervention is required from data collection to authentication, increasing efficiency by over 80%.

[0073] 2. High-precision calculation: Multimodal data fusion reduces the error rate of carbon sink estimation to less than 10% (traditional methods are 25%).

[0074] 3. Dynamic adaptability: Grazing overload events can trigger carbon credit correction within 12 hours.

[0075] 4. Multi-dimensional coverage: Simultaneously calculate above-ground and underground carbon pools to support a complete assessment of grassland carbon sinks.

[0076] 5. Trusted Evidence: Blockchain technology ensures that data cannot be tampered with and complies with international carbon verification standards (such as VCS).

[0077] Through multi-source data fusion and intelligent algorithm optimization, this invention provides an efficient, accurate and transparent grassland carbon credit calculation solution, which can be widely used in carbon market transactions, grassland ecological protection and sustainable grazing management.

[0078] The grassland carbon sink calculation method provided in this embodiment can adopt a variety of optimization schemes for different application scenarios.

[0079] For small and medium-sized pastures, only satellite remote sensing and grazing data can be used to reduce sensor deployment costs. For large-scale erasures, multi-source data fusion (satellite + drone + ground sensor) is used to improve data coverage. For extreme environments, low-power Internet of Things (LoRa) is used to transmit data in real time to improve the system's dynamic response capabilities. Of course, for different schemes, the parameters of each module in the grassland carbon sink calculation model need to be adjusted accordingly. For example, in extreme weather, an adaptive online learning algorithm needs to be started in real time to improve the system's dynamic response capabilities. Those skilled in the art can manually adjust the parameters of each module as needed. Of course, multiple models can be trained with corresponding types of data to adapt to each scenario one by one. Those skilled in the art can make choices as needed, and I will not go into details here.

[0080] As used below, the term “unit,” “sub-unit,” or “module” may be a combination of software and / or hardware that implements a predetermined function.

[0081] Based on the same inventive concept, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method as described in any one of embodiments one to three.

[0082] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, or can be embodied through the implementation process of data migration. All or part of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, mobile communication terminals, multi-processor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0083] Although the present application has been described with reference to the embodiments, those skilled in the art will appreciate that there are many modifications and variations to the present application without departing from the spirit of the present application. It is intended that the appended claims include these modifications and variations without departing from the spirit of the present application.

Claims

1. A grassland carbon sink calculation model, characterized in that , including a bidirectional long short-term memory network and a multi-task learning model based on the Transformer architecture. The multi-task learning model includes: aboveground biomass carbon storage module, underground root carbon storage module and soil organic carbon content module; the training data of the model includes grassland ecology knowledge and historical carbon flux data.

2. The grassland carbon sink calculation model according to claim 1, characterized in that: It also includes an attention mechanism module for optimizing calculation weights.

3. A grassland carbon credit calculation method based on a grassland carbon sink calculation model, characterized in that: The grassland carbon sequestration calculation model as claimed in any one of claims 1 to 2 is used to calculate the carbon sequestration amount of the target area.

4. The grassland carbon credit calculation method according to claim 3, characterized in that: The following steps are involved: Use the multimodal data acquisition module to collect basic data of the target area; Performing spatiotemporal alignment and / or noise filtering processing on the basic data using a data fusion module; The basic data that has undergone spatiotemporal alignment and / or noise filtering processing is input into the grassland carbon sink calculation model to calculate the carbon sink amount.

5. The grassland carbon credit calculation method according to claim 4, characterized in that: The method of collecting basic data of the target area using the multimodal data collection module includes: Acquiring remote sensing information of a target area using a hyperspectral satellite remote sensing unit, wherein the remote sensing information includes above-ground plant mass and vegetation coverage; Use the multispectral scanning unit of the UAV to obtain the surface degradation information of the target area; An underground carbon monitoring network is used to obtain underground information, including soil temperature, humidity, and root biomass.

6. The grassland carbon credit calculation method according to claim 4, characterized in that: The performing of spatiotemporal alignment and / or noise filtering processing on the basic data by using a data fusion module includes: unifying the temporal and spatial references of the remote sensing information, the surface degradation information, and the subsurface information using a spatiotemporal alignment engine; The mapping relationship between vegetation index and soil carbon content was established using multi-source data association units based on the basic data after unified benchmarking; Use variational autoencoders to repair abnormal and / or missing data.

7. The grassland carbon credit calculation method according to claim 4, characterized in that: Also includes: Use the dynamic carbon sink correction module to adjust carbon sinks based on precipitation, temperature, and grazing data; Use online incremental learning module to predict future carbon sequestration trends.

8. The grassland carbon credit calculation method according to claim 4, characterized in that: It also includes accessing livestock GPS positioning collar data to analyze grazing paths and densities; combined with grassland rotational grazing management systems, it evaluates livestock feeding intensity and vegetation recovery.

9. The grassland carbon credit calculation method according to any one of claims 4 to 8, characterized in that: Use the result output and evidence storage module to output and display the carbon sink amount.

10. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method according to any one of claims 3 to 9 is implemented.

Images