A method and system for automatically generating a wetland structure network and assessing carbon sequestration

Abstract

The application discloses a wetland structure network automatic generation and carbon burial evaluation method and system, relates to the technical field of data analysis, and comprises the following steps: generating a SHP format tidal creek vector coordinate file according to a satellite remote sensing image of a research area; extracting basic morphological parameters and topological correlation parameters of the tidal creek from the tidal creek vector coordinate file; constructing a standardized training set according to the vector coordinates, the basic morphological parameters and the topological correlation parameters of the tidal creek in the tidal creek vector coordinate file; training a tidal creek network adaptive generation model according to the standardized training set; generating a target tidal creek network of the research area according to the trained tidal creek network adaptive generation model; simulating the dynamic landforms of the research area by using the target tidal creek network; and evaluating the carbon burial of the research area according to the dynamic landforms. The application can provide a target tidal creek network with the characteristics of large range, high precision, customization and easy adaptation, so that the carbon burial of the research area can be efficiently evaluated through numerical simulation.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a method and system for automatic generation of wetland structure networks and carbon sequestration assessment. Background Technology

[0002] Coastal wetlands, as unique ecosystems in the transitional zone between land and sea, are crucial blue carbon sinks for the Earth. Their organic carbon sequestration processes play a vital role in the global carbon cycle and are one of the core ecological carriers for achieving the dual-carbon strategy goals. Research on the carbon cycle and carbon balance of coastal wetlands has become an important research area in Earth system science. Accurately assessing their carbon sink function is of great significance for ecological protection, climate change response, and regional development planning.

[0003] However, coastal wetlands are generally characterized by their vast area, flat terrain, and complex environment. Traditional on-site survey methods require a large investment of manpower and resources, which is not only inefficient but also limited by natural conditions such as tides and topography. This makes it difficult to achieve large-scale, dynamic monitoring of carbon cycle processes and fails to meet the timeliness and comprehensiveness requirements of carbon sink assessment. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method and system for automatic generation of wetland structure networks and carbon burial assessment, so as to efficiently and accurately assess the carbon burial of coastal wetlands.

[0005] One aspect of this application provides a method for automatic generation of wetland structure networks and carbon sequestration assessment, the method comprising the following steps:

[0006] Generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area;

[0007] The basic morphological parameters and topological correlation parameters of the tidal channel are extracted from the tidal channel vector coordinate file;

[0008] A standardized training set is constructed based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file;

[0009] Train the adaptive generation model of the tidal channel network based on the standardized training set;

[0010] The target tidal channel network for the study area is generated based on the trained adaptive generation model of the tidal channel network.

[0011] The dynamic geomorphology of the study area was obtained by simulating the target tidal channel network.

[0012] Carbon burial in the study area was assessed based on the dynamic geomorphology.

[0013] In some embodiments, generating a tidal channel vector coordinate file in SHP format based on satellite remote sensing images of the study area includes the following steps:

[0014] The satellite remote sensing images of the study area were sequentially subjected to radiometric calibration, FLAASH atmospheric correction, spectral enhancement of tidal channel areas, Gram-Schmidt panchromatic and multispectral band fusion, and topographically constrained geometric fine correction to obtain preprocessed remote sensing images.

[0015] An initial tidal channel network is extracted from the preprocessed remote sensing image;

[0016] The initial tidal channel network is subjected to topology verification and visual interpretation optimization to generate a tidal channel vector coordinate file in SHP format.

[0017] In some embodiments, extracting the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file includes the following steps:

[0018] The length, density, curvature, and dilatation rate of the tidal channel are extracted from the tidal channel vector coordinate file as the basic morphological parameters.

[0019] The main-branch length ratio, branching angle, and number of branch levels are extracted from the tidal channel vector coordinate file as the topological association parameters.

[0020] In some embodiments, constructing a standardized training set based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file includes the following steps:

[0021] An initial training set is constructed based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file;

[0022] The initial training set is subjected to parameter logic verification so that the density and division rate conform to the natural law of fractal dimension.

[0023] The initial training set is subjected to morphological topology verification to remove abnormal morphological samples with intersections or breaks.

[0024] For the difficult example samples with fine tributaries and complex branching in the initial training set, data augmentation is performed using a topology-preserving enhancement method involving rotation and scaling.

[0025] The regional feature normalization method is used to eliminate regional landform differences, and the standardized training set is finally obtained by quantizing and encoding the landform background parameters.

[0026] In some embodiments, the network adaptive generative model adopts a combined architecture of the SegNeXt spatial generative model and the graph attention network parameter regressor;

[0027] The SegNeXt spatial generation model is used to optimize the encoding and decoding network structure. By adding a weight module at the main and branch channel levels of the tidal channel, the encoding layer focuses on extracting the long-distance continuity features of the main channel, while the decoding layer captures the details of the branch channel through multi-scale feature aggregation. Simultaneously, a tidal channel fractal window adaptation mechanism is embedded, and the window size is dynamically adjusted according to the fractal dimension to avoid the fractal structure being fragmented.

[0028] The graph attention network parameter regressor is used to construct the tidal channel parameter graph structure. The main channel parameters are set as core nodes, the tributary channel parameters are set as child nodes, and the landform background is set as edge attributes. The coupling law between key parameters is strengthened through the attention mechanism to achieve adaptive parameter adjustment.

[0029] The network adaptive generation model incorporates biodynamic geomorphological coupling constraints to achieve two-way linkage between spatial morphology generation and parameter matching, ensuring that the generation results have both spatial rationality and parameter coordination.

[0030] In some embodiments, training the adaptive generative model of the tidal channel network based on the standardized training set includes the following steps:

[0031] The first stage involves training to generate the main gully, using the main gully continuity loss as the loss function:

[0032] ;

[0033] in, The main sulcus continuity loss value is used to quantify the difference between the predicted main sulcus and the actual main sulcus in terms of continuity and smoothness of direction; N is the number of training samples. Let be the predicted continuous length of the main channel for the i-th sample; λ is the true continuous length of the main groove of the i-th sample; λ is the curvature smoothness constraint weight; smooth( ) is the function for calculating the curvature smoothness of the predicted main gully; Cpred,i is the predicted curvature value of the main gully for the i-th sample;

[0034] The second stage involves branch generation and dual-architecture fusion training, using a composite loss function:

[0035] ;

[0036] in, The total composite loss value; α is the morphological loss weight; Hierarchical FocalTverskyLoss is used to optimize tidal channel morphology details and class imbalance; β is the parameter matching loss weight. γ is the parameter matching loss value, used to quantify the deviation between the predicted parameters and the actual parameters; γ is the topological rationality loss weight. This is the topological rationality loss value, used to constrain the connection relationship between the main and branch ditches;

[0037] The third stage implements a training optimization strategy. In this stage, the performance is evaluated using a validation set after each round of training. If the loss does not decrease after five consecutive rounds of validation, an early stopping strategy is initiated to prevent overfitting. Throughout the training process, three specific indicators are recorded: the accuracy of main and branch channel connection, the matching rate of branch points, and the parameter deviation rate. Finally, the optimal model weights based on the comprehensive analysis of these specific indicators are saved to obtain the trained adaptive generation model of the tidal channel network.

[0038] In some embodiments, the method further includes the following steps:

[0039] Using Delft3D code as a framework, the following steps are integrated and encapsulated: generating a target tidal channel network for the study area based on the trained adaptive generation model of the tidal channel network; simulating the dynamic geomorphology of the study area using the target tidal channel network; and assessing carbon burial in the study area based on the dynamic geomorphology. This results in an encapsulated module.

[0040] The user-inputted basic configuration information is entered into the encapsulation module to obtain the carbon burial assessment results for the study area.

[0041] Another aspect of this application provides a system for automatic generation of wetland structure networks and carbon sequestration assessment, the system comprising:

[0042] The image processing unit is used to generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area;

[0043] The parameter extraction unit is used to extract the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file.

[0044] The training set construction unit is used to construct a standardized training set based on the vector coordinates of the tidal channel, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file.

[0045] The model training unit is used to train an adaptive generative model of the tidal channel network based on the standardized training set.

[0046] The tidal channel generation unit is used to generate the target tidal channel network of the study area based on the trained adaptive generation model of the tidal channel network.

[0047] The geomorphological simulation unit is used to simulate the dynamic geomorphology of the study area using the target tidal channel network.

[0048] A carbon burial assessment unit is used to assess the carbon burial of the study area based on the dynamic geomorphology.

[0049] Another aspect of this application embodiment provides an electronic device, including a processor and a memory;

[0050] The memory is used to store programs;

[0051] The processor executes the program to implement any of the methods described above.

[0052] Another aspect of this application provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described in any of the above embodiments.

[0053] This application includes at least the following beneficial effects:

[0054] This application can generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area; extract the basic morphological parameters and topological correlation parameters of the tidal channels from the tidal channel vector coordinate files; construct a standardized training set based on the vector coordinates, basic morphological parameters, and topological correlation parameters of the tidal channels in the tidal channel vector coordinate files; train an adaptive generation model of the tidal channel network based on the standardized training set; generate a target tidal channel network for the study area based on the trained adaptive generation model of the tidal channel network; simulate the dynamic geomorphology of the study area using the target tidal channel network; and conduct carbon burial assessment of the study area based on the dynamic geomorphology. This application can provide a large-scale, high-precision, customizable, and easily adaptable target tidal channel network, thereby improving the accurate data foundation for carbon burial in the study area and enabling efficient assessment of carbon burial in the study area through numerical simulation. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating an automatic generation method for wetland structure networks and carbon burial assessment provided in this application embodiment;

[0057] Figure 2 An example flowchart of a method for automatic generation of wetland structure networks and carbon burial assessment provided in this application embodiment;

[0058] Figure 3 An example diagram of a tidal channel network provided in an embodiment of this application;

[0059] Figure 4 An example diagram of an extracted tidal channel network provided in an embodiment of this application;

[0060] Figure 5 Another example diagram of the extracted tidal channel network provided in this application embodiment;

[0061] Figure 6 Example diagram of the user interface provided in the embodiments of this application;

[0062] Figure 7 An example diagram illustrating the visualization of tidal channel network extraction results provided in this application embodiment;

[0063] Figures 8 to 14 A graph showing the carbon burial-related results provided in the embodiments of this application;

[0064] Figure 15 This is a structural block diagram of a wetland structure network automatic generation and carbon burial assessment system provided in an embodiment of this application. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0066] Before providing a detailed description of the embodiments of this application, some related technologies involved in the embodiments of this application will be described first, as follows:

[0067] Terminology Explanation:

[0068] Coastal wetlands: unique ecosystems located in the transition zone between land and sea, including sub-units such as salt marshes and tidal flats, are important blue carbon reservoirs on Earth. Their organic carbon burial process is synergistically regulated by multiple factors such as landforms, hydrology, and vegetation, which is the application scenario of this application.

[0069] Tidal channel network: A dendritic or network-like micro-geomorphic system formed by tidal erosion and deposition on coastal wetland tidal flats. It is the core channel for the migration of tidal water, sediment and organic carbon, and is the key innovative element introduced in this application.

[0070] Tidal channel length: Generally refers to the total length of all tidal channels within the study area. Different studies assign different meanings to the length of a single tidal channel. The average length of tidal channels varies significantly across different areas.

[0071] Tidal channel density: refers to the total length of tidal channels per unit area of ​​tidal flat, used to describe the density of tidal channel development and the coverage of the tidal channel network.

[0072] Tidal channel bifurcation rate: refers to the number of bifurcation points of a tidal channel per unit area of ​​tidal flat, reflecting the stability of the tidal channel. When studying the bifurcation rate of a tidal channel system, the total bifurcation rate should be the weighted average of the bifurcation rates of each level of tidal channel.

[0073] Tidal channel curvature: refers to the ratio of the actual length of the tidal channel to the straight line length between its two ends, reflecting the degree of curvature of the tidal channel. The greater the curvature of the tidal channel, the more meandering the tidal channel is; when the curvature of the tidal channel is equal to 1, the tidal channel is straight.

[0074] Organic carbon sequestration: The process by which organic carbon in coastal wetlands is generated, migrated, and deposited, and then stored in sediments for a long period of time. It originates from carbon inputs such as vegetation residues and suspended organic matter, and is the core link for wetlands to perform their carbon sequestration function.

[0075] Carbon sink assessment: Based on the principles of carbon cycle and standardized accounting methods, this study quantifies the total amount, rate, and spatial distribution characteristics of organic carbon buried in coastal wetlands, reveals its impact patterns, and provides scientific support for enhancing carbon sink functions and ecological restoration.

[0076] Numerical simulation has become the mainstream technical means for assessing carbon sinks in coastal wetlands. By integrating modules such as hydrodynamics, sediment transport, and ecological processes, it enables quantitative analysis of the entire process of organic carbon generation, migration, decomposition, and burial.

[0077] As the most active micro-geomorphic unit in coastal wetlands, the tidal channel network is a core channel connecting land and sea for the exchange of materials and energy. Its morphological characteristics, such as density, bifurcation rate, and curvature, directly regulate tidal flooding patterns, sediment transport efficiency, and the spatial distribution of organic carbon. Existing studies have confirmed that the developmental state of the tidal channel network affects soil redox environment, vegetation distribution, and microbial activity by altering wetland hydrological connectivity, thereby significantly regulating the spatial heterogeneity of organic carbon burial. The main channel and bifurcation nodes of the tidal channel are often hotspots for organic carbon burial, and the lateral transport process of the tidal channel is a key link in the carbon cycle, with its carbon flux on the same order of magnitude as the vertical burial flux. Therefore, incorporating the tidal channel network into the carbon sink simulation framework is a core prerequisite for improving the realism and accuracy of the simulation.

[0078] However, current mainstream carbon sink simulation models generally adopt the initial geomorphological assumption of uniformly exposed tidal flats and do not integrate the tidal channel network as an independent key geomorphological unit into the simulation system. The core reason is that the acquisition of tidal channel data faces significant technical bottlenecks. Currently, the technical solutions for acquiring tidal channel networks both domestically and internationally mainly fall into three categories: Field surveying directly acquires data by setting up transects and measuring cross-sections in the field. While this method can obtain high-precision local parameters, it is limited by environmental factors such as the soft texture of tidal flats and periodic tidal inundation, covering only a small area and failing to capture the complete tidal channel network. Furthermore, the data update cycle is long. Remote sensing extraction, based on optical remote sensing, SAR imagery, or UAV imagery, extracts tidal channel information through algorithms such as exponential calculation and threshold segmentation. While this method can achieve large-scale macroscopic observation, it is easily affected by clouds, fog, and vegetation obstruction, resulting in insufficient accuracy in identifying small tributary tidal channels and complex branching structures. The extracted results require extensive manual correction. Manual design and drawing involves manually planning or designing tidal channel networks using drawing software. This method is suitable for small-scale artificial tidal channel excavation design, but it heavily relies on the designer's experience and lacks adaptation to the fractal patterns of natural tidal channels and the biodynamic geomorphological coupling process. The generated tidal channel network differs significantly from the actual natural tidal channel morphology.

[0079] Although numerical simulation has become the core method for assessing carbon sinks in coastal wetlands, its shortcomings are becoming increasingly apparent when combined with the practical needs of tidal channel data acquisition and model application. This directly leads to significant deviations between the simulation results of existing carbon sink models and the actual situation, failing to meet the requirements for accurate assessment. Specifically, this is reflected in the following four aspects:

[0080] (1) The field survey method has limited coverage and insufficient dynamic monitoring capabilities. Due to the dual limitations of tidal flat environmental conditions and labor costs, field surveys can only obtain local small-scale tidal channel data and cannot capture the morphological characteristics of large-scale complete tidal channel networks; moreover, the data update cycle is long, making it difficult to reflect the dynamic evolution process of tidal channel networks and unable to support the simulation needs of carbon sink models for the long-term evolution effects of tidal channels, thus severely limiting the applicability of the data.

[0081] (2) Remote sensing extraction methods lack accuracy and flexibility and rely on external data. They have weak ability to identify narrow tidal channels and complex branching points, and are easily affected by factors such as image resolution, weather conditions, and vegetation cover, resulting in low data processing efficiency. More importantly, remote sensing technology can only acquire the morphology of existing tidal channels and cannot generate tidal channel networks with specific parameters according to the needs of simulated scenarios (such as different vegetation cover, hydrological conditions, and restoration targets). This results in extremely poor flexibility and makes it difficult to support multi-scenario carbon sequestration comparative analysis.

[0082] (3) The manual design and drawing method is highly subjective and has poor ecological adaptability. Relying on the designer's experience and static parameter calculation, the generated tidal channel network lacks consideration of the fractal growth law of natural tidal channels and the bio-geomorphic feedback mechanism, and deviates significantly from the morphological characteristics and evolution law of real tidal channels; moreover, the design process is time-consuming and labor-intensive, and it is impossible to generate tidal channel networks with different parameter combinations in batches, which makes it difficult to meet the large-scale and customized needs of tidal channel data for large-scale carbon sink simulation or regional ecological restoration planning.

[0083] (4) Poor model-data compatibility and high technical threshold. Existing technologies lack an efficient conversion path between tidal channel parameters and model inputs. Tidal channel data formats need to be manually converted step by step to adapt to the simulation model, which is cumbersome and prone to introducing errors. At the same time, the coupling of tidal channel networks and carbon sink models requires professional technicians to complete, involving complex operations such as multi-software collaboration and parameter debugging. The high technical threshold makes it difficult to effectively convert the ecological effects of tidal channel landforms into carbon sink assessment results, which cannot meet the convenience needs of scientific research and engineering practice.

[0084] In summary, existing tidal channel data acquisition technologies cannot provide large-scale, high-precision, customizable, and easily adaptable tidal channel network data for carbon sink simulation. This leads to mainstream carbon sink models failing to accurately reproduce real wetland geomorphological characteristics due to the unreasonable assumption of uniformly exposed tidal flats, resulting in significant deviations between simulated carbon burial spatial distribution results and actual conditions. Therefore, the core objective of this application is to address the aforementioned technical deficiencies. Specifically, the technical problems to be solved include: overcoming the limitations of existing technologies in terms of coverage and accuracy to achieve automatic generation of tidal channel networks that conform to natural laws; enabling customized input of tidal channel parameters to support the generation of target tidal channel networks under different scenarios; establishing an automated adaptation mechanism between tidal channel data and carbon sink models to lower the technical barrier to use; and deeply integrating the automatically generated tidal channel network with the organic carbon burial model, encapsulating it into an easy-to-use black-box tool to achieve a closed loop from tidal channel parameter input to carbon sink assessment result output.

[0085] Reference Figure 1 This application provides a method for automatically generating wetland structure networks and assessing carbon burial, specifically including the following steps S100~S160:

[0086] S100: Generates SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area;

[0087] S110: Extract the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file;

[0088] S120: Construct a standardized training set based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file;

[0089] S130: Train the adaptive generative model of the tidal channel network based on the standardized training set;

[0090] S140: Generate the target tidal channel network for the study area based on the trained adaptive generation model of the tidal channel network;

[0091] S150: The dynamic geomorphology of the study area is obtained by simulating the target tidal channel network;

[0092] S160: Conduct a carbon burial assessment of the study area based on the dynamic geomorphology.

[0093] The following section will provide a detailed introduction and explanation of the solutions in the embodiments of this application, using specific application examples.

[0094] refer to Figure 2 This embodiment addresses the core pain points in existing coastal wetland carbon sink simulations, such as initial topography deviating from the real scene, difficulty in obtaining tidal channel data, and high technical barriers to use. It proposes an integrated technical solution for adaptive generation of coastal wetland tidal channel networks and assessment of organic carbon burial, aiming to achieve the dual goals of accurate adaptation to real landforms and efficient assessment of carbon sinks.

[0095] Based on the coupling law of biodynamic geomorphology, and relying on publicly available datasets of typical coastal wetlands such as the Yellow River Delta and the Yangtze River Delta, this project integrates multi-source information such as 3m resolution digital elevation, tides, and vegetation. Through machine learning algorithms, it deeply mines the morphological-topological correlation features and natural evolution laws of tidal channels, constructing a dedicated adaptive generation model for tidal channels. Users only need to input basic information or target parameters of the simulated area, and the model can automatically verify its rationality, dynamically match natural laws, and generate a tidal channel network that closely resembles the real scene, significantly reducing reliance on measured or remote sensing data. Furthermore, this generation module is deeply integrated with a mature organic carbon burial simulation model. Through a dedicated linkage mechanism between tidal channels and the carbon cycle, automated data adaptation, and grid topology matching, an integrated tool is formed that integrates the entire process of initial geomorphological generation and carbon sink assessment. This achieves highly efficient and automated operation from basic parameter input to realistic tidal channel topography construction to organic carbon burial assessment result output.

[0096] This solution not only bridges the gap between the existing model's assumption of uniform tidal flats and the actual landform by using multi-source data fusion and a tidal channel-specific adaptive algorithm, thus ensuring the scientific validity of carbon sink assessment, but also simplifies the operation of complex technical steps and lowers the technical threshold through integrated process design and intelligent encapsulation. This promotes the widespread application of accurate carbon sink assessment technology in scientific research and engineering practice, providing professional and practical technical support for coastal wetland protection and ecological restoration planning.

[0097] (1). Acquisition and standardization of public datasets.

[0098] This section focuses on the training requirements of the adaptive generative model for tidal channel networks. Through multi-source data collaborative collection and targeted tidal channel processing, it completes data collection, refined processing, and the construction of a high-quality training set, ensuring the accuracy, completeness, and adaptability of the data to the tidal channel scenario. The specific steps are as follows:

[0099] (a) Study area and selection of multi-source data.

[0100] Typical coastal wetland areas such as the Yellow River Delta and the Yangtze River Delta were selected to construct a multi-source data collaborative support system. Remote sensing images from Jilin-1 wide-swath 01B / C satellites and Gaofen-02 series satellites were acquired. The selected image sensor was a multispectral / panchromatic integrated imaging payload with a panchromatic band resolution of 0.5m-0.75m and a multispectral band resolution of 2m-3m. The spectral range covered the 450nm-800nm ​​panchromatic band, 450nm-510nm blue band, 520nm-590nm green band, 630nm-690nm red band, and 770nm-890nm near-infrared band. Single images... The imagery spans from 11.6km to 40km, with data in TIFF format, including standardized naming information such as center latitude and longitude and imaging time. Simultaneously, a 3m resolution digital elevation model, tidal observation data, and vegetation distribution vector data for the study area are acquired, providing geomorphological and hydrological constraints for tidal channel extraction and parameter correlation. Strict imagery acquisition conditions are applied, selecting clear weather during low tide (tide level 1.0m below the local average tide level) with cloud cover ≤10%, covering spring, summer, and autumn. Seasonal images corresponding to different developmental stages of the tidal channels are matched to ensure comprehensive capture of tidal channel morphological evolution characteristics and spectral response differences, laying the foundation for subsequent extraction work.

[0101] (b). Remote sensing image preprocessing and tidal channel enhancement.

[0102] Standardized preprocessing was performed on the acquired remote sensing images of the target area, including radiometric calibration, FLAASH atmospheric correction, spectral enhancement of the tidal channel area, Gram-Schmidt panchromatic and multispectral band fusion, and topographically constrained geometric fine correction. Radiometric calibration eliminated sensor response differences, and atmospheric correction removed atmospheric scattering and absorption interference. To address the spectral overlap between the tidal channel and bare beaches / vegetation, narrowband filtering and spectral angle matching algorithms were used to enhance the spectral signal of the tidal channel water and suppress background interference. Band fusion balanced spatial detail and spectral fidelity, improving the spatial resolution of small tributaries. Topographically constrained geometric fine correction was performed by combining digital elevation models and tidal observation data to correct topographic deformation errors caused by tides, ensuring the accuracy of spatial positioning of the tidal channel. Ultimately, this significantly improved the image quality and recognizability of the tidal channel area, providing high-quality data support for the fine extraction of tidal channels.

[0103] (c) Fine extraction and coordinate calibration of tidal channels.

[0104] Tidal channel extraction was conducted using a combination of multi-feature fusion extraction and topology rule verification. First, a multi-feature fusion extraction model was constructed, incorporating normalized water index, topographic slope, and texture features. Then, leveraging the spectral differences between tidal channels and bare beaches / vegetation, the topographic slope constraints related to erosion development, and continuous linear texture features, the tidal channel regions were quickly separated, completing the initial extraction of the tidal channel network. The initial extraction results underwent tidal channel-specific topology verification, including main-branch connectivity detection, bifurcation point rationality verification (bifurcation angles conforming to the natural range of 30-60°), and no-overlap verification, with automatic correction. The system addresses issues such as edge misjudgment and breakage. Visual interpretation is then used to optimize the automatically verified results, specifically correcting omissions of small tributaries and breaks caused by shadows. A well-developed and interconnected tidal channel network is selected as the basic data source for simulation. Subsequently, key nodes such as tidal channel edge feature points, branching points, and endpoints are manually extracted. Following a precision standard of 0.5m spacing between main channel nodes, 1.0m spacing between tributary nodes, and precise labeling of branching points, SHP format tidal channel vector coordinate files are generated and output to ensure accurate recording of tidal channel spatial morphology information.

[0105] (d) Extraction of morphological and topological parameters and construction of training set.

[0106] Based on the aforementioned SHP format tidal channel vector data, a self-developed coastal wetland tidal channel network topology-morphology collaborative extraction method was used to accurately extract multi-dimensional parameters. The parameter system covers four basic morphological parameters: length, density, curvature, and bifurcation rate, as well as topological correlation parameters such as the ratio of main channel to branch channel length, bifurcation angle, and the number of branch channel levels. Among them, the tidal channel length is obtained through actual measurement of the centerline, the density is calculated based on the total length of the tidal channel per unit area, the curvature is the ratio of the actual length of a single tidal channel to the straight-line distance between its two ends, the bifurcation rate is the number of bifurcation points per unit area, the bifurcation angle is calculated by the angle between the centerlines of the main and branch channels on both sides of the bifurcation point, the ratio of main channel to branch channel length is the ratio of the total length of the branch channel to the length of the corresponding main channel, and the number of branch channel levels is defined sequentially from the branch level of the main channel to the terminal branch channel.

[0107] Taking into account the differences in tidal channel development stages (juvenile, mature, and declining stages), geomorphological backgrounds (high and low tidal flats, different vegetation cover areas), and topological complexity (simple dendritic, complex network) between the two study areas, a total of 150 natural tidal channels were selected, along with their corresponding vector coordinates, basic morphological parameters, and topological correlation parameters, to construct an initial training set. The initial training set underwent rigorous double validation: parameter logic validation ensured that density and branching rate conformed to the natural law of fractal dimension 1.3-1.8; morphological and topological validation eliminated abnormal morphological samples such as intersections and breaks. Simultaneously, for difficult samples such as small tributaries and complex branching, data expansion was performed using topology preservation enhancement methods such as rotation and scaling to improve the model's learning performance in challenging scenarios. A regional feature normalization method was used to eliminate regional geomorphological differences between the Yellow River Delta and the Yangtze River Delta. By quantifying and encoding geomorphological background parameters, the initial training set acquired cross-regional generalization capabilities, providing comprehensive, accurate, and highly adaptable data support for the subsequent training of the adaptive generation model of the tidal channel network, ultimately resulting in a standardized training set.

[0108] (2). Training of adaptive generative model of tidal channel network.

[0109] This step, based on a standardized training set, utilizes a machine learning architecture adapted to the spatial topological features and parameter correlation patterns of tidal channels to uncover the evolutionary characteristics of natural tidal channels. It constructs a generative model with adaptive parameter adjustment capabilities, providing core algorithmic support for the accurate reconstruction of realistic landforms. The specific steps are as follows:

[0110] (a). Model architecture design.

[0111] A combined architecture of the SegNeXt spatial generation model and a Graph Attention Network (GAT) parameter regressor is adopted to balance the generation of tidal channel spatial morphology with dynamic parameter matching, ensuring that the results conform to the laws of natural evolution. The SegNeXt spatial generation model optimizes the encoding and decoding network structure, adds hierarchical weight modules for the main and branch channels of the tidal channel, the encoding layer focuses on extracting long-distance continuity features of the main channel, and the decoding layer captures the details of branch channel bifurcation through multi-scale feature aggregation. At the same time, a fractal window adaptation mechanism for the tidal channel is embedded, and the window size is dynamically adjusted according to the fractal dimension to avoid the fractal structure being fragmented.

[0112] The Graph Attention Network (GAT) parameter regressor constructs the tidal channel parameter map structure, setting the main channel parameters as core nodes, the tributary parameters as child nodes, and the geomorphic background as edge attributes. Through an attention mechanism, it strengthens the coupling relationship between key parameters, providing precise algorithmic support for adaptive parameter adjustment. A dual-architecture approach embeds biodynamic geomorphic coupling constraints, enabling bidirectional linkage between spatial morphology generation and parameter matching, ensuring that the generated results possess both spatial rationality and parameter coordination.

[0113] (b) Training data preprocessing.

[0114] The standardized training set was specifically transformed to eliminate the impact of data heterogeneity and convert it into an input format that the model could directly learn. The centerline and boundary data of the tidal channels in SHP format were converted into binary raster data at a resolution of 0.5m-0.75m to match the resolution of the remote sensing images. The tidal channel regions were assigned a value of 1, and the background regions were assigned a value of 0. Simultaneously, tidal channel topology enhancement processing was performed, and main and branch channel markers were added to the raster data to form a spatial feature map containing hierarchical information, which was used as the input of SegNeXt.

[0115] The four core parameters were normalized and mapped to the 0-1 range to eliminate dimensional differences. The geomorphic background data was quantized and encoded, then concatenated with the parameter features to form a combined vector, which served as the input feature for the Graph Attention Network (GAT). The 150 tidal channel samples were divided into training, validation, and test sets in a 7:2:1 ratio to ensure that each dataset covered samples from different developmental stages and geomorphic backgrounds, thus improving the model's generalization ability.

[0116] (c) Phased training process.

[0117] We adopt a tidal channel growth-based phased training strategy and rely on the PyTorch framework to gradually optimize the model. The core innovations are the tidal channel-specific composite loss function and dynamic training rules.

[0118] Phase 1: Main channel generation training (rounds 1-50).

[0119] Input a spatial feature map containing hierarchical information, focusing on training the long-distance continuity and orientation rationality of the main sulcus. The optimizer used is Adan, with an initial learning rate of 0.001. The learning rate is dynamically bound to the main sulcus breakage rate: when the breakage rate exceeds 3%, the learning rate is automatically reduced by 40%; when the breakage rate is below 1%, the learning rate remains unchanged.

[0120] The loss function uses the main channel continuity loss ( ):

[0121] ;

[0122] : Main ditch continuity loss value, used to quantify the difference between the predicted main ditch and the actual main ditch in terms of continuity and smoothness of direction. N: Number of training samples, i.e., the total number of tidal ditch main ditch samples participating in this round of training. : The predicted continuous length (m) of the main channel for the i-th sample. : The true continuous length of the main gully for the i-th sample (m). λ: Curvature smoothness constraint weight (0.3), based on the conventional weight range set for linear landforms generated in the industry, balancing length matching and morphological smoothness. smooth( ): The function for calculating the curvature smoothness of the main channel, which calculates the curvature fluctuation value through a sliding window to constrain the main channel to avoid excessive meandering. Cpred,i: The predicted curvature value of the main channel for the i-th sample.

[0123] The above formula compares the continuous length of the predicted main ditch with that of the actual main ditch, and combines the curvature smoothness constraint to ensure that the main ditch conforms to the natural direction of tidal erosion. λ is set to 0.3 because the continuity of the main ditch is the core objective, and smoothness is used as an auxiliary optimization to avoid the main ditch being too straight due to excessive weight.

[0124] Phase 2: Branch generation and dual-architecture fusion training (rounds 51-80).

[0125] Based on the trained main gully features, tributary gully generation training is triggered. The SegNeXt encoding layer features are concatenated with the parameter-landform combination vector and then input into GAT. The optimizer still uses Adan, with the learning rate adjusted to 0.0005. The learning rate decay is tied to the fractal dimension bias: when the bias is less than 5%, it decays by 50% every 10 rounds; when the bias exceeds 10%, it decays by 20% every 10 rounds, prioritizing the preservation of fractal features.

[0126] The core innovation is a composite loss function specific to tidal channels ( Taking into account morphological accuracy, parameter matching, and topological rationality:

[0127] ;

[0128] : Composite total loss value, comprehensively measuring morphological accuracy, parameter matching degree, and topological rationality. α: Morphological loss weight (value 0.6), based on the technical logic that morphology is the foundation of carbon sink simulation, prioritizing that the tidal channel morphology conforms to natural laws. : Hierarchical FocalTverskyLoss, used to optimize tidal channel morphology details and address category imbalance issues. β: Parameter matching loss weight (value 0.3), second only to morphology optimization priority, ensuring that parameter coupling patterns conform naturally. : Parameter matching loss value, quantifying the deviation between predicted parameters and actual parameters. γ: Topological rationality loss weight (value 0.1), an auxiliary optimization term to avoid excessive constraints leading to shape distortion. : Topological rationality loss value, constraining the connection relationship between main and branch channels.

[0129] The ratio of α:β:γ = 0.6:0.3:0.1 follows the technical priority of first ensuring reasonable morphology, then optimizing parameter adaptation, and finally improving topological details. This meets the actual needs of tidal channel generation and carbon sink simulation. The weight values ​​are all within the conventional reasonable range (0.1-0.7) of multi-objective optimization in the industry.

[0130] (1). Hierarchical FocalTverskyLoss ( ).

[0131] The main issues addressed are the imbalance between different types of tidal channels and the optimization of details. The formula is:

[0132] ;

[0133] k: Type identifier of the core area of ​​the tidal channel (k=1 corresponds to the branching point, k=2 corresponds to the main channel, k=3 corresponds to the tributary channel). Loss weights for different regions ( =1.2、 =1.0、 =0.8), based on the regional importance setting, the branching point is the core node of the tidal channel network, with the highest weight. : The number of true positives in the k-th region (the number of pixels predicted as tidal channels and actually being tidal channels). : The number of false positives in the k-th region (the number of pixels predicted as tidal grooves but actually being background). : False negatives in the k-th class region (the number of pixels that are actually tidal channels but are predicted as background). δ: Balance coefficient (value 0.6), referring to the standard value for handling class imbalance in FocalTverskyLoss, balancing precision and recall.

[0134] The weight of the branch point is 1.2 because it is a key connecting node in the tidal channel network, and its accuracy directly affects the rationality of the overall topology. The weight of the tributary is 0.8, but the narrow tributary samples are multiplied by an additional 1.5 times the fit coefficient (based on the class imbalance characteristic of narrow tributaries having a low proportion in the samples) to ensure that small tributaries are not missed. δ=0.6 focuses more on recall rate and is adapted to the extraction needs of narrow tidal channel landforms.

[0135] (2) Parameter matching loss ).

[0136] The formula used to optimize the parameter prediction accuracy of GAT is:

[0137] ;

[0138] Parameter description: j: Identifier of core parameters of tidal channel (j=1 corresponds to length, j=2 corresponds to density, j=3 corresponds to dilatation ratio, j=4 corresponds to curvature); : The predicted value of the j-th parameter of the i-th sample (normalized to the range of 0-1); : The true value of the j-th parameter of the i-th sample (normalized to the range of 0-1); N: The number of training samples; The averaging coefficients of the four main parameters ensure that each parameter has an equal weight.

[0139] The root mean square error (RMSE) is used to quantify the parameter deviation. The four parameters are averaged because they have an equally important impact on tidal channel formation. The formula design conforms to the conventional evaluation logic of parameter prediction in machine learning, and is also adapted to the coupling characteristics of the four core parameters of tidal channel.

[0140] (3) Loss of topological rationality ( ).

[0141] The key constraint is the connection between the main and branch ditches, and the formula is:

[0142] ;

[0143] M: Number of branching points, i.e., the total number of nodes connecting the main and branch channels in the current sample; Predicted bifurcation angle at the m-th bifurcation point (unit: °); : The ideal bifurcation angle (unit: °) of the m-th bifurcation point, with a value of 30-60°. Based on the published conclusions of measured data from tidal channels in the Yellow River Delta and the Yangtze River Delta, this range conforms to the natural bifurcation pattern formed by tidal erosion.

[0144] By calculating the average absolute error between the predicted bifurcation angle and the ideal angle, the connection between the main and branch channels is constrained. When the angle deviation exceeds 20°, the loss value is automatically amplified (achieved through dynamic weights in the training strategy) to avoid unreasonable bifurcation angles that are too steep or too gentle, ensuring that the topology of the tidal channel conforms to the laws of natural evolution.

[0145] Phase 3: Training and optimization strategies.

[0146] After each training round, the performance is evaluated using a validation set. If the loss does not decrease after 5 consecutive validation rounds, an early stopping strategy is initiated to prevent overfitting. Throughout the training process, three specific indicators are recorded: the accuracy of main-branch connection, the matching rate of branch points, and the parameter deviation rate. Finally, the model weights with the best comprehensive combination of these specific indicators are saved.

[0147] (d) Construction of adaptive adjustment mechanism.

[0148] Based on the trained combined model, a two-level adjustment logic of parameter self-verification and terrain self-adaptation is constructed to achieve dynamic matching between input parameters and the real scene. Regarding parameter self-verification, when the user inputs only some core parameters, GAT automatically matches the reasonable range of the remaining parameters through learned coupling rules; when multiple input parameters have logical contradictions, the system automatically triggers fractal dimension verification based on the fractal dimension formula (…). =1.3+0.2×ln(D×B), where D is density and B is diffraction rate) Output recommended adjustment value to ensure that the fractal dimension is within the natural range of 1.3-1.8 (this range is the conclusion of publicly available research on the fractal characteristics of coastal wetland tidal channels).

[0149] In terms of terrain self-adaptation, after the user inputs the terrain background of the simulated area, the model calls the features of tidal ditch samples of similar terrain in the training set to fine-tune the generation parameters and spatial morphology: in high elevation gradient areas (>1.0m), the density of tidal ditch is automatically reduced by 10%-15% (based on the natural law that the higher the elevation, the weaker the tidal ditch development); in shoreline bends (curvature>1.5), the number of bifurcation points is increased by 8%-12% (based on the terrain characteristics of complex tidal dynamics and easy bifurcation formation in shoreline bends); in areas with suitable tidal ditch development (elevation -0.5~1.0m), the parameters are kept unchanged; and in unsuitable areas (elevation>1.5m or <-1.0m), tidal ditch generation is blocked to ensure that the tidal ditch network is accurately adapted to the terrain of the target area.

[0150] (e) Model performance verification.

[0151] Multi-dimensional quantitative indicators were used to verify the reliability of the model, ensuring that the generated results met the real-world geomorphological input requirements for carbon sink simulation. In the spatial morphology verification, the test set geomorphological background was input into the model, generating tidal channel raster data and converting it back to vector format. Compared with real tidal channel vector data, the centerline overlap was ≥89%, the branching point matching rate was ≥84%, and there were no intersections, breaks, or other morphologies that did not conform to natural laws. The recognition rate of small tributaries was significantly improved compared with traditional models.

[0152] In parameter accuracy verification, the four core parameters generated by the model were compared with the real parameters in the test set. The average relative error was ≤10%, and the maximum relative error of a single parameter was ≤15%. The parameter coupling pattern was consistent with the measured data of natural tidal channels. In scene adaptation verification, an independent area of ​​the Yellow River Delta that was not involved in the training was selected as the test scene. After inputting basic geomorphological information, the tidal channel network generated by the model was verified by field survey data. The difference between the key parameters and the real values ​​was ≤15%, which meets the geomorphological input requirements for carbon sink simulation.

[0153] Through the above training and verification, the model has the dual capabilities of accurate spatial morphology generation and adaptive parameter adjustment. It can quickly output tidal channel network data that fits the real scene based on the basic information or target parameters input by the user, laying the core algorithm foundation for subsequent coupling with organic carbon burial models.

[0154] (3) Automatic generation of tidal channel network and organic carbon burial assessment are integrated into a single package.

[0155] This section involves in-depth independent modification and cross-module integration based on the Delft3D source code. Through a dedicated data interaction protocol for tidal channels and an intelligent batch processing workflow, it achieves a fully integrated operation encompassing adaptive generation of tidal channel networks, simulation of dynamic geomorphological evolution, and accurate assessment of organic carbon burial. The specific steps are as follows:

[0156] (a) Cooperative coupling architecture design.

[0157] Using a modified Delft3D source code as the core framework, and based on the carbon cycle module division logic of relevant organic carbon burial models, four core modules—organic carbon generation, transport and exchange, decomposition, and burial—are embedded in the Delft3D dynamic geomorphological simulation kernel. A tidal channel feature-carbon process linkage intermediate layer is simultaneously constructed, forming a synergistic coupling architecture integrating geomorphology, hydrology, and carbon cycle. By reconstructing the process call logic of the Delft3D source code, a real-time bidirectional feedback mechanism for hydrodynamics, sediment transport, topographic evolution, and organic carbon cycle is established: the flow field vector, spatiotemporal distribution of sediment flux, and dynamic elevation change data output by Delft3D are converted into precise driving inputs for the organic carbon module via the intermediate layer; the results calculated by the organic carbon module, such as the spatial distribution of carbon concentration, burial rate gradient, and carbon pool conversion efficiency, are synchronously fed back to the topographic evolution module, dynamically adjusting the sediment-carbon co-deposition coefficient and micro-geomorphological evolution rate. The output interface of the adaptive generation model of tidal channel network is deeply integrated, so that the generated tidal channel vector data (including basic morphological parameters and topological correlation parameters) can be directly adapted to the terrain initialization process of Delft3D. Based on the micro-topographic data of the 3m resolution digital elevation model, the model automatically divides the tidal channel into functional units such as the main channel, the branching zone, and the floodplain, and configures differentiated carbon transport and burial parameters. Through the dynamic mapping rules between the tidal channel topological parameters and carbon module parameters, the model realizes exclusive adaptation logic such as the enhancement of carbon sink hotspots at the branching points and the amplification of lateral carbon flux in the main channel. This constructs a deeply coupled technical architecture of tidal channel generation, dynamic geomorphological simulation and organic carbon assessment.

[0158] (b) Integrated parameter transfer.

[0159] A combined approach of standardized file parsing and dynamic parameter mapping is employed to ensure accurate and efficient transmission of all input parameters to the integrated system. The system features a modular, visual user interface, allowing users to input basic configuration information without manually modifying any underlying model files. Inputs include slope settings, study area size, grid resolution, computation time, sediment particle size, geomorphological acceleration factor, multi-dimensional parameters of the tidal channel network (including basic morphology and topology parameters), and core organic carbon parameters. After receiving user input, the system completes three layers of parameter processing through a built-in intelligent parsing script: automatically generating or modifying the Delft3D MDF configuration file, accurately inputting basic parameters such as hydrodynamics, sediment, vegetation, and organic carbon into the corresponding configuration segments; automatically identifying tidal channel parameters or imported POL format files, and completing the refined definition of tidal channel area water depth in the QUICKIN module based on a 3m resolution digital elevation model, with the main channel water depth being 0.6-0.8m deeper than the surrounding tidal flats, and the tributary channel water depth dynamically adjusted according to the ratio of the main and tributary channel lengths, combined with slope settings to construct an initial landform containing the fine topology of the tidal channel; automatically converting the multi-dimensional parameters of the tidal channel into a format compatible with the organic carbon module, achieving precise adaptation through preset dynamic mapping rules, adjusting the carbon transport and diffusion coefficient stepwise according to the density of the bifurcation points, the ratio of the main and tributary channel lengths affecting the lateral carbon flux distribution ratio, and the bifurcation angle constraining the spatial distribution range of carbon deposition hotspots, and simultaneously updating the parameter configuration segments of the transport and burial modules.

[0160] (c) Automated operation encapsulation.

[0161] By writing multi-threaded intelligent batch processing scripts, the entire process logic of the integrated system is encapsulated, completely eliminating dependence on the Delft3D graphical user interface and achieving high-concurrency, low-intervention automated operation. The script incorporates three core control logic layers: a time-series logic following tidal channel generation → terrain initialization → hydrodynamic simulation → sediment transport calculation → terrain evolution update → carbon process coupling → result feedback. It sequentially calls the adaptive generation model of the tidal channel network and the self-modified Delft3D coupling source code, ensuring real-time data interaction and synchronized calculation steps across modules. In addition to routine parameter integrity and format correctness checks, it simultaneously performs dynamic quality control of tidal channel morphology and verification of carbon process rationality. During the simulation, it monitors tidal channel connectivity and bifurcation angles in real time (maintaining a natural 30-60° angle). The simulation measures the integrity of the connection between the main and branch channels and the tidal channel. If morphological abnormalities occur, local calculation rollback and parameter fine-tuning are automatically triggered. At the same time, the synergy between carbon concentration distribution and topographic evolution is verified. During operation, intermediate data is recorded at preset time steps, and simulation parameters are dynamically adjusted based on the evolution status of the tidal channel. When the tidal channel branching rate increases by more than 10%, the simulation accuracy of lateral carbon transport is automatically improved. When the elevation change exceeds 0.1m, the depth decay coefficient of organic carbon decomposition rate is updated synchronously to ensure the synergistic consistency of tidal channel evolution, dynamic geomorphology and carbon cycle process. No human intervention is required throughout the process.

[0162] (d) Results extraction and visualization.

[0163] After the system is running, it automatically generates a multi-format result file package in the preset output path. The core is a .mat format native data file, which contains total organic carbon burial, spatial distribution concentration, burial rate gradient, spatiotemporal sequence of tidal channel-carbon transport flux, and dynamic data of topographic evolution. The data format is fully compatible with the native output of Delft3D and supports cross-platform secondary analysis. The visualization presentation offers multi-level interactive functions: users can import .mat files into the Delft3D FLOW-QUICKPLOT interface and view the spatial distribution characteristics of organic carbon burial and carbon transport paths along tidal channels by customizing rendering parameters. It supports dynamic playback and spatial slicing analysis at multiple time steps. The system has a built-in Python intelligent analysis script that automatically extracts the correlation features between tidal channel topology and carbon burial, generating carbon burial hotspot maps at bifurcation points, carbon transport flux comparison curves between main and branch channels, and correlation heatmaps of tidal channel density and carbon burial rate. The script can batch read key data and convert it into standardized datasets in Excel or CSV format, including core indicators of organic carbon burial, tidal channel topological parameters, and correlation analysis indicators between the two (such as the contribution rate of carbon burial at bifurcation points and the proportion of carbon transport in the main channel). This provides professional and practical quantitative data support for the accurate assessment of carbon sinks in coastal wetlands and the quantification of ecological restoration effects.

[0164] After the above encapsulation, the system possesses core features such as intelligent input parameters, collaborative operation, correlated output results, and low operational threshold. Users only need to input basic configuration information through the interface to start the full-process simulation, significantly reducing the technical barrier to use. At the same time, through independent modification of the source code and the exclusive coupling design of tidal channels, it achieves deep linkage between tidal channel topography, dynamic geomorphological evolution, and organic carbon processes, ensuring the accuracy and scientific nature of carbon sequestration assessment results, and providing a highly adaptable technical tool for the assessment of carbon sink functions and ecological protection planning of coastal wetlands.

[0165] An optional implementation method is as follows:

[0166] This embodiment takes the coastal wetlands of the Yellow River Delta as the research object, sets a rectangular computational domain of 2500m×1000m, and uses a 1m×1m grid resolution to achieve fine characterization, fully reproducing the entire process from basic information acquisition to carbon burial assessment. The specific steps are as follows:

[0167] (1). Acquisition of basic information about the study area.

[0168] Reference Figure 3This study focuses on the typical tidal flat landforms of the Yellow River Delta. The computational domain encompasses three levels of geomorphic units: high tide, mid-tidal flat, and low tide. The landward boundary connects to the Spartina alterniflora salt marsh area, while the seaward boundary extends to the intertidal zone and the shallow sea transition zone. Based on a 30m resolution digital elevation model (DEM), the elevation range of the computational domain is determined to be -1.5m to 1.8m, with a preset uniform slope of 1 / 1000, consistent with the gentle geomorphic properties of the region's tidal flats. Hydrological and sediment information is derived from measured data from long-term observation stations in the Yellow River Delta coastal wetlands. The tides are dominated by the M2 confluence, with a period of 12.42 hours. The average tidal range in spring is 3.0m, and the low tide level is 1.0m below the average tidal level. The median sediment particle size is 0.08mm, the boundary sediment concentration is 0.11kg / m³, the critical erosion shear force is 0.12Pa, and the Chezy coefficient is 66m¹ / ² / s. The basic data on vegetation and organic carbon were obtained through field surveys. The core vegetation was Spartina alterniflora, with a maximum aboveground biomass growth rate of 1.25 kg / (m²・a), a root-to-shoot ratio of 4.2-6.0 that varied linearly with elevation gradient, and a soil organic carbon background value of 10-32 g / kg. Based on 25 soil core samples, the carbon pools were divided into fast-decomposing carbon pools (33%), slow-decomposing carbon pools (52%), and inert carbon pools (15%), providing experimental support for model parameterization.

[0169] (2). Parameterized generation of tidal channel network.

[0170] Reference Figure 4 , Figure 5 Users input target parameters through the system interface, setting the tidal channel density to 0.6 km / km² and the branching rate to 2.5 channels / km². After selecting the parameterized automatic generation mode, the system automatically matches matching parameters with a curvature of 1.1-1.3 and an average length of 1.0 km for a single tidal channel based on preset association rules. Simultaneously, it determines topological association parameters such as the main-branch channel length ratio of 1:0.6, the branching angle of 35-55°, and the number of branch channel levels of 2-3, generating a parameter configuration scheme for user confirmation. The system then invokes a parameterized tidal channel network automatic generation model (a combination of the SegNeXt spatial generation model and the graph attention network (GAT) parametric regressor architecture) to quickly generate vector data that conforms to the development pattern of tidal channels in the Yellow River Delta, based on the DEM data of the study area and input parameters. The data is automatically exported as an SHP format file, containing the coordinates of the tidal channel centerline, boundary range, and water depth attributes. The water depth in the tidal channel area is 0.6-0.8m deeper than the surrounding tidal flats, and the water depth of the main channel decreases by 0.1-0.2m according to the tributary channel levels, adapting to the actual development depth characteristics of the tidal channels in the region. The format conversion is automatically completed by the system's built-in Python script. The SHP format data is converted to JSON and parsed from TXT coordinates to generate a POL format file compatible with the Delft3D QUICKIN module. The entire process requires no manual intervention, ensuring the accuracy and efficiency of data format adaptation.

[0171] (3) System parameter input and automated operation.

[0172] Reference Figure 6 Users can supplement and enter basic calculation parameters and organic carbon module parameters through a visual interface. The calculation time is 3650 days (10 years), the landform acceleration factor is 100, the time step is 0.1 min, the carbon conversion coefficient is 0.45, the decomposition rate of the fast decomposition carbon pool is 0.4a⁻¹, the decomposition rate of the slow decomposition carbon pool is 0.04a⁻¹, the decomposition rate of the inert carbon pool is 0.001a⁻¹, and the decomposition rate decreases by 13% for every 10 cm increase in depth. After receiving all input parameters, the system initiates an automatic configuration process. First, it modifies the Delft3D MDF configuration file, accurately inputting parameters such as hydrodynamics, sediment, vegetation, and organic carbon into the corresponding fields. Second, it imports the POL format tidal channel file into the QUICKIN module, and combines it with DEM data to complete the initial geomorphological construction of the computational domain, achieving precise topological matching between the tidal channel and the 1m×1m grid. Finally, it automatically extracts the tidal channel morphology parameters and topological correlation parameters, converts them into a format compatible with the organic carbon transport module according to preset dynamic mapping rules, adjusts the carbon transport and diffusion coefficient corresponding to the tributary density in a stepwise manner, and updates the ratio of main and branch channel lengths to the lateral carbon flux distribution ratio, synchronously updating the parameter configuration section to establish a dynamic correlation between tidal channel morphology and carbon cycle process. The system runs automatically via a batch script (.bat). After the user clicks the start calculation button on the interface, the script sequentially calls the tidal channel generation module, the self-modified Delft3D coupling source code, and the organic carbon cycle module. Before running, it automatically checks the rationality of the parameter logic and the integrity of the files. If there are no abnormalities, it starts the full-process calculation. During the process, it outputs intermediate logs daily, and the user can view the progress in real time. No manual intervention is required throughout the entire process.

[0173] (4) Results extraction, visualization and verification.

[0174] After the system finishes running, it automatically generates a raw result file in .mat format along a preset path. This file contains total organic carbon burial, spatial distribution concentration, spatiotemporal series of tidal channel-carbon transport flux, and 10-year topographic evolution data. Simultaneously, it automatically converts the data into a standardized Excel dataset using a built-in script, covering carbon burial rates for each grid unit, tidal channel morphology and topological parameters, and correlation analysis indicators (bifurcation point carbon burial contribution rate, main channel carbon transport proportion), etc. (Refer to...) Figure 7The visualization process eliminates the need for manual import into Delft3DGUI; the system directly presents the spatial distribution map of organic carbon burial and carbon transport pathways along tidal channels through the interface. It supports annual dynamic playback and spatial slice analysis, simultaneously outputting carbon burial hotspot maps at branch points and carbon transport flux comparison curves between main and tributary channels. Multi-dimensional verification results show that the system-generated tidal channel vector data, compared with 60 tidal channels surveyed in the field, exhibits an 89% overlap of centerlines, an 84% matching rate of branch points, and an average relative error of ≤12% between the main-tributary channel length ratio and branch angles and measured values. No intersections, breaks, or other irregularities contrary to natural laws are observed. Selecting 12 soil core sampling points, the average relative error between the simulated and measured carbon burial rate values ​​is 10.8%, with a maximum error of 16.5%, meeting the accuracy requirements. The simulated total organic carbon burial value for the study area is 1.32 × 10⁻⁶. 4 t / year, compared with the results of the Yellow River Delta region carbon sink assessment literature (1.25×10 4 -1.38×10 4 The carbon sequestration contribution rate at the t / year level was consistent with that at the tributary points, reaching 32%, and the carbon transport rate at the main channel was 68%, which is consistent with the regulation law of tidal channels on carbon sinks and verifies the system's scenario adaptability.

[0175] Figures 8 to 14 A graph showing the results related to carbon burial.

[0176] The beneficial effects of this embodiment include:

[0177] This embodiment addresses the core shortcomings of existing coastal wetland tidal channel data acquisition methods, the disconnect between carbon sink simulation and real landforms, and the high technical threshold for operation. It upgrades existing technologies by constructing an integrated technical approach for automatic tidal channel network generation and organic carbon burial assessment, with the following specific benefits:

[0178] (1) Improve the rationality and scenario adaptability of tidal channel data acquisition.

[0179] To address the limitations of field surveys (narrow coverage), remote sensing extraction (insufficient accuracy and scenario adaptability), and manual design (high subjectivity), this embodiment utilizes multi-source data collaboration and the mining of natural tidal channel evolution patterns to construct an adaptive generative model. Users only need to input core control parameters to generate a tidal channel network that conforms to regional geomorphological characteristics and natural development patterns, without relying on measured or remote sensing data. This reduces the empirical bias of manual design and allows for flexible parameter adjustment based on different simulation scenarios, generating customized tidal channel data that adapts to the needs of multi-scenario carbon sequestration assessment and ecological restoration planning.

[0180] (2) Strengthen the scientific rigor and accuracy of organic carbon burial assessment.

[0181] Existing carbon sink simulations generally employ the idealized assumption of uniform tidal flats, neglecting the regulatory role of tidal channels in material migration and energy exchange. This embodiment uses an automatically generated tidal channel network as a key geomorphic unit, deeply integrating it with an organic carbon burial simulation model to establish a linkage mechanism between tidal channel topological features and carbon cycle processes. By reconstructing the regulatory effects of tidal channels on organic carbon generation, migration, and burial, it corrects assessment biases caused by unreasonable geomorphic assumptions, making the organic carbon burial assessment results more consistent with the actual ecological processes of coastal wetlands.

[0182] (3) Reduce the barriers to use and the complexity of operation of carbon sink assessment technology.

[0183] To address the issue that existing technologies require specialized personnel to perform complex operations such as multi-software collaboration, manual data format conversion, and step-by-step parameter adjustments, this embodiment integrates complex processes such as data processing, model invocation, and result output through an integrated workflow design and automated encapsulation. Users only need to input basic parameters through a visual interface to start the entire process, without interfering with underlying technical aspects. This significantly reduces reliance on specialized skills, making it easier for non-specialized personnel to conduct accurate carbon sequestration assessments and adapting to the diverse needs of scientific research and engineering practice.

[0184] (4) Enhance the practical value and transformation efficiency of technological achievements.

[0185] Existing technologies typically output data in raw formats, requiring secondary processing before use, and visualization relies on additional tools. This embodiment optimizes the output process, automatically generating standardized datasets and visual maps that can be directly used in decision-making scenarios such as carbon sequestration accounting and ecological restoration effectiveness assessment without additional processing. Furthermore, the intuitive visualization facilitates rapid analysis of the correlation between tidal channels and carbon burial, improving the efficiency of technological transformation and ease of application.

[0186] (5) Enhance the generalizability and scalability of technical solutions.

[0187] The core parameters and model architecture of this embodiment can be flexibly adjusted according to the geomorphological, hydrological, and vegetation characteristics of different regions, without the need for targeted reconstruction of the technical solution, thus adapting to the carbon sequestration assessment needs of different types of coastal wetlands. The model architecture reserves space for functional expansion, allowing for the subsequent inclusion of more ecological and geomorphological parameters to further enhance the comprehensiveness of the assessment and significantly improve the applicability and scalability of the technical solution.

[0188] Reference Figure 15 This application provides an automatic wetland structure network generation and carbon burial assessment system, comprising:

[0189] The image processing unit is used to generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area;

[0190] The parameter extraction unit is used to extract the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file.

[0191] The training set construction unit is used to construct a standardized training set based on the vector coordinates of the tidal channel, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file.

[0192] The model training unit is used to train an adaptive generative model of the tidal channel network based on the standardized training set.

[0193] The tidal channel generation unit is used to generate the target tidal channel network of the study area based on the trained adaptive generation model of the tidal channel network.

[0194] The geomorphological simulation unit is used to simulate the dynamic geomorphology of the study area using the target tidal channel network.

[0195] A carbon burial assessment unit is used to assess the carbon burial of the study area based on the dynamic geomorphology.

[0196] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0197] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for automatic generation of wetland structure networks and assessment of carbon burial, characterized in that, The method includes the following steps: Generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area; The basic morphological parameters and topological correlation parameters of the tidal channel are extracted from the tidal channel vector coordinate file; A standardized training set is constructed based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file; Train the adaptive generation model of the tidal channel network based on the standardized training set; The target tidal channel network for the study area is generated based on the trained adaptive generation model of the tidal channel network. The dynamic geomorphology of the study area was obtained by simulating the target tidal channel network. Carbon burial in the study area was assessed based on the dynamic geomorphology.

2. The method for automatic generation of wetland structure networks and carbon burial assessment according to claim 1, characterized in that, The process of generating a tidal channel vector coordinate file in SHP format based on satellite remote sensing images of the study area includes the following steps: The satellite remote sensing images of the study area were sequentially subjected to radiometric calibration, FLAASH atmospheric correction, spectral enhancement of tidal channel areas, Gram-Schmidt panchromatic and multispectral band fusion, and topographically constrained geometric fine correction to obtain preprocessed remote sensing images. An initial tidal channel network is extracted from the preprocessed remote sensing image; The initial tidal channel network is subjected to topology verification and visual interpretation optimization to generate a tidal channel vector coordinate file in SHP format.

3. The method for automatic generation of wetland structure networks and carbon sequestration assessment according to claim 1, characterized in that, The extraction of the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file includes the following steps: The length, density, curvature, and dilatation rate of the tidal channel are extracted from the tidal channel vector coordinate file as the basic morphological parameters. The main-branch length ratio, branching angle, and number of branch levels are extracted from the tidal channel vector coordinate file as the topological association parameters.

4. The method for automatic generation of wetland structure networks and carbon sequestration assessment according to claim 1, characterized in that, The process of constructing a standardized training set based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file includes the following steps: An initial training set is constructed based on the tidal channel vector coordinates, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file; The initial training set is subjected to parameter logic verification so that the density and division rate conform to the natural law of fractal dimension. The initial training set is subjected to morphological topology verification to remove abnormal morphological samples with intersections or breaks. For the difficult example samples with fine tributaries and complex branching in the initial training set, data augmentation is performed using a topology-preserving enhancement method involving rotation and scaling. The regional feature normalization method is used to eliminate regional landform differences, and the standardized training set is finally obtained by quantizing and encoding the landform background parameters.

5. The method for automatic generation of wetland structure networks and carbon sequestration assessment according to claim 1, characterized in that, The network adaptive generative model adopts a combined architecture of the SegNeXt spatial generative model and the graph attention network parameter regressor. The SegNeXt spatial generation model is used to optimize the encoding and decoding network structure. By adding a weight module at the main and branch channel levels of the tidal channel, the encoding layer focuses on extracting the long-distance continuity features of the main channel, while the decoding layer captures the details of the branch channel through multi-scale feature aggregation. Simultaneously, a tidal channel fractal window adaptation mechanism is embedded, and the window size is dynamically adjusted according to the fractal dimension to avoid the fractal structure being fragmented. The graph attention network parameter regressor is used to construct the tidal channel parameter graph structure. The main channel parameters are set as core nodes, the tributary channel parameters are set as child nodes, and the landform background is set as edge attributes. The coupling law between key parameters is strengthened through the attention mechanism to achieve adaptive parameter adjustment. The network adaptive generation model incorporates biodynamic geomorphological coupling constraints to achieve two-way linkage between spatial morphology generation and parameter matching, ensuring that the generation results have both spatial rationality and parameter coordination.

6. The method for automatic generation of wetland structure networks and carbon sequestration assessment according to claim 5, characterized in that, The step of training the adaptive generative model of the tidal channel network based on the standardized training set includes the following steps: The first stage involves training to generate the main gully, using the main gully continuity loss as the loss function: ； in, The main sulcus continuity loss value is used to quantify the difference between the predicted main sulcus and the actual main sulcus in terms of continuity and smoothness of direction; N is the number of training samples. Let be the predicted continuous length of the main channel for the i-th sample; λ is the true continuous length of the main groove of the i-th sample; λ is the curvature smoothness constraint weight; smooth( ) is the function for calculating the curvature smoothness of the predicted main gully; Cpred,i is the predicted curvature value of the main gully for the i-th sample; The second stage involves branch generation and dual-architecture fusion training, using a composite loss function: ； in, The total composite loss value; α is the morphological loss weight; Hierarchical FocalTverskyLoss is used to optimize tidal channel morphology details and class imbalance; β is the parameter matching loss weight. γ is the parameter matching loss value, used to quantify the deviation between the predicted parameters and the actual parameters; γ is the topological rationality loss weight. This is the topological rationality loss value, used to constrain the connection relationship between the main and branch ditches; The third stage implements a training optimization strategy. In this stage, the performance is evaluated using a validation set after each round of training. If the loss does not decrease after five consecutive rounds of validation, an early stopping strategy is initiated to prevent overfitting. Throughout the training process, three specific indicators are recorded: the accuracy of main and branch channel connection, the matching rate of branch points, and the parameter deviation rate. Finally, the optimal model weights based on the comprehensive analysis of these specific indicators are saved to obtain the trained adaptive generation model of the tidal channel network.

7. A method for automatic generation of wetland structure networks and carbon sequestration assessment according to any one of claims 1 to 6, characterized in that, The method further includes the following steps: Using Delft3D code as a framework, the following steps are integrated and encapsulated: generating a target tidal channel network for the study area based on the trained adaptive generation model of the tidal channel network; simulating the dynamic geomorphology of the study area using the target tidal channel network; and assessing carbon burial in the study area based on the dynamic geomorphology. This results in an encapsulated module. The user-inputted basic configuration information is entered into the encapsulation module to obtain the carbon burial assessment results for the study area.

8. A system for automatically generating wetland structure networks and assessing carbon sequestration, characterized in that, The system includes: The image processing unit is used to generate SHP format tidal channel vector coordinate files based on satellite remote sensing images of the study area; The parameter extraction unit is used to extract the basic morphological parameters and topological correlation parameters of the tidal channel from the tidal channel vector coordinate file. The training set construction unit is used to construct a standardized training set based on the vector coordinates of the tidal channel, the basic morphological parameters, and the topological correlation parameters in the tidal channel vector coordinate file. The model training unit is used to train an adaptive generative model of the tidal channel network based on the standardized training set. The tidal channel generation unit is used to generate the target tidal channel network of the study area based on the trained adaptive generation model of the tidal channel network. The geomorphological simulation unit is used to simulate the dynamic geomorphology of the study area using the target tidal channel network. A carbon burial assessment unit is used to assess the carbon burial of the study area based on the dynamic geomorphology.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 7.

Images