A mangrove carbon sink prediction method, system, device and medium based on seasonal dynamic hydrograph and niche fusion community embedding

Abstract

The application discloses a mangrove carbon sink prediction method, system, equipment and medium based on seasonal dynamic hydrograph and niche fusion community embedding, comprising the following steps: extracting mangrove patches and obtaining node feature time series; constructing a comprehensive static ecological map according to the spatial distance between each mangrove patch and the habitat similarity; clustering the comprehensive static ecological map to obtain a community membership vector and embedding it into the node feature time series; obtaining a seasonal dynamic hydrological connectivity map according to a historical water body index time series; in the static channel, graph convolution is performed on the enhanced node feature matrix and the comprehensive static ecological map to obtain static features; in the dynamic channel, graph convolution is performed on the enhanced node feature matrix and the target seasonal dynamic hydrological connectivity map; and the static features and the dynamic features are fused by using a fusion layer to predict a mangrove carbon storage prediction value. The application can fuse multiple source environmental factors, automatically adapt to seasonal hydrological changes and reflect the niche overlap characteristics.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, system, device, and medium for predicting mangrove carbon sequestration based on seasonal dynamic hydrological maps and ecological niche integration with community embedding. Background Technology

[0002] Mangroves, as unique ecological barriers transitioning from land to sea, are considered important blue carbon ecosystems for mitigating global climate change due to their exceptional carbon sequestration potential. Dynamic and precise monitoring of mangrove carbon storage and long-term trend prediction have become key technological requirements for quantitatively assessing the effectiveness of mangrove ecological restoration, developing scientific restoration plans, and supporting the conversion of blue carbon value. However, mangrove ecosystems exhibit significant spatial heterogeneity and environmental sensitivity, and their carbon sequestration function is subject to complex coupling effects from vegetation growth, hydrological dynamics, soil physicochemical properties, and extreme weather events (such as typhoons).

[0003] Currently, predictions of mangrove carbon sequestration effectiveness mainly rely on time series models or traditional pixel-based machine learning methods. However, these methods have the following shortcomings: 1) Traditional methods treat each mangrove patch as an independent observation point, ignoring key ecological processes such as nutrient exchange and seed dispersal between patches via tidal hydrodynamics. 2) Existing graph neural network methods typically use fixed adjacency matrices, which cannot reflect the significant monsoon climate characteristics of regions such as Hainan Island. For example, the connectivity between patches differs fundamentally between dry season water flow blockage and typhoon season flooding, leading to prediction bias when using static graphs. 3) Existing ecological zoning is mostly based on hard geographical divisions, ignoring the overlap of community functions in the "ecological ecotone" and failing to effectively utilize community affiliation characteristics to assist model learning.

[0004] Therefore, there is an urgent need for a mangrove carbon sink prediction method that can integrate multiple environmental factors, automatically adapt to seasonal hydrological changes, and reflect the characteristics of overlapping ecological niches. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for predicting mangrove carbon sinks based on seasonal dynamic hydrological maps and ecological niche integration and community embedding, so as to integrate multiple environmental factors, automatically adapt to seasonal hydrological changes, and reflect the characteristics of ecological niche overlap.

[0006] In a first aspect, the present invention provides a method for predicting mangrove carbon sinks based on seasonal dynamic hydrological maps and ecological niche-integrated community embedding, the method comprising: Acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area; Multiple mangrove patches were extracted from remote sensing image data, and the mangrove patches were used as graph nodes to construct the node feature time series corresponding to each graph node. Habitat similarity is obtained from environmental factor data, and a comprehensive static ecological map is constructed based on the spatial distance between mangrove patches and habitat similarity. Clustering of the comprehensive static ecological map yields the community membership vector corresponding to each mangrove patch; and the community membership vector is embedded as a prior feature into the node feature time series to obtain the enhanced node feature matrix. Based on the historical water index time series, the time-lag correlation coefficients of each mangrove patch in different seasonal periods were determined, and a seasonal dynamic hydrological connectivity map was constructed based on the time-lag correlation coefficients. The integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map are input into a pre-trained multi-channel dynamic graph convolutional network model, which includes a static channel, a dynamic channel, and a fusion layer. In the static channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the comprehensive static ecological map as the first adjacency relationship to obtain static features; the target seasonal dynamic hydrological connectivity map is determined according to the current season, and in the dynamic channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship to obtain dynamic features; By using a fusion layer to fuse static and dynamic features, and based on the fused features, the predicted value of mangrove carbon storage at future moments is obtained.

[0007] In some embodiments of the present invention, the multi-channel dynamic graph convolutional network model further includes a temporal convolutional layer, the fusion layer is an attention fusion layer, and the fused features are spatiotemporal features; After obtaining static and dynamic features, the method also includes: Temporal convolutional layers are used to extract temporal dependencies between static and dynamic features; The fusion layer is used to fuse static and dynamic features, including: The attention weights of static and dynamic features are calculated using an attention fusion layer, and the static and dynamic features are then weighted and fused based on these attention weights. By combining the weighted fused features with time dependencies, spatiotemporal features are obtained.

[0008] In some embodiments of the present invention, habitat similarity is obtained from environmental factor data, and a comprehensive static ecological map is constructed based on the spatial distance between mangrove patches and habitat similarity, including: Calculate the Euclidean distance between the centroids of any two mangrove patches, and determine the spatial association weight between the corresponding two mangrove patches based on the Euclidean distance and the Gaussian kernel function to obtain the spatial adjacency matrix; The habitat feature vectors of each mangrove patch are extracted from the environmental factor data, and the cosine similarity between the habitat feature vectors of any two mangrove patches is calculated to obtain the habitat similarity matrix. The spatial adjacency matrix and habitat similarity matrix are normalized respectively, and then the normalized spatial adjacency matrix and the normalized habitat similarity matrix are weighted and fused to obtain a comprehensive static ecological map.

[0009] In some embodiments of the present invention, clustering of the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch includes: The number of potential ecological functional zones is determined, and the community membership vector corresponding to each mangrove patch is initialized. Based on the node connection relationships and edge weights in the comprehensive static ecological graph, a likelihood function of the observation graph is constructed. By maximizing the likelihood function of the observation map, the community membership vector corresponding to each mangrove patch is iteratively updated until the convergence condition is met. The community membership vector that meets the convergence condition is taken as the final community membership vector.

[0010] In some embodiments of the present invention, based on historical water index time series, the time-lag correlation coefficients of each mangrove patch in different seasonal periods are determined, and a seasonal dynamic hydrological connectivity map is constructed based on the time-lag correlation coefficients, including: Extract the normalized water index time series corresponding to each mangrove patch from the historical water index time series; Based on the climate characteristics of the mangrove monitoring area, the normalized water index time series is divided into multiple seasonal periods; Within each seasonal period, the cross-correlation coefficient of the normalized water index time series corresponding to any two mangrove patches is calculated within a preset time lag window, and the maximum cross-correlation coefficient within the preset time lag window is determined as the time lag correlation coefficient. Based on the time-lag correlation coefficients of each mangrove patch in different seasons, hydrological adjacency matrices corresponding to different seasons are constructed to obtain seasonal dynamic hydrological connectivity maps.

[0011] In some embodiments of the present invention, the multiple seasonal periods include the dry season, the rainy season, and the typhoon season, and the hydrological adjacency matrix includes the dry season hydrological adjacency matrix, the rainy season hydrological adjacency matrix, and the typhoon season hydrological adjacency matrix.

[0012] In some embodiments of the present invention, the multi-channel dynamic graph convolutional network model is based on the following loss function. train:

[0013] In the formula, MSE is the mean square error. Attention weight vector The square of the L2 norm, The regularization coefficient is used. For predicted values, This is the label value.

[0014] Secondly, the present invention also provides a mangrove carbon sink prediction system based on seasonal dynamic hydrological maps and ecological niche integration with community embedding, the system comprising: The data acquisition module is used to acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area; The node construction module is used to extract multiple mangrove patches from remote sensing image data and use the mangrove patches as graph nodes to construct the node feature time series corresponding to each graph node. The static ecological map construction module is used to obtain habitat similarity from environmental factor data and construct a comprehensive static ecological map based on the spatial distance between mangrove patches and habitat similarity. The community embedding module is used to cluster the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch; and to embed the community membership vector as a prior feature into the node feature time series to obtain the enhanced node feature matrix. The dynamic hydrological map construction module is used to determine the time-lag correlation coefficient of each mangrove patch in different seasonal periods based on the historical water index time series, and to construct a seasonal dynamic hydrological connectivity map based on the time-lag correlation coefficient. The model input module is used to input the integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map into the pre-trained multi-channel dynamic graph convolutional network model. The multi-channel dynamic graph convolutional network model includes static channels, dynamic channels, and a fusion layer. The multi-channel convolution module is used to perform graph convolution in the static channel, using the enhanced node feature matrix as the node signal and the comprehensive static ecological map as the first adjacency relationship, to obtain static features; and to determine the target seasonal dynamic hydrological connectivity map according to the current season, and to perform graph convolution in the dynamic channel, using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship, to obtain dynamic features. The prediction module is used to fuse static and dynamic features using a fusion layer, and based on the fused features, to predict the mangrove carbon storage value at future times.

[0015] Thirdly, the present invention also provides an electronic device, including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to perform operations in the mangrove carbon sink prediction method based on seasonal dynamic hydrographic maps and niche fusion community embedding provided in the first aspect.

[0016] Fourthly, the present invention also provides a storage medium storing a plurality of instructions adapted for loading by a processor to execute the steps in the mangrove carbon sink prediction method based on seasonal dynamic hydrographic maps and niche fusion community embedding provided in the first aspect.

[0017] In the mangrove carbon sink prediction method, system, device, and medium based on seasonal dynamic hydrological map and niche fusion community embedding provided by this invention, in the static channel, graph convolution is performed using an enhanced node feature matrix as node signal and a comprehensive static ecological map as the first adjacency relationship to obtain static features; the target seasonal dynamic hydrological connectivity map is determined according to the current season, and in the dynamic channel, graph convolution is performed using an enhanced node feature matrix as node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship to obtain dynamic features; the static and dynamic features are fused using a fusion layer, and based on the fused features, the predicted value of mangrove carbon storage at future times is obtained. This enables the multi-channel dynamic graph convolutional network (STGCN) model to perceive the material transport network that is completely different between the dry season and the typhoon season, significantly improving the prediction accuracy under complex monsoon climates. Furthermore, by simultaneously considering information from three dimensions—spatial location, habitat background, and hydrodynamics—the STGCN model avoids neglecting complex but crucial hydrological mechanisms, thus enhancing its generalization ability to sudden environmental events such as red tides and typhoons. In addition, the two-layer cascaded structure of static and dynamic channels ensures the capture of micro-patch and macro-community relationships while significantly reducing computational resource consumption, making it suitable for rapid deployment over large-scale areas. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and ecological niche fusion community embedding provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the mangrove carbon sink prediction system based on seasonal dynamic hydrological maps and ecological niche fusion community embedding provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention; Figure 4This is a schematic diagram of the network structure of the multi-channel dynamic graph convolutional network model provided in the embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.

[0023] The use of "applies to" or "configured to" in this invention implies an open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more conditions or values ​​may in practice be based on additional conditions or values ​​beyond those conditions.

[0024] In this invention, the term "exemplary" is used to mean "serving as an example, illustration, or description." Any embodiment described as "exemplary" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0025] The following description, in conjunction with the accompanying drawings, introduces the mangrove carbon sink prediction method, system, equipment, and medium based on seasonal dynamic hydrological maps and ecological niche fusion community embedding provided by embodiments of the present invention.

[0026] like Figure 1As shown in the figure, this invention provides a method for predicting mangrove carbon sinks based on seasonal dynamic hydrological maps and ecological niche community embedding. The method includes the following steps: S101, acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area.

[0027] S102, extract multiple mangrove patches from remote sensing image data, and use the mangrove patches as graph nodes to construct the node feature time series corresponding to each graph node.

[0028] S103. Habitat similarity is obtained from environmental factor data, and a comprehensive static ecological map is constructed based on the spatial distance between mangrove patches and habitat similarity.

[0029] S104: Cluster the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch; and embed the community membership vector as a prior feature into the node feature time series to obtain the enhanced node feature matrix.

[0030] In some examples, the K-dimensional community membership vector of each patch is used. As static prior features that do not change over time, at each time step, the node feature time series of the patch is compared with the time series of the patch. By concatenating the components, we obtain the enhanced node feature matrix. .

[0031] S105. Based on the historical water index time series, the time-lag correlation coefficients of each mangrove patch in different seasonal periods are determined, and a seasonal dynamic hydrological connectivity map is constructed based on the time-lag correlation coefficients.

[0032] S106 inputs the integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map into the pre-trained multi-channel dynamic graph convolutional network STGCN model.

[0033] Among them, the multi-channel dynamic graph convolutional network model is as follows: Figure 4 As shown, it includes static channels, dynamic channels, and a fusion layer.

[0034] S107. In the static channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the comprehensive static ecological map as the first adjacency relationship to obtain static features. Based on the current season, the target seasonal dynamic hydrological connectivity map is determined, and in the dynamic channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship to obtain dynamic features.

[0035] In some examples, the calendar date information corresponding to the current prediction time step is obtained; based on the climate and phenological characteristics of the mangrove monitoring area, the season to which the current prediction time step belongs is determined from the dry season, rainy season, or typhoon season, and is taken as the current season; from the dry season hydrological adjacency matrix, rainy season hydrological adjacency matrix, and typhoon season hydrological adjacency matrix contained in the seasonal dynamic hydrological connectivity map, the hydrological adjacency matrix corresponding to the current season is selected as the target seasonal dynamic hydrological connectivity map.

[0036] S108 utilizes a fusion layer to fuse static and dynamic features, and based on the fused features, predicts the future carbon storage value of mangroves.

[0037] The mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and niche fusion community embedding provided in this invention obtains static features by performing graph convolution in the static channel, using an enhanced node feature matrix as node signals and a comprehensive static ecological map as the first adjacency relationship. Based on the current season, a target seasonal dynamic hydrological connectivity graph is determined, and in the dynamic channel, graph convolution is performed using an enhanced node feature matrix as node signals and the target seasonal dynamic hydrological connectivity graph as the second adjacency relationship to obtain dynamic features. A fusion layer is used to fuse the static and dynamic features, and based on the fused features, the predicted value of mangrove carbon storage at future times is obtained. This enables the multi-channel dynamic graph convolutional network (STGCN) model to perceive the significantly different material transport networks during the dry season and typhoon season, significantly improving prediction accuracy under complex monsoon climates. Furthermore, by simultaneously considering information from three dimensions—spatial location, habitat background, and hydrodynamics—the multi-channel dynamic graph convolutional network (STGCN) model avoids ignoring complex but crucial hydrological mechanisms, improving its generalization ability to sudden environmental events (such as red tides and typhoons). Furthermore, by using a two-layer cascaded approach of static and dynamic channels, the system can capture the relationship between microscopic patches and macroscopic communities while significantly reducing computational resource consumption, making it suitable for rapid deployment in large-scale areas.

[0038] In some embodiments of the present invention, such as Figure 4 As shown, the multi-channel dynamic graph convolutional network model also includes a temporal convolutional layer, and the fusion layer is an attention fusion layer. The fused features are spatiotemporal features.

[0039] After obtaining static and dynamic features, the method also includes: Temporal convolutional layers are used to extract temporal dependencies between static and dynamic features.

[0040] In some examples, the temporal convolutional layer includes causal dilated convolution, which performs causal dilated convolution on static and dynamic features to extract temporal dependencies.

[0041] The fusion layer is used to fuse static and dynamic features, including: Attention weights for static and dynamic features are calculated using an attention fusion layer. Based on these attention weights, static and dynamic features are weighted and fused. The weighted fused features are then combined with temporal dependencies to obtain spatiotemporal features.

[0042] Specifically, the network structure of the multi-channel dynamic graph convolutional network model includes a sequentially connected input layer, parallel static and dynamic channels, a temporal convolutional layer, an attention fusion layer, and a prediction output layer.

[0043] The input layer receives the enhanced node feature matrix, the integrated static ecological map, and the seasonal dynamic hydrological connectivity map. The static channel is a graph convolutional layer, using the enhanced node feature matrix as the node signal and the spatial adjacency matrix A as the input. geo Habitat similarity matrix A env The first adjacency relationship is used for graph convolution to output static features. The dynamic channel is a graph convolution layer that automatically selects the target seasonal hydrological adjacency matrix based on the season of the current prediction time step. Using the enhanced node feature matrix as the node signal and the target seasonal hydrological adjacency matrix as the second adjacency relationship, graph convolution is performed to output dynamic features. The temporal convolution layer uses causal dilated convolution to convolve both static and dynamic features along the time dimension, extracting temporal dependencies. The attention fusion layer calculates the attention weights for static and dynamic features, performs weighted fusion based on these weights, and combines the weighted fused features with the temporal dependencies to output spatiotemporal features. The prediction output layer is a fully connected layer that outputs the predicted mangrove carbon storage value for future times based on the spatiotemporal features.

[0044] In some embodiments of the present invention, a comprehensive static ecological map is constructed based on the spatial distance between mangrove patches and the similarity of their habitats, including the following sub-steps: S201. Calculate the Euclidean distance between the centroids of any two mangrove patches, and determine the spatial association weights between the corresponding two mangrove patches based on the Euclidean distance and the Gaussian kernel function to obtain the spatial adjacency matrix. .

[0045] S202, extract habitat feature vectors for each mangrove patch from environmental factor data, and calculate the cosine similarity between the habitat feature vectors of any two mangrove patches to obtain the habitat similarity matrix. .

[0046] Among them, the habitat feature vector includes average salinity, elevation, and dominant species type.

[0047] S203, normalize the spatial adjacency matrix and habitat similarity matrix respectively, and then normalize the spatial adjacency matrix. and the normalized habitat similarity matrix Weighted fusion is performed to obtain a comprehensive static ecological map. This comprehensive static ecological map Characterizing the stable structural relationships of mangrove patches in geographic space and ecological niche. Schematic, .

[0048] Understandably, in the integrated static ecological map, nodes represent each specific patch within the mangrove monitoring area. These patches are extracted from remote sensing imagery, and each patch is treated as an independent node. It's an adjacency matrix, defining an adjacency table of the strength of connections between patches. The specific value (a score combining spatial distance and habitat similarity) directly defines the weight of that edge. Edges are based on... The established logical connections, used to guide the propagation of spatial features in the subsequent STGCN model, represent the intensity of niche overlap and potential ecological energy exchange between mangrove patches.

[0049] In some embodiments of the present invention, the BigClam clustering algorithm is used to cluster the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch, including the following sub-steps: S301, set the number of potential ecological functional zones K, and initialize the community membership vector corresponding to each mangrove patch. The elements in the initialized community membership vector represent the probability that the patch belongs to a certain ecological functional zone (such as a high-salinity intertidal zone or an estuarine scour zone).

[0050] S302, based on the node connection relationships and edge weights in the comprehensive static ecological graph, constructs the likelihood function of the observation graph.

[0051] In some examples, the observation graph likelihood function is derived by modifying the weighted integrated static ecological graph Acomb based on the Bernoulli likelihood of the BigClam community discovery model: the continuous edge weight Acomb(i,j) of node pair (i,j) in the integrated static ecological graph Acomb is used as the observation intensity of the edge between nodes i and j, and the community membership vector F of patches i and j is used as the observation intensity. i F j The inner product is modeled by the exponential link function to determine the edge probability, i.e., the edge probability P. ij =1-exp(-F i ·F jT ), F jT F represents j The transpose of ; the logarithmic form of the likelihood function of the observation plot constructed accordingly is: lnL(F)=Σ(i,j)[Acomb(i,j)·ln(1-exp(-F i ·F jT ))-(1-Acomb(i,j))·F i ·F jT ] Among them, F i F j It is a non-negative community membership vector, meaning that none of its components are less than 0.

[0052] S303, by maximizing the likelihood function of the observation map, iteratively updates the community membership vector corresponding to each mangrove patch until the convergence condition is met.

[0053] To illustrate, the community membership vector is iteratively updated using the gradient descent algorithm until convergence.

[0054] S304, the community membership vector that meets the convergence condition is used as the final community membership vector. The final community membership vector This reflects the functional positioning of patches under multiple ecological gradients.

[0055] The mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and ecological niche fusion community embedding provided in this invention runs the BigClam clustering algorithm on a spatially combined habitat fusion map and inputs the output membership vector as a soft feature into the neural network. This not only preserves the fuzzy boundary information of the mangrove ecotone, but also guides the model to better learn the commonalities of growth within the same functional group.

[0056] In some embodiments of the present invention, the time-lag correlation coefficients of each mangrove patch in different seasonal periods are determined based on historical water index time series, and a seasonal dynamic hydrological connectivity map is constructed based on the time-lag correlation coefficients, including the following sub-steps: S401, extract the Normalized Difference Water Index (NDWI) time series corresponding to each mangrove patch from the historical water index time series.

[0057] S402, based on the climate characteristics of the mangrove monitoring area, divides the normalized water index time series into multiple seasonal periods.

[0058] In some examples, multiple seasonal periods include the dry season, the rainy season, and the typhoon season.

[0059] S403, within each seasonal period t, calculate the cross-correlation coefficient of the normalized water index time series corresponding to any two mangrove patches i and j within a preset time lag window τ, and determine the maximum cross-correlation coefficient within the preset time lag window as the time lag correlation coefficient. .

[0060] Indicatively, ,s∈{dry season ,rainy season Typhoon season τ reflects the physical delay in the propagation of tidal currents between patches.

[0061] S404. Based on the time-delay correlation coefficients of each mangrove patch in different seasons, hydrological adjacency matrices corresponding to different seasons are constructed to obtain seasonal dynamic hydrological connectivity maps.

[0062] In some examples, the hydrological adjacency matrix includes the dry season hydrological adjacency matrix. Rainy season hydrological adjacency matrix Hydrological adjacency matrix during typhoon season .

[0063] In some embodiments of the present invention, the multi-channel dynamic graph convolutional network model is based on the following loss function. train:

[0064] In the formula, MSE is the mean square error. Attention weight vector The square of the L2 norm, The regularization coefficient is used. This is the predicted value (i.e., the predicted mangrove carbon storage at future moments output by the multi-channel dynamic graph convolutional network model). The label value (i.e., the actual carbon storage observation value in the training data).

[0065] Specifically, the attention weight vector α contains channel weight components corresponding to the three mechanisms of spatial proximity, habitat similarity in the static channel, and hydrological connectivity in the dynamic channel, respectively, i.e., α = [α geo ,α env ,α hydro ], α geo Corresponding to the spatial adjacency matrix A geo The spatial proximity mechanism represented by α env Corresponding to habitat similarity matrix A env The habitat similarity mechanism represented by α hydro Corresponding to the seasonal hydrological adjacency matrix A tide The hydrological connectivity mechanism is characterized.

[0066] The attention weight vector α is obtained by adaptive calculation based on static and dynamic features by the attention fusion layer mentioned above. α represents the contribution ratio of the above three mechanisms to the predicted value of mangrove carbon storage.

[0067] loss function The regularization term λ·‖α‖2² is the L2 regularization term for the attention weight vector α. It suppresses the overly sharp distribution of the attention weight vector α and prevents the attention weights from collapsing to a single mechanism. This forces the multi-channel dynamic graph convolutional network model to make balanced use of the three mechanisms of spatial proximity, habitat similarity and hydrological connectivity during training, thereby preventing the multi-channel dynamic graph convolutional network model from over-relying on a single static graph structure and improving the robustness of the model under extreme hydrological conditions.

[0068] In some embodiments of the present invention, the training process of a multi-channel dynamic graph convolutional network model includes the following steps: S501. Acquire remote sensing image training data and environmental factor training data, extract multiple mangrove patches from the remote sensing image training data, and use the mangrove patches as graph nodes to construct the node feature time series corresponding to each graph node.

[0069] S502, based on the spatial distance and habitat similarity between mangrove patches, a comprehensive static ecological map is constructed.

[0070] S503 clusters the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch; and embeds the community membership vector as a prior feature into the node feature time series to obtain the enhanced node feature matrix.

[0071] S504. Based on the historical water index time series, the time-lag correlation coefficients of each mangrove patch in different seasonal periods are determined, and a seasonal dynamic hydrological connectivity map is constructed based on the time-lag correlation coefficients.

[0072] S505, build a multi-channel dynamic graph convolutional network STGCN model.

[0073] S506 trains the STGCN model based on a comprehensive static ecological map, a seasonal dynamic hydrological connectivity map, and an enhanced node feature matrix. It then automatically learns the weights of different channels in the STGCN model through an attention regularization mechanism until the model converges, thus obtaining the trained STGCN model.

[0074] like Figure 2 As shown, the present invention also provides a mangrove carbon sink prediction system based on seasonal dynamic hydrological map and ecological niche fusion community embedding. The system includes a data acquisition module 201, a node construction module 202, a static ecological map construction module 203, a community embedding module 204, a dynamic hydrological map construction module 205, a model input module 206, a multi-channel convolution module 207, and a prediction module 208.

[0075] The data acquisition module 201 is used to acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area.

[0076] The node construction module 202 is used to extract multiple mangrove patches from remote sensing image data and use the mangrove patches as graph nodes to construct the node feature time series corresponding to each graph node.

[0077] The static ecological map construction module 203 is used to obtain habitat similarity from environmental factor data and construct a comprehensive static ecological map based on the spatial distance between mangrove patches and habitat similarity.

[0078] The community embedding module 204 is used to cluster the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch; and to embed the community membership vector as a prior feature into the node feature time series to obtain the enhanced node feature matrix.

[0079] The dynamic hydrological map construction module 205 is used to determine the time-lag correlation coefficient of each mangrove patch in different seasonal periods based on the historical water index time series, and to construct a seasonal dynamic hydrological connectivity map based on the time-lag correlation coefficient.

[0080] The model input module 206 is used to input the integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map into the pre-trained multi-channel dynamic graph convolutional network model. The multi-channel dynamic graph convolutional network model includes static channels, dynamic channels, and a fusion layer.

[0081] The multi-channel convolution module 207 is used to perform graph convolution in the static channel, using the enhanced node feature matrix as the node signal and the comprehensive static ecological map as the first adjacency relationship, to obtain static features; and to determine the target seasonal dynamic hydrological connectivity map according to the current season, and to perform graph convolution in the dynamic channel, using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship, to obtain dynamic features.

[0082] The prediction module 208 is used to fuse static and dynamic features using a fusion layer, and based on the fused features, to predict the mangrove carbon storage value at future times.

[0083] The mangrove carbon sink prediction system based on seasonal dynamic hydrological maps and niche-integrated community embedding provided in this embodiment corresponds to the mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and niche-integrated community embedding provided in any of the above embodiments, and will not be described again here.

[0084] Based on any of the above embodiments, another embodiment of this application also provides an electronic device, such as... Figure 3As shown, the electronic device may include a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute the aforementioned mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and niche-integrated community embedding.

[0085] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] On the other hand, embodiments of this application also provide a storage medium storing multiple instructions adapted for loading by a processor to execute the mangrove carbon sink prediction method based on seasonal dynamic hydrological maps and niche fusion community embedding as provided in the above embodiments.

[0087] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0089] The foregoing has provided a detailed description of a mangrove carbon sequestration prediction method, system, device, and medium based on seasonal dynamic hydrological maps and ecological niche fusion community embedding provided by embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting mangrove carbon sequestration based on seasonal dynamic hydrological maps and ecological niche community embedding, characterized in that, The method includes: Acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area; Multiple mangrove patches are extracted from the remote sensing image data, and the mangrove patches are used as graph nodes to construct the node feature time series corresponding to each graph node. Habitat similarity is obtained from the environmental factor data, and a comprehensive static ecological map is constructed based on the spatial distance between each mangrove patch and the habitat similarity. Clustering of the comprehensive static ecological map yields a community membership vector for each mangrove patch; the community membership vector is then embedded as a prior feature into the node feature time series to obtain an enhanced node feature matrix. Based on the historical water index time series, the time lag correlation coefficients of each mangrove patch in different seasonal periods are determined, and a seasonal dynamic hydrological connectivity map is constructed based on the time lag correlation coefficients. The integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map are input into a pre-trained multi-channel dynamic graph convolutional network model, which includes a static channel, a dynamic channel, and a fusion layer. In the static channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the integrated static ecological map as the first adjacency relationship to obtain static features; the target seasonal dynamic hydrological connectivity map is determined according to the current season, and in the dynamic channel, graph convolution is performed using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship to obtain dynamic features; The static and dynamic features are fused using the fusion layer, and the predicted value of mangrove carbon storage at future times is obtained based on the fused features.

2. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and ecological niche fusion community embedding as described in claim 1, characterized in that, The multi-channel dynamic graph convolutional network model also includes a temporal convolutional layer, the fusion layer is an attention fusion layer, and the fused features are spatiotemporal features; After obtaining the static features and the dynamic features, the method further includes: The temporal convolutional layer is used to extract the temporal dependencies between the static and dynamic features; The process of fusing the static features and the dynamic features using the fusion layer includes: The attention fusion layer is used to calculate the attention weights of the static features and the dynamic features, and the static features and the dynamic features are weighted and fused based on the attention weights. The weighted and fused features are combined with the time dependency relationship to obtain the spatiotemporal features.

3. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and ecological niche community embedding as described in claim 1, characterized in that, The step of obtaining habitat similarity from the environmental factor data and constructing a comprehensive static ecological map based on the spatial distance and habitat similarity between the mangrove patches includes: Calculate the Euclidean distance between the centroids of any two mangrove patches, and determine the spatial association weights between the corresponding two mangrove patches based on the Euclidean distance and the Gaussian kernel function to obtain the spatial adjacency matrix; The habitat feature vectors of each mangrove patch are extracted from the environmental factor data, and the cosine similarity between any two mangrove patch habitat feature vectors is calculated to obtain a habitat similarity matrix. The spatial adjacency matrix and the habitat similarity matrix are normalized respectively, and the normalized spatial adjacency matrix and the normalized habitat similarity matrix are weighted and fused to obtain the comprehensive static ecological map.

4. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and ecological niche fusion community embedding as described in claim 1, characterized in that, The clustering of the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch includes: The number of potential ecological functional zones is set, and the community membership vector corresponding to each mangrove patch is initialized; Based on the node connection relationships and edge weights in the comprehensive static ecological graph, an observation graph likelihood function is constructed; By maximizing the likelihood function of the observation map, the community membership vector corresponding to each mangrove patch is iteratively updated until the convergence condition is met. The community membership vector that meets the convergence condition is taken as the final community membership vector.

5. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and ecological niche community embedding as described in claim 1, characterized in that, The step of determining the time-lag correlation coefficient of each mangrove patch in different seasonal periods based on the historical water index time series, and constructing a seasonal dynamic hydrological connectivity map based on the time-lag correlation coefficient, includes: Extract the normalized water index time series corresponding to each mangrove patch from the historical water index time series; Based on the climate characteristics of the mangrove monitoring area, the normalized water index time series is divided into multiple seasonal periods; Within each of the aforementioned seasonal periods, the cross-correlation coefficient of the normalized water index time series corresponding to any two mangrove patches within a preset time lag window is calculated, and the maximum cross-correlation coefficient within the preset time lag window is determined as the time lag correlation coefficient. Based on the time-lag correlation coefficients of each mangrove patch in different seasonal periods, hydrological adjacency matrices corresponding to different seasonal periods are constructed to obtain the seasonal dynamic hydrological connectivity map.

6. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and ecological niche fusion community embedding as described in claim 5, characterized in that, The multiple seasonal periods include the dry season, the rainy season, and the typhoon season, and the hydrological adjacency matrix includes the dry season hydrological adjacency matrix, the rainy season hydrological adjacency matrix, and the typhoon season hydrological adjacency matrix.

7. The mangrove carbon sequestration prediction method based on seasonal dynamic hydrological maps and niche-embedded community integration according to any one of claims 1 to 6, characterized in that, The multi-channel dynamic graph convolutional network model is based on the following loss function. train: In the formula, MSE is the mean square error. Attention weight vector The square of the L2 norm, The regularization coefficient is used. For predicted values, This is the label value.

8. A mangrove carbon sequestration prediction system based on seasonal dynamic hydrological maps and ecological niche integration with community embedding, characterized in that, The system includes: The data acquisition module is used to acquire remote sensing image data, environmental factor data, and historical water index time series of the mangrove monitoring area; The node construction module is used to extract multiple mangrove patches from the remote sensing image data, and use the mangrove patches as graph nodes to construct the node feature time series corresponding to each graph node. A static ecological map construction module is used to obtain habitat similarity from the environmental factor data and construct a comprehensive static ecological map based on the spatial distance between each mangrove patch and the habitat similarity. The community embedding module is used to cluster the comprehensive static ecological map to obtain the community membership vector corresponding to each mangrove patch; and to embed the community membership vector as a prior feature into the node feature time series to obtain an enhanced node feature matrix. The dynamic hydrological map construction module is used to determine the time-lag correlation coefficient of each mangrove patch in different seasonal periods based on the historical water index time series, and to construct a seasonal dynamic hydrological connectivity map based on the time-lag correlation coefficient. The model input module is used to input the integrated static ecological map, the enhanced node feature matrix, and the seasonal dynamic hydrological connectivity map into a pre-trained multi-channel dynamic graph convolutional network model, which includes a static channel, a dynamic channel, and a fusion layer. A multi-channel convolution module is used to perform graph convolution in the static channel using the enhanced node feature matrix as the node signal and the integrated static ecological map as the first adjacency relationship to obtain static features; and to determine the target seasonal dynamic hydrological connectivity map according to the current season, and to perform graph convolution in the dynamic channel using the enhanced node feature matrix as the node signal and the target seasonal dynamic hydrological connectivity map as the second adjacency relationship to obtain dynamic features. The prediction module is used to fuse the static features and the dynamic features using the fusion layer, and based on the fused features, to predict the mangrove carbon storage value at future times.

9. An electronic device, characterized in that, It includes a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to perform the steps in the mangrove carbon sink prediction method based on seasonal dynamic hydrographic maps and niche-integrated community embedding as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores multiple instructions adapted for loading by a processor to execute the steps in the mangrove carbon sink prediction method based on seasonal dynamic hydrographic maps and niche-embedded community integration as described in any one of claims 1 to 7.

Images