A wetland ecosystem carbon sink data monitoring system and method

Abstract

This invention discloses a wetland ecosystem carbon sink data monitoring system and method, belonging to the field of data monitoring technology. The system includes an intelligent sensing node network, where intelligent sensing nodes collect soil parameters, vegetation parameters, and water parameters to obtain initial sampling data; a sampling and transmission optimization module, which identifies spatial variation regions based on a prior map of wetland spatial heterogeneity and generates optimized sampling data; a multi-scale data fusion module, which generates spatial representative weights and generates carbon sink parameter distribution data based on these weights; a carbon sink assessment module, which generates carbon sink assessment results based on the carbon sink parameter distribution data and spatial representative weights; and a system optimization module, which receives the carbon sink assessment results and updates the prior map of wetland spatial heterogeneity. This invention improves the spatial interpolation accuracy of extrapolating continuous carbon sink distribution fields from discrete data by proposing a node spatial representative weight calculation method based on feature similarity and deviation.

Description

Technical Field

[0001] This invention relates to the field of data monitoring technology, and in particular to a data monitoring system and method for carbon sinks in wetland ecosystems. Background Technology

[0002] Wetland ecosystems refer to areas where the surface is excessively wet or perennially waterlogged, and where wetland plants grow. Their soils are chronically anaerobic, making them among the most productive ecosystems on Earth. They possess a powerful carbon sequestration capacity and play a crucial "carbon sink" role in the global carbon cycle and in addressing climate change. With the advancement of "dual carbon" goals, high-precision, dynamic monitoring and quantitative assessment of wetland carbon sinks have become a core requirement for ecological research and environmental management.

[0003] Existing methods for monitoring wetland carbon sinks primarily rely on field surveys of fixed plots or model inversion based on remote sensing imagery. Field surveys involve establishing fixed plots in key areas and periodically sampling soil and measuring vegetation biomass to obtain local carbon flux or storage data. Remote sensing inversion, on the other hand, utilizes large-scale surface information (such as vegetation indices and surface temperature) obtained by satellites or drones, combined with empirical or semi-empirical models, to extrapolate regional-scale carbon sink conditions. These methods constitute the fundamental data sources and technical means for current carbon sink assessments.

[0004] However, existing technologies have two significant drawbacks: First, field monitoring methods often use fixed sampling frequencies, which cannot intelligently identify and focus on "hotspot areas" where carbon sink characteristics change drastically in wetland environments. This leads to unreasonable allocation of monitoring resources, and when faced with high spatial heterogeneity in wetlands, single-point observations are insufficient to represent large-scale area information. Second, traditional monitoring and assessment processes are in an open-loop state, i.e., data collection, carbon sink calculation, and result output. They lack the ability to dynamically optimize the monitoring network itself and cannot automatically adjust sampling strategies or update the understanding of wetland heterogeneity based on assessment results, thus limiting the accuracy and efficiency of long-term monitoring.

[0005] Therefore, there is an urgent need to provide a wetland ecosystem carbon sink data monitoring system and method to solve the above problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a wetland ecosystem carbon sink data monitoring system and method.

[0007] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: to provide a wetland ecosystem carbon sink data monitoring system, comprising: The intelligent sensing node network collects soil parameters, vegetation parameters, and water parameters, and, based on a pre-constructed wetland spatial heterogeneity prior map, executes node deployment and sampling strategy configuration to obtain initial sampling data. The sampling and transmission optimization module identifies spatial variation regions based on the prior map of wetland spatial heterogeneity, adjusts the sampling frequency of the intelligent sensing nodes within the spatial variation regions, and generates optimized sampling data. The multi-scale data fusion module acquires remote sensing data of the area where the intelligent sensing node is located, performs spatial alignment and feature fusion on the spatiotemporally optimized sampling data and the remote sensing data, generates spatial representative weights based on the fusion results, and generates carbon sink parameter distribution data based on the spatial representative weights. The carbon sink assessment module calculates the total amount and spatial distribution of carbon in the wetland ecosystem based on the carbon sink parameter distribution data and the spatial representative weights, and performs uncertainty quantification on the calculation results based on the spatial representative weights and the remote sensing data to generate a carbon sink assessment result with uncertainty information. The system optimization module is used to receive the carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, generate node adjustment strategies and remote sensing revisit instructions, feed back the node adjustment strategies to the intelligent sensing node network to update its deployment and sampling strategies, and update the wetland spatial heterogeneity prior map based on the carbon sink assessment results.

[0008] The present invention is further configured such that: the intelligent sensing node network is deployed in the target wetland ecosystem; The prior map of wetland spatial heterogeneity is generated in advance by fusing historical remote sensing data, topographic data, and wetland type data; The initial sampling data generation steps are as follows: the intelligent sensing nodes in the intelligent sensing node network are deployed in the target wetland ecosystem according to the node deployment locations determined by the wetland spatial heterogeneity prior map, and based on the sampling strategy corresponding to the wetland spatial heterogeneity prior map, the soil parameters, the vegetation parameters, and the water parameters are subjected to time-synchronous collection, outlier removal, and sequential rearrangement processing to form a standardized sampling data sequence that corresponds one-to-one with the area where the intelligent sensing node is located, thereby generating the initial sampling data.

[0009] The present invention is further configured such that the method for identifying the spatial variation region in the sampling and transmission optimization module is: S1. Based on the standardized sampling data sequence in the initial sampling data, the standardized sampling data sequence is spatially grouped according to the region where the intelligent sensing node is located, and the variation amplitude and direction of the soil parameters, vegetation parameters and water parameters are extracted in each spatial group to generate a regional change feature sequence. S2. Based on the regional change feature sequence, calculate the consistency index of each pair of adjacent intelligent sensing nodes in each spatial group in terms of the change amplitude and the change direction, and calculate the change difference value representing the difference between nodes according to the consistency index, and generate a node difference association set. S3. Based on the node difference association set, perform continuous filtering on the change difference values ​​between the adjacent smart sensing nodes, and mark the set of smart sensing nodes whose change difference values ​​are continuously greater than the difference threshold within a preset continuous spatial range as a candidate mutation node set. S4. Based on the candidate mutated node set, calculate the temporal fluctuation consistency index of the regional change feature sequence of each intelligent sensing node in the set within multiple consecutive sampling periods. Mark the nodes whose temporal fluctuation consistency index exceeds a preset threshold as temporally consistent nodes, and determine the intelligent sensing nodes in the candidate mutated node set that simultaneously meet the conditions of spatial continuity and temporally consistent nodes as the spatial mutated region.

[0010] The present invention is further configured such that the step of generating the optimized sampling data is as follows: S5. Based on the spatial variation region, calculate the variation intensity of the regional change feature sequence of the intelligent sensing node in the spatial variation region, generate a variation intensity value by combining the change amplitude and the change direction in the regional change feature sequence according to a preset combination rule, and classify and label the intelligent sensing node based on the variation intensity value to form a hierarchical set of variation nodes. S6. Based on the hierarchical set of mutated nodes, a corresponding sampling frequency adjustment strategy is matched for each of the intelligent sensing nodes at different levels. The sampling frequency adjustment strategy is dynamically corrected in combination with the temporal fluctuation consistency index of the spatial mutated region. The dynamically corrected sampling frequency adjustment strategy is then applied to the intelligent sensing node to generate optimized sampling data corresponding to the spatial mutated region.

[0011] The present invention is further configured such that: in the multi-scale data fusion module, the remote sensing data of the area where the intelligent sensing node is located is obtained by performing spatial positioning retrieval in a preset remote sensing data source based on the geographical coordinates of the area where the intelligent sensing node is located; and the remote sensing image data is subjected to time matching filtering and spatial resolution unification processing to generate standardized remote sensing data that corresponds one-to-one with the area where the intelligent sensing node is located and the spatiotemporal optimized sampling data. The multi-scale data fusion module performs spatial alignment and feature fusion on the spatiotemporally optimized sampling data and the remote sensing data. Specifically, it includes: matching the coordinate system and spatial location of the two types of data based on the geographic coordinate information contained in the standardized remote sensing data and the location information of the intelligent sensing nodes corresponding to the spatiotemporally optimized sampling data; extracting time-series statistical features related to carbon sinks from the spatiotemporally optimized sampling data within the same matched area; and extracting vegetation index features and surface temperature features corresponding to the area where the intelligent sensing nodes are located from the standardized remote sensing data. Then, it constructs a fused feature vector based on the time-series statistical features, the vegetation index features, and the surface temperature features, and performs consistency verification and difference correction processing on the fused feature vector to generate a fused feature set.

[0012] The present invention is further configured such that the method for calculating the spatial representative weight is as follows: T1. Based on the fused feature set, the fused feature set is spatially partitioned according to the region where the intelligent sensing node is located, and the feature vector set corresponding to each intelligent sensing node is extracted in each spatial partition. Based on the feature vector set, the feature similarity between any two intelligent sensing nodes in the partition is calculated to generate a node similarity set. T2. Based on the node similarity set, calculate the average feature similarity between each intelligent sensing node and all other nodes in the spatial partition, and use the average feature similarity as the feature distribution consistency index of the node. Nodes with the feature distribution consistency index higher than a preset consistency threshold are marked as high consistency nodes, and nodes with the consistency index lower than the consistency threshold are marked as low consistency nodes, thus generating a node consistency classification set. T3. Based on the node consistency classification set, for the high consistency node and the low consistency node, calculate the Euclidean distance between the feature vector of the node and the regional feature center of the spatial partition, and define the Euclidean distance as the feature deviation degree of the high consistency node or the low consistency node; then, by calculating the ratio of the feature distribution consistency index to the feature deviation degree, generate an initial representative evaluation value for each intelligent sensing node. T4. Based on the initial representative evaluation values ​​of all the intelligent sensing nodes, perform normalization processing on the nodes within the same spatial partition, and use the normalized results as the spatial representative weights of each node.

[0013] The present invention is further configured such that the method for generating the carbon sink parameter distribution data is as follows: R1. Based on the spatial representative weight, perform weight mapping processing on the feature vectors in the fusion feature set corresponding to each intelligent sensing node in the fusion feature set, adjust the contribution ratio of the fusion feature vectors, obtain a weighted fusion feature set, and maintain the correspondence between the intelligent sensing node and the spatial partition in the weighted fusion feature set. R2. Based on the weighted fusion feature set, perform partition aggregation processing with weight constraints in each spatial partition according to the spatial representative weight to generate regional carbon sink feature representations for each spatial partition, and retain the contribution ratio relationship of the spatial representative weight in the regional carbon sink feature representations. R3. Based on the regional carbon sink feature representation, perform a continuity check on the regional carbon sink feature representation between adjacent spatial partitions based on the consistency of feature change direction and change magnitude, connect the regional carbon sink feature representations that satisfy the continuity constraint, and perform a transition construction process based on spatial adjacency relationship on the regional carbon sink feature representations that do not satisfy the continuity constraint to generate a spatial continuous carbon sink feature field. R4. Based on the spatial continuous carbon sink feature field, perform numerical mapping processing based on spatial location according to the correspondence between the spatial continuous carbon sink feature field and spatial location to generate carbon sink parameter distribution data covering the area where the intelligent sensing node is located, and maintain the mapping relationship between the spatial representative weight and the contribution of the carbon sink parameter in the carbon sink parameter distribution data.

[0014] The present invention is further configured such that the total carbon content and spatial distribution of the wetland ecosystem are generated through the following steps: U1. Based on the carbon sink parameter distribution data, perform cumulative calculations based on spatial grid division on the carbon sink parameter distribution data in each spatial partition to generate the partition carbon sink amount of each spatial partition. U2. Perform a summation calculation on the carbon sink of all the spatial partitions to generate the total carbon sink of the wetland ecosystem; U3. Based on the carbon sink amount of each spatial partition and its corresponding spatial location information, perform spatial interpolation and rendering processing to generate a spatial distribution map of carbon sink in the wetland ecosystem that characterizes the spatial variation gradient of carbon sink amount. The uncertainty quantification generates carbon sink assessment results with accompanying uncertainty information through the following steps: V1. Based on the spatial representative weight, calculate the local variance of the spatial representative weight within each spatial partition, and use the local variance as a first type of uncertainty index caused by the difference in node representativeness. V2. Based on the standardized remote sensing data, extract the pixel quality score and atmospheric correction residual of the remote sensing image corresponding to each spatial partition, and fuse the pixel quality score and atmospheric correction residual to generate a second type of uncertainty index caused by the quality of remote sensing data. V3. The first type of uncertainty index and the second type of uncertainty index are weighted and fused to generate a comprehensive uncertainty value for each spatial partition. Based on the comprehensive uncertainty value of all partitions, an uncertainty distribution field consistent with the spatial range of the carbon sink parameter distribution data is generated by spatial interpolation. V4. The total carbon content of the wetland ecosystem, the spatial distribution map of the carbon sink of the wetland ecosystem, and the uncertain distribution field are associated and encapsulated to generate the carbon sink assessment result with the attached uncertain information.

[0015] The present invention is further configured such that: the node adjustment strategy in the system optimization module generates a node addition / reduction adjustment and sampling strategy update scheme based on the abnormal areas and uncertainty exceeding the limit areas identified by the carbon sink assessment results; The remote sensing revisit instruction generates a remote sensing data update scheduling scheme based on the abnormal region. The node adjustment strategy and the remote sensing revisit command work together on the intelligent sensing node network and the wetland spatial heterogeneity prior map.

[0016] A method for monitoring carbon sink data in wetland ecosystems includes the following steps: W1. Deploy a network of intelligent sensing nodes in the target wetland ecosystem, and configure node deployment and sampling strategies based on the prior map of wetland spatial heterogeneity to collect soil parameters, vegetation parameters and water parameters to generate initial sampling data. W2. Based on the prior map of wetland spatial heterogeneity and the initial sampling data, identify spatial variation regions, and adjust the sampling frequency of intelligent sensing nodes in the region to generate optimized sampling data; W3. Obtain remote sensing data of the area where the intelligent sensing node is located, and spatially align and fuse it with the optimized sampling data to generate a fused feature set; W4. Calculate the spatial representative weight of each of the intelligent sensing nodes based on the fused feature set, and use the spatial representative weight to generate carbon sink parameter distribution data; W5. Calculate the total carbon content and spatial distribution of the wetland ecosystem based on the carbon sink parameter distribution data, and quantify the uncertainty of the calculation results by combining the spatial representative weights and remote sensing data to generate carbon sink assessment results. W6. Analyze the carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, generate node adjustment strategies and remote sensing revisit instructions to update the deployment of the intelligent sensing node network, sampling strategies and the prior map of wetland spatial heterogeneity, and realize closed-loop optimization of monitoring.

[0017] The beneficial effects of this invention are as follows: 1. This invention constructs a priori maps of wetland spatial heterogeneity and dynamically identifies spatially variable areas by combining real-time monitoring data, thereby enabling intelligent deployment of monitoring nodes and differentiated adjustment of sampling frequencies. This design solves the problem that traditional fixed sampling cannot focus on key areas of change, and significantly improves the utilization efficiency of monitoring resources and the targeting of data acquisition. 2. This invention proposes a node spatial representative weight calculation method based on feature similarity and deviation by integrating ground-based sensor data and remote sensing data. This algorithm effectively solves the representativeness bias problem caused by "using points to represent areas" in wetland carbon sink monitoring, and significantly improves the spatial interpolation accuracy and the reliability of the evaluation results in extrapolating the continuous carbon sink distribution field from discrete data. 3. This invention constructs a complete closed loop from data acquisition, fusion evaluation to system optimization, which can automatically generate node adjustment strategies and update prior maps based on the abnormal situations and uncertainties in carbon sink assessment. This mechanism breaks through the limitations of traditional open-loop monitoring, endows the system with the ability to learn and adapt dynamically, and realizes the continuous improvement of the long-term accuracy and intelligence level of the wetland carbon sink monitoring system. Attached Figure Description

[0018] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0020] Please see Figure 1 - Figure 2 A wetland ecosystem carbon sequestration data monitoring system, comprising: The intelligent sensing node network collects soil parameters, vegetation parameters, and water parameters, and executes node deployment and sampling strategy configuration based on a pre-constructed wetland spatial heterogeneity prior map to obtain initial sampling data. The sampling and transmission optimization module is connected to the intelligent sensing node network. Based on the prior map of wetland spatial heterogeneity, it identifies spatial variation regions, adjusts the sampling frequency of intelligent sensing nodes within the spatial variation regions, and generates optimized sampling data. The multi-scale data fusion module is connected to the adaptive sampling and transmission optimization module to acquire remote sensing data of the area where the intelligent sensing node is located. It performs spatial alignment and feature fusion on the spatiotemporally optimized sampling data and remote sensing data, generates spatial representative weights based on the fusion results, and generates carbon sink parameter distribution data based on the spatial representative weights. The carbon sink assessment module is connected to the multi-scale data fusion and representativeness calibration module. Based on the carbon sink parameter distribution data and spatial representative weights, it calculates the total amount and spatial distribution of carbon in the wetland ecosystem. Based on the spatial representative weights and remote sensing data, it quantifies the uncertainty of the calculation results and generates carbon sink assessment results with uncertainty information. The system optimization module, connected to the carbon sink assessment and uncertainty quantification module, is used to receive carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, and generate node adjustment strategies and remote sensing revisit instructions. The node adjustment strategies are fed back to the intelligent sensing node network to update its deployment and sampling strategies. At the same time, the prior map of wetland spatial heterogeneity is updated based on the carbon sink assessment results, realizing closed-loop adaptive optimization of the monitoring system.

[0021] Among them, the intelligent sensing node network is deployed in the target wetland ecosystem; The prior map of wetland spatial heterogeneity is generated in advance by integrating historical remote sensing data, topographic data and wetland type data; The specific steps for generating a priori maps of wetland spatial heterogeneity are as follows: First, acquire historical multi-period multispectral remote sensing images, digital elevation model (DEM) data, and wetland type vector data of the target wetland area. Perform atmospheric and geometrical fine-correction on the historical multispectral remote sensing images to generate standardized historical images. Then, extract normalized vegetation index (NVI), normalized water index (NDI), and surface temperature (STI) from the standardized historical images as historical remote sensing feature layers. Next, spatially register and resample the historical remote sensing feature layers, DEM data, and wetland type vector data in a unified geographic coordinate system to generate spatially aligned multi-source data cubes. Finally, utilize a pre-defined deep learning algorithm... The encoder extracts and reduces the dimensionality of multi-source data cubes to generate a high-dimensional latent feature tensor. The high-dimensional latent feature tensor is then input into a trained spatial heterogeneity prediction model. This model is trained on a supervised training dataset constructed based on historical remote sensing feature layers and measured carbon sink sample data. The model learns the nonlinear mapping relationship between features and carbon sink potential by fusing convolution and attention mechanisms, and outputs a predicted value of carbon sink heterogeneity intensity for each pixel. The predicted values ​​of carbon sink heterogeneity intensity are then spatially smoothed and classified to finally generate a wetland spatial heterogeneity prior map that represents the spatial differentiation pattern of carbon sinks in grayscale or levels.

[0022] The intelligent sensing node network specifically includes multiple intelligent sensing nodes with wireless communication and edge computing capabilities, one or more aggregation gateways, and a network management platform. Each intelligent sensing node integrates a multi-parameter soil sensor, a vegetation canopy analyzer, a miniature water quality spectrometer, a positioning module, a microprocessor, and a wireless transmission module. The multi-parameter soil sensor measures soil temperature, humidity, conductivity, and organic carbon content; the vegetation canopy analyzer acquires leaf area index and canopy spectral reflectance; and the miniature water quality spectrometer detects dissolved oxygen, chemical oxygen demand, and chlorophyll concentration in water. The aggregation gateway receives data collected from multiple intelligent sensing nodes and has protocol conversion and data caching functions. The network management platform monitors the status of the intelligent sensing node network, configures sampling tasks, and receives the aggregated data.

[0023] Soil parameters specifically include soil temperature, soil volumetric water content, soil electrical conductivity, and soil organic carbon content; vegetation parameters specifically include leaf area index, normalized difference vegetation index, photochemical reflectance index, and chlorophyll fluorescence intensity; water parameters specifically include water temperature, pH value, dissolved oxygen concentration, chemical oxygen demand, chlorophyll a concentration, and turbidity.

[0024] The initial sampling data generation steps are as follows: the intelligent sensing nodes in the intelligent sensing node network are deployed in the target wetland ecosystem according to the node deployment locations determined by the wetland spatial heterogeneity prior map, and based on the sampling strategy corresponding to the wetland spatial heterogeneity prior map, the soil parameters, vegetation parameters and water parameters are collected synchronously over time, outlier is removed and the order is rearranged to form a standardized sampling data sequence that corresponds one-to-one with the area where the intelligent sensing nodes are located, thereby generating the initial sampling data.

[0025] One embodiment of the present invention is that the method for identifying spatially variable regions in the sampling and transmission optimization module is as follows: S1. Based on the standardized sampling data sequence in the initial sampling data, the standardized sampling data sequence is spatially grouped according to the area where the intelligent sensing node is located, and the variation amplitude and direction of soil parameters, vegetation parameters and water parameters are extracted in each spatial group to generate a regional change feature sequence. Specifically, the method for extracting the variation range and direction of soil parameters, vegetation parameters, and water parameters is as follows: For the time series data corresponding to each intelligent sensing node in the standardized sampling data sequence, calculate the difference between the values ​​of each parameter between two adjacent sampling times, and define the absolute value of the difference as the variation range of the parameter during this time period; at the same time, construct a multi-dimensional feature space with each parameter as the coordinate axis, represent the state of the intelligent sensing node at adjacent times as two points in this space, calculate the displacement vector from the previous time point to the current time point, and define the unit vector of the displacement vector as the comprehensive state change direction of the node within this time window.

[0026] S2. Based on the regional change feature sequence, calculate the consistency index of each pair of adjacent intelligent sensing nodes in each spatial group in terms of change magnitude and change direction, and calculate the change difference value representing the difference between nodes according to the consistency index, and generate a node difference association set. Specifically, the consistency index of each pair of adjacent intelligent sensing nodes in terms of the magnitude and direction of change is calculated as follows: For a pair of adjacent intelligent sensing nodes, obtain their change magnitude vector and change direction vector within the same time period; Calculate the ratio of the corresponding components of the change magnitude vectors of two nodes, and take the geometric mean of the ratios of all components to generate the magnitude similarity factor; Calculate the cosine similarity between the change direction vectors of two nodes to generate a direction consistency factor; Multiplying the amplitude similarity factor by the direction consistency factor yields a consistency index that characterizes the overall similarity of the node change patterns.

[0027] The steps for calculating the change difference value are as follows: Based on the calculated consistency index, take the reciprocal of the index to obtain the basic difference degree; obtain the geographical distance between a pair of adjacent intelligent sensing nodes; perform a weighted summation of the basic difference degree and the normalized value of the geographical distance, in which the basic difference degree is given a higher weight to emphasize the dominant role of the change pattern difference, and define the result of the weighted summation as the change difference value between the pair of nodes.

[0028] S3. Based on the node difference association set, perform continuous filtering on the change difference values ​​between adjacent smart sensing nodes, and mark the set of smart sensing nodes whose change difference values ​​are continuously greater than the difference threshold within a preset continuous space range as the candidate mutation node set. The continuity screening process for the variation differences between adjacent intelligent sensing nodes is as follows: For each intelligent sensing node, all directly connected neighboring nodes are searched within a circular area defined by its communication radius; the variation difference between each node and each neighboring node is determined to be greater than a preset difference threshold; the number of connections of the node that exceed the difference threshold is counted; if this number exceeds a preset minimum number of connections, the node is marked as a potential anomalous node; finally, spatial clustering analysis based on density clustering is performed on all marked potential anomalous nodes, and all nodes within clusters that are spatially continuous and have reached the minimum cluster size are collectively marked as a candidate mutation node set.

[0029] Preset continuous spatial range: The preset continuous spatial range is defined by the wireless communication radius of the intelligent sensing node, which is usually 50-200 meters. This range is set according to the characteristic spatial scale of significant changes in typical wetland vegetation patches or soil properties to ensure that physically adjacent nodes are related in ecological processes.

[0030] Difference threshold: The difference threshold is determined by calculating the historical distribution percentile of the difference values ​​of all nodes in the entire network. The preferred value is the 75th percentile. This value is based on the fact that it can effectively distinguish abnormal connections whose change patterns deviate significantly from the overall background, and ensure sensitivity to real variations while controlling the false alarm rate.

[0031] S4. Based on the candidate mutated node set, calculate the temporal fluctuation consistency index of the regional change feature sequence of each intelligent sensing node in the set within multiple consecutive sampling periods. Mark the nodes whose temporal fluctuation consistency index exceeds the preset threshold as temporally consistent nodes, and determine the intelligent sensing nodes in the candidate mutated node set that simultaneously meet the conditions of spatial continuity and temporally consistent nodes as spatial mutated regions.

[0032] Sampling period: The sampling period is preferably set to 1 hour. This value is based on the ability to capture the typical diurnal variation patterns of key physiological and ecological processes such as photosynthesis, respiration, and evapotranspiration in wetland ecosystems, as well as environmental factors (such as temperature and light), while taking into account the energy consumption and data load of intelligent sensing nodes, and achieving a balance between time resolution and system sustainability.

[0033] The calculation method of the time fluctuation consistency index is as follows: For each intelligent sensing node in the candidate mutated node set, extract its regional change feature sequence within N consecutive sampling periods to form an N-row M-column feature matrix, where M is the feature dimension; calculate the cosine similarity of the feature vector change direction between each pair of consecutive sampling periods in the time dimension of this feature matrix, and obtain a sequence containing N-1 similarity values; calculate the standard deviation of this similarity sequence, and normalize its reciprocal. The obtained value is the time fluctuation consistency index. The lower the index, the more unstable the change pattern of the node in the time dimension and the greater the fluctuation.

[0034] Preset threshold: Based on the statistical characteristics of the distribution of the consistency index of time fluctuation between stable nodes and fluctuating nodes in historical data, the preferred value is the 25th percentile of the historical distribution of the index. This threshold can filter out nodes that show a relatively stable change pattern in the time series, thereby distinguishing between "noise variation" caused by short-term disturbances and continuous, stable "true variation".

[0035] The intelligent sensing nodes that satisfy spatial continuity in the candidate mutation node set refer to all nodes that, in terms of spatial distribution, can be connected to form a connected subgraph containing at least a preset minimum number of nodes (such as 3) through the adjacency relationship between nodes. These nodes constitute anomalous regions that appear in patches in space.

[0036] The steps for optimizing the generation of sampling data are as follows: S5. Based on the spatial variation region, calculate the variation intensity of the regional change feature sequence of the intelligent sensing node in the spatial variation region, generate variation intensity value by combining the change amplitude and change direction in the regional change feature sequence according to the preset combination rule, and classify and label the intelligent sensing node based on the variation intensity value to form a hierarchical set of variation nodes. The variation intensity value is generated by combining the variation amplitude and direction in the regional variation feature sequence according to a preset combination rule. The preset combination rule is as follows: take the square root of the sum of the squares of each component of the variation amplitude vector to obtain the total variation amplitude scalar; sum the weighted components of the variation direction vector, with the weight being the prior importance coefficient of each parameter to the carbon sink process, to obtain the weighted direction index; multiply the absolute value of the total variation amplitude scalar and the weighted direction index, and then multiply by a stability coefficient positively correlated with the consistency index of time fluctuation, to finally generate the comprehensive variation intensity value.

[0037] For example, using the "amplitude-weighted directional magnitude" rule, the magnitude (L2 norm) of the amplitude vector is calculated as the product of the weighted sum of the absolute values ​​of the normalized components of the directional vector. Assume the amplitude vector is... The direction vector, after normalization, is The parameter weight vector is Then the variation intensity value is:

[0038] in, Characterizing the drasticness of change, The "net projection" representing the direction of change on the importance dimension ensures that the intensity value is positive and reflects the significance of the directional deviation.

[0039] The beneficial effect of this combination rule is that it not only quantifies the severity of change, but also introduces ecological prior knowledge through directional weighting, so that the final calculated variation intensity value can more accurately characterize those ecological "hot spots" that have undergone drastic and directional changes in parameter dimensions that have an important impact on carbon sinks, thereby improving the ecological significance of the identification results and the pertinence of subsequent resource allocation.

[0040] The intelligent sensing nodes are classified and labeled based on their mutation intensity values. The classification criteria are as follows: calculate the mutation intensity value of all intelligent sensing nodes in the spatial mutation region, and determine the 33rd and 66th percentiles of their global distribution as classification thresholds; mark nodes with mutation intensity values ​​below the 33rd percentile as "low" mutation level, mark nodes between the 33rd and 66th percentiles as "medium" mutation level, and mark nodes above the 66th percentile as "high" mutation level, thus forming a classification set of mutated nodes.

[0041] S6. Based on the hierarchical set of mutated nodes, a corresponding sampling frequency adjustment strategy is matched for intelligent sensing nodes of different levels. The sampling frequency adjustment strategy is dynamically corrected in combination with the temporal fluctuation consistency index of the spatial mutated region. The dynamically corrected sampling frequency adjustment strategy is applied to the intelligent sensing nodes to generate optimized sampling data corresponding to the spatial mutated region.

[0042] Sampling frequency adjustment strategy: This is a set of instructions tied to the mutation level, used to modify the basic sampling frequency of a node. Preset strategies matching different levels include: for "high" mutation level nodes, doubling the basic sampling frequency (e.g., increasing from 1 hour / time to 30 minutes / time); for "medium" mutation level nodes, increasing the basic sampling frequency by 50% (e.g., increasing from 1 hour / time to 40 minutes / time); and for "low" mutation level nodes, maintaining the original basic sampling frequency.

[0043] The specific method for dynamically correcting the sampling frequency adjustment strategy based on the temporal fluctuation consistency index of the spatial variation region is as follows: Calculate the average temporal fluctuation consistency index of all temporally consistent nodes in the entire spatial variation region; if the average index is lower than the preset stability threshold, it is determined that the overall change in the region is unstable in time, and the preset strategy is "aggressively corrected". For example, the frequency doubling strategy of "high" level nodes is further improved to "three times", and the strategy of "medium" level nodes is improved to "double"; if the average index is higher than the stability threshold, it is determined that the regional change is relatively stable, and the preset strategy is "conservatively corrected" or not corrected. For example, "high" level nodes are kept at "double", and "medium" level nodes are adjusted to "increase by 25%" to achieve a balance between capturing dynamics and saving energy.

[0044] One embodiment of the present invention is that the remote sensing data of the area where the intelligent sensing node is located in the multi-scale data fusion module obtains the remote sensing image data of the corresponding area by performing spatial positioning retrieval in the preset remote sensing data source based on the geographical coordinates of the area where the intelligent sensing node is located, and performs time matching filtering and spatial resolution unification processing on the remote sensing image data to generate standardized remote sensing data that corresponds one-to-one with the temporal and spatial optimized sampling data of the area where the intelligent sensing node is located. Preset remote sensing data sources include, but are not limited to, multispectral satellite data such as the Landsat series, Sentinel-2, and Gaofen (GF) series, as well as medium-resolution time-series data products such as MODIS and VIIRS. These data sources provide public or authorized data access interfaces through their official data distribution platforms or commercial cloud service platforms.

[0045] The multi-scale data fusion module performs spatial alignment and feature fusion on spatiotemporally optimized sampled data and remote sensing data. Specifically, it includes: performing coordinate system matching and spatial location matching on the two types of data based on the geographic coordinate information contained in the standardized remote sensing data and the location information of the intelligent sensing nodes corresponding to the spatiotemporally optimized sampled data; extracting time-series statistical features related to carbon sinks from the spatiotemporally optimized sampled data within the same matched area; and extracting vegetation index features and land surface temperature features corresponding to the area where the intelligent sensing nodes are located from the standardized remote sensing data. Then, it constructs a fused feature vector based on the time-series statistical features, vegetation index features, and land surface temperature features, and performs consistency verification and difference correction on the fused feature vector to generate a fused feature set for calculating spatial representative weights.

[0046] The steps for constructing the fused feature vector are as follows: For the region where a matched intelligent sensing node is located, multiple temporal statistical features (such as mean, variance, and trend) extracted from the spatiotemporally optimized sampling data are arranged in order to form a ground observation feature sub-vector; vegetation index feature values ​​and surface temperature feature values ​​of the corresponding pixel or its neighborhood are extracted from the standardized remote sensing data to form a remote sensing feature sub-vector; the ground observation feature sub-vector and the remote sensing feature sub-vector are concatenated end to end to form a long vector; the concatenated long vector is Z-score standardized to eliminate dimensional differences, and finally, the fused feature vector corresponding to the node is generated.

[0047] One embodiment of the present invention is as follows: the method for calculating the spatial representation weight is as follows: T1. Based on the fusion feature set, the fusion feature set is spatially partitioned according to the region where the intelligent sensing node is located, and the feature vector set corresponding to each intelligent sensing node is extracted in each spatial partition. Based on the feature vector set, the feature similarity between any two intelligent sensing nodes in the partition is calculated to generate a node similarity set. Specifically, the formula for calculating feature similarity is:

[0048] in, Represents a node With nodes Feature similarity between and These are the feature vectors of the two nodes, Represents the dot product. This represents the L2 norm of a vector.

[0049] Example: Select two intelligent sensing nodes A and B within a certain spatial partition, and obtain their fused feature vectors respectively. and ,in for , for The numerator is obtained by calculating the vector dot product. Calculate the L2 norm of each vector and multiply them together to get the feature similarity. Divide the numerator by the denominator to obtain the feature similarity. for This allows us to determine that the two nodes have a high degree of feature consistency within the spatial partition.

[0050] This example achieves unified comparison of multi-dimensional features through vector similarity calculation, avoiding the impact of single parameter bias on the results and improving the stability and robustness of node representativeness evaluation.

[0051] T2. Based on the node similarity set, calculate the average feature similarity between each intelligent sensing node and all other nodes in its spatial partition, and use the average feature similarity as the feature distribution consistency index of the node. Nodes with feature distribution consistency indices higher than a preset consistency threshold are marked as high consistency nodes, and nodes with consistency indices lower than the consistency threshold are marked as low consistency nodes, thus generating a node consistency classification set. Specifically, the formula for calculating the characteristic distribution consistency index is as follows:

[0052] in, Represents a node The consistency index of the characteristic distribution, This represents the total number of nodes within the spatial partition containing this node.

[0053] Example: A spatial partition contains 5 intelligent sensing nodes. (Regarding the nodes...) Calculate the feature similarity between it and the other four nodes respectively. , , , The average value is obtained by summing the four similarities and dividing by the number of nodes minus one. Use this value as a node Feature distribution consistency index .

[0054] This method can evaluate node stability from the perspective of overall distribution, avoid interference from local anomalies on the evaluation of individual nodes, and improve the overall consistency of spatial representation weight calculation.

[0055] T3. Based on the node consistency classification set, for high consistency nodes and low consistency nodes, calculate the Euclidean distance between their feature vectors and the regional feature centers of their respective spatial partitions, and define the Euclidean distance as the feature deviation degree of high consistency nodes or low consistency nodes; then, by calculating the ratio of feature distribution consistency index to feature deviation degree, generate the initial representative evaluation value of each intelligent sensing node. Specifically, the formula for calculating the degree of feature deviation and the initial representativeness evaluation value is as follows:

[0056]

[0057] in, Represents a node The degree of deviation of the characteristics This is the mean vector of the feature vectors of all nodes within the spatial partition, i.e., the feature center of the region. For Euclidean distance, Represents a node The initial representative evaluation value, For a very small positive number (such as This is to prevent the denominator from being zero.

[0058] Example: Calculate the feature center vector of a region within a spatial partition. for ,node Feature vector for The degree of deviation is calculated using Euclidean distance. for Meanwhile, the node's consistency index for Divide the consistency index by the degree of deviation plus a small value. Obtain the initial representative evaluation value for .

[0059] This method couples consistency with the degree of deviation, which ensures node stability, suppresses the influence of outliers, and improves the discrimination accuracy of representative weights.

[0060] T4. Based on the initial representative evaluation values ​​of all intelligent sensing nodes, perform normalization processing on nodes within the same spatial partition, and use the normalized results as the spatial representative weights of each node.

[0061] Specifically, the formula for calculating the space-representing weight is as follows:

[0062] in, Represents a node The final space represents the weights, and the denominator is the sum of the initial representative evaluation values ​​of all nodes within the partition.

[0063] Example: Given a spatial partition containing 4 nodes, what is their initial representative evaluation value? They are respectively , , , Divide the evaluation value of each node by the sum. To obtain the spatial representation weight They are respectively , , , .

[0064] This normalization process ensures that the total weight is 1, achieving a reasonable allocation of the contribution of different nodes to carbon sink calculation and improving calculation stability.

[0065] The method for generating carbon sink parameter distribution data is as follows: R1. Based on spatial representation weights, perform weight mapping processing on the feature vectors in the fusion feature set corresponding to each intelligent sensing node in the fusion feature set, adjust the contribution ratio of the fusion feature vectors, obtain a weighted fusion feature set, and maintain the correspondence between the intelligent sensing node and its spatial partition in the weighted fusion feature set. The steps for generating the weighted fusion feature set are as follows: For each intelligent sensing node in the fusion feature set, obtain its calculated spatial representative weight; multiply each feature value in the fusion feature vector of the node by the spatial representative weight of the node to generate a new weighted feature vector; collect and organize the weighted feature vectors of all nodes to ensure that each weighted feature vector is still bound to its original node identifier and spatial partition information, thereby forming the weighted fusion feature set.

[0066] R2. Based on the weighted fusion feature set, perform weighted partition aggregation processing according to the spatial representative weight in each spatial partition to generate the regional carbon sink feature representation of each spatial partition, and retain the contribution ratio relationship of the spatial representative weight in the regional carbon sink feature representation. The method for generating the regional carbon sink feature representation is as follows: For each spatial partition, extract the weighted feature vectors of all nodes belonging to that partition from the weighted fusion feature set; treat these weighted feature vectors as a batch of samples within that partition; calculate the mean vector of the weighted feature vectors of this batch of samples, which is the regional carbon sink feature representation of that partition. In the calculation process, the differences in the representativeness of each node have been considered through spatial representative weights, so that nodes with strong representativeness contribute more to the mean.

[0067] R3. Based on the regional carbon sink feature representation, perform continuity verification on the regional carbon sink feature representation between adjacent spatial partitions based on the consistency of feature change direction and change magnitude, connect the regional carbon sink feature representations that meet the continuity constraints, and perform transition construction processing based on spatial adjacency relationship on the regional carbon sink feature representations that do not meet the continuity constraints to generate a spatial continuous carbon sink feature field. The specific content of performing continuity verification based on the consistency of feature change direction and change magnitude for the regional carbon sink feature representation between adjacent spatial partitions is as follows: For any two adjacent spatial partitions, calculate the difference vector between their regional carbon sink feature representation vectors; calculate the magnitude of the difference vector as the feature change magnitude; and extract the direction (unit vector) of the difference vector.

[0068] In the pre-calculated global feature change direction pattern library, find the known pattern that is most similar to the geographical location relationship of the adjacent partition and obtain its standard change direction; calculate the cosine value of the angle between the measured difference vector direction and the standard change direction; if the feature change amplitude is less than the preset amplitude threshold and the cosine value of the angle is greater than the preset direction consistency threshold, then the adjacent partition is determined to meet the continuity constraint; otherwise, it is determined not to meet the constraint.

[0069] Among them, the global feature change direction pattern library is constructed based on the historical regional carbon sink feature change sequence.

[0070] The specific steps for connecting the regional carbon sink feature representations that satisfy the continuity constraint are as follows: mark all adjacent partition pairs that satisfy the continuity constraint to form a connected graph, where nodes are spatial partitions and edges are connection relationships that satisfy the constraint; find all connected components in the connected graph; for each connected component, seamlessly stitch the regional carbon sink feature representations of all its contained partitions in spatial location according to the partition boundaries to form a larger, continuous feature representation block.

[0071] The method for generating the spatially continuous carbon sink characteristic field is as follows: After completing all connections that satisfy the constraints, you will get several consecutive feature representation blocks and some isolated or partitioned feature representations that do not satisfy the connection conditions. For each consecutive feature representation block, the feature representation within the block is directly used within it; For the gap regions between blocks or between blocks and isolated partitions, a distance-based inverse distance weighted interpolation method is adopted. The spatial distance from the gap location to the surrounding known feature representation blocks or partition features is used to calculate the weight and perform interpolation, thereby assigning a feature vector to each geographical location of the entire target wetland ecosystem, and finally generating a spatially continuous carbon sink feature field that covers the entire area and is spatially continuous and gradually changes.

[0072] R4. Based on the spatial continuous carbon sink feature field, perform numerical mapping processing based on spatial location according to the correspondence between the spatial continuous carbon sink feature field and spatial location, generate carbon sink parameter distribution data covering the area where the intelligent sensing node is located, and maintain the mapping relationship between the spatial representative weight and the contribution of carbon sink parameters in the carbon sink parameter distribution data.

[0073] The specific steps for generating carbon sink parameter distribution data are as follows: First, obtain a spatially continuous carbon sink feature field, which associates a high-dimensional feature vector with each geographic grid point. Second, input the high-dimensional feature vector of each grid point into a pre-trained carbon sink parameter inversion model. This model is a deep neural network capable of mapping high-dimensional features to carbon sink parameter values. Third, the carbon sink parameter inversion model calculates for each grid point and outputs an estimated value of carbon sink per unit area. Fourth, iterate through all geographic grid points within the target area, arranging the output values ​​of all points according to their spatial location to generate a two-dimensional spatial matrix. This matrix represents the carbon sink parameter distribution data, where each pixel value represents the carbon sink intensity at the corresponding ground location.

[0074] One embodiment of the present invention involves generating the total carbon content and spatial distribution of a wetland ecosystem through the following steps: U1. Based on the carbon sink parameter distribution data, perform cumulative calculations based on spatial grid division on the carbon sink parameter distribution data in each spatial partition to generate the partition carbon sink amount of each spatial partition. Specifically, the formula for calculating the carbon sequestration by zone is as follows:

[0075] in, Indicates the first Carbon sequestration in each spatial partition. This represents the number of geographic grid cells contained in this partition. The first in the carbon sink parameter distribution data Estimated carbon sequestration per unit area of ​​each grid point For the first The actual ground area represented by each grid point.

[0076] Example: A spatial partition contains 3 grid points. Its carbon sequestration per unit area is... They are respectively , , Grid area They are respectively , , The carbon sequestration of a region is obtained by summing the products point by point. for .

[0077] This method transforms spatial distribution data into total indicators, achieving physical interpretability and quantifiability in carbon sink calculation.

[0078] U2. Perform a summation calculation on the carbon sink of all spatial zones to generate the total carbon sink of the wetland ecosystem; U3. Based on the carbon sink amount of each spatial partition and its corresponding spatial location information, perform spatial interpolation and rendering to generate a spatial distribution map of wetland ecosystem carbon sink that characterizes the spatial variation gradient of carbon sink amount. The spatial distribution map of carbon sinks in wetland ecosystems is produced by importing carbon sink parameter distribution data into geographic information system software, selecting appropriate color bands for color rendering based on their numerical values, overlaying basic geographic information layers such as wetland boundaries, water systems, and roads, and adding cartographic elements such as scale bars, north arrows, and legends, ultimately outputting a standard map product.

[0079] Uncertainty quantification generates carbon sink assessment results with accompanying uncertainty information through the following steps: V1. Based on the spatial representative weight, calculate the local variance of the spatial representative weight within each spatial partition, and use the local variance as a first-type uncertainty index caused by the difference in node representativeness. The formula for calculating the local variance of the spatial representative weights within each spatial partition is as follows:

[0080] in, Indicates the first The first type of uncertainty index (local variance) for each spatial partition. This represents the number of intelligent sensing nodes within this partition. For the first The space of each node represents a weight. This represents the average of the spatial weights of all nodes within the partition.

[0081] Example: Node weights within a spatial partition for , , , Calculate the average value for Sum the squared differences between each weight and the average value, then divide by the number of nodes. Obtain local variance for .

[0082] This method can quantify the differences in the representative distribution of nodes, which can be used to identify areas of data imbalance and improve the accuracy of uncertainty assessment.

[0083] V2. Based on standardized remote sensing data, the pixel quality score and atmospheric correction residual of the remote sensing image corresponding to each spatial partition are extracted, and the pixel quality score and atmospheric correction residual are fused to generate the second type of uncertainty index caused by the quality of remote sensing data. The method for extracting pixel quality scores and atmospheric correction residuals is as follows: directly read the quality assessment identifier of each pixel, such as cloud, cloud shadow, snow, water body, etc., from the metadata or quality assessment bands attached to the standardized remote sensing data product, and convert it into a numerical score between 0 and 1 as the pixel quality score; at the same time, obtain the root mean square error between the reflectance of each pixel after atmospheric correction and the reflectance of the theoretical model from the log or intermediate data product generated during the atmospheric correction process, as the atmospheric correction residual of that pixel.

[0084] The steps for generating the second type of uncertainty index are as follows: For each spatial partition, extract the pixel quality score and atmospheric correction residual of all remote sensing pixels within its corresponding range; calculate the average quality score of all pixels in the partition as the average quality score; calculate the average atmospheric correction residual of all pixels in the partition as the average correction residual; invert the average quality score (e.g., subtract the average from 1) to obtain the quality uncertainty component; use the average correction residual as the correction uncertainty component; perform a weighted summation of the quality uncertainty component and the correction uncertainty component, with the weights determined based on the historical sensitivity analysis of both to the final carbon sink inversion accuracy. The summation result is the second type of uncertainty index for that partition.

[0085] V3. The first type of uncertainty index and the second type of uncertainty index are weighted and fused to generate the comprehensive uncertainty value of each spatial partition. Based on the comprehensive uncertainty value of all partitions, an uncertainty distribution field consistent with the spatial range of carbon sink parameter distribution data is generated by spatial interpolation. The weights for weighted fusion are determined based on the historical reliability of the first type of uncertainty index and the second type of uncertainty index.

[0086] V4. The total carbon sequestration of wetland ecosystems, the spatial distribution map of wetland ecosystem carbon sequestration, and the uncertain distribution field are associated and encapsulated to generate carbon sequestration assessment results with uncertain information.

[0087] One embodiment of the present invention is that the node adjustment strategy in the system optimization module generates a node addition / reduction adjustment and sampling strategy update scheme based on the abnormal areas and uncertainty exceeding the limit areas identified by the carbon sink assessment results. Specifically, when a high-value area with abnormal spatial distribution of carbon sinks is identified and its uncertainty exceeds the threshold, a "supplementary strategy" is generated to deploy 2 to 3 new intelligent sensing nodes in the center of the area, and a "densified sampling strategy" is generated at the same time to increase the sampling frequency of existing nodes in the area by one level. This strategy enables targeted strengthening of monitoring efforts in areas with carbon sink "hotspots" and questionable data reliability. By increasing the spatial and temporal density of observation samples, it effectively reduces the uncertainty in these local areas, providing a more reliable data foundation for accurate assessment in the next cycle. It also enables dynamic adaptive allocation of monitoring resources according to the focus of scientific issues.

[0088] The remote sensing revisit instruction generates a remote sensing data update scheduling scheme based on abnormal regions. The specific method for generating the remote sensing data update scheduling scheme is as follows: Based on the abnormal areas identified by the carbon sink assessment results, their spatial boundary coordinates and area information are obtained; Based on the size of the area, determine the suitable remote sensing data source (for large areas, medium- and low-resolution high-frequency satellites are preferred, while for small, fine areas, high-resolution satellites are preferred). Query the mission queues and transit forecasts of cooperative satellite data centers or drone platforms, and calculate the optimal transit time window that can cover the target area in the next few days; Generate a standardized data acquisition task instruction, i.e., a remote sensing data update scheduling scheme, which includes parameters such as target area coordinates, preferred data source, expected imaging time window, minimum cloud cover requirement, and data priority level, and automatically submit it to the corresponding data service platform.

[0089] The node adjustment strategy and remote sensing revisit instructions work together on the intelligent sensing node network and the prior map of wetland spatial heterogeneity to achieve closed-loop optimization of the monitoring system.

[0090] Example In a target wetland ecosystem, the fused feature set is divided into four spatial partitions. Five intelligent sensing nodes are selected in the first spatial partition, and the average feature distribution consistency index is obtained by calculating the node similarity set. for The degree of deviation of corresponding node features for To obtain the initial representative evaluation value for Normalization is performed on all nodes within the spatial partition to generate spatial representative weights. Within this spatial partition, a weighted fusion feature set is generated by performing weight mapping processing on the fusion feature set based on spatial representation weights. This is further used to generate a regional carbon sink feature representation. After performing continuity verification on adjacent spatial partitions, a spatially continuous carbon sink feature field is constructed. Carbon sink parameter distribution data is then generated based on this spatially continuous carbon sink feature field. Finally, spatial integration is performed within this spatial partition to calculate the partitioned carbon sink amount. for The total carbon content of the wetland ecosystem is obtained by summing all spatial partitions. An uncertainty distribution field is generated by combining spatial representation weights with standardized remote sensing data. When the spatial partition is identified as an anomalous area and the uncertainty exceeds the limit, a node adjustment strategy and a remote sensing revisit instruction are generated. The node adjustment strategy is fed back to the intelligent sensing node network, and the wetland spatial heterogeneity prior map is updated to achieve closed-loop optimization.

[0091] This embodiment uses spatial representation of weights. Driven by partition fusion and continuity constraint modeling, high-precision representation of carbon sink distribution is achieved. Combined with uncertainty quantification and dynamic adjustment strategies, the ability to identify abnormal areas and the adaptive optimization level of the monitoring system are significantly improved.

[0092] A method for monitoring carbon sink data in wetland ecosystems includes the following steps: W1. Deploy a network of intelligent sensing nodes in the target wetland ecosystem, and configure node deployment and sampling strategies based on the prior map of wetland spatial heterogeneity to collect soil parameters, vegetation parameters and water parameters to generate initial sampling data. W2. Based on the prior map of wetland spatial heterogeneity and the initial sampling data, identify spatially variable regions and adjust the sampling frequency of intelligent sensing nodes in the region to generate optimized sampling data. W3. Acquire remote sensing data of the area where the intelligent sensing node is located, spatially align and fuse it with the optimized sampling data to generate a fused feature set; W4. Calculate the spatial representative weight of each intelligent sensing node based on the fusion feature set, and use the spatial representative weight to generate carbon sink parameter distribution data; W5. Calculate the total amount and spatial distribution of carbon in wetland ecosystems based on carbon sink parameter distribution data, and quantify the uncertainty of the calculation results by combining spatial representative weights and remote sensing data to generate carbon sink assessment results. W6. Analyze the carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, generate node adjustment strategies and remote sensing revisit instructions to update the deployment of the intelligent sensing node network, sampling strategies and wetland spatial heterogeneity prior maps, and achieve closed-loop optimization of monitoring.

[0093] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A wetland ecosystem carbon sink data monitoring system, characterized in that: include: The intelligent sensing node network collects soil parameters, vegetation parameters, and water parameters, and, based on a pre-constructed wetland spatial heterogeneity prior map, executes node deployment and sampling strategy configuration to obtain initial sampling data. The sampling and transmission optimization module identifies spatial variation regions based on the prior map of wetland spatial heterogeneity, adjusts the sampling frequency of the intelligent sensing nodes within the spatial variation regions, and generates optimized sampling data. The multi-scale data fusion module acquires remote sensing data of the area where the intelligent sensing node is located, performs spatial alignment and feature fusion on the spatiotemporally optimized sampling data and the remote sensing data, generates spatial representative weights based on the fusion results, and generates carbon sink parameter distribution data based on the spatial representative weights. The carbon sink assessment module calculates the total amount and spatial distribution of carbon in the wetland ecosystem based on the carbon sink parameter distribution data and the spatial representative weights, and performs uncertainty quantification on the calculation results based on the spatial representative weights and the remote sensing data to generate a carbon sink assessment result with uncertainty information. The system optimization module is used to receive the carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, generate node adjustment strategies and remote sensing revisit instructions, feed back the node adjustment strategies to the intelligent sensing node network to update its deployment and sampling strategies, and update the wetland spatial heterogeneity prior map based on the carbon sink assessment results. 2.The wetland ecosystem carbon sink data monitoring system of claim 1, wherein: The intelligent sensing node network is deployed in the target wetland ecosystem; The prior map of wetland spatial heterogeneity is generated in advance by fusing historical remote sensing data, topographic data, and wetland type data; The initial sampling data generation steps are as follows: the intelligent sensing nodes in the intelligent sensing node network are deployed in the target wetland ecosystem according to the node deployment locations determined by the wetland spatial heterogeneity prior map, and based on the sampling strategy corresponding to the wetland spatial heterogeneity prior map, the soil parameters, the vegetation parameters, and the water parameters are subjected to time-synchronous collection, outlier removal, and sequential rearrangement processing to form a standardized sampling data sequence that corresponds one-to-one with the area where the intelligent sensing node is located, thereby generating the initial sampling data.

3. The carbon sink data monitoring system of wetland ecosystem according to claim 1, characterized in that: The method for identifying the spatial variation region in the sampling and transmission optimization module is as follows: S1. Based on the standardized sampling data sequence in the initial sampling data, the standardized sampling data sequence is spatially grouped according to the region where the intelligent sensing node is located, and the variation amplitude and direction of the soil parameters, vegetation parameters and water parameters are extracted in each spatial group to generate a regional change feature sequence. S2. Based on the regional change feature sequence, calculate the consistency index of each pair of adjacent intelligent sensing nodes in each spatial group in terms of the change amplitude and the change direction, and calculate the change difference value representing the difference between nodes according to the consistency index, and generate a node difference association set. S3. Based on the node difference association set, perform continuous filtering on the change difference values ​​between the adjacent smart sensing nodes, and mark the set of smart sensing nodes whose change difference values ​​are continuously greater than the difference threshold within a preset continuous spatial range as a candidate mutation node set. S4. Based on the candidate mutated node set, calculate the temporal fluctuation consistency index of the regional change feature sequence of each intelligent sensing node in the set within multiple consecutive sampling periods. Mark the nodes whose temporal fluctuation consistency index exceeds a preset threshold as temporally consistent nodes, and determine the intelligent sensing nodes in the candidate mutated node set that simultaneously meet the conditions of spatial continuity and temporally consistent nodes as the spatial mutated region.

4. The wetland ecosystem carbon sink data monitoring system of claim 3, wherein: The steps for generating the optimized sampling data are as follows: S5. Based on the spatial variation region, calculate the variation intensity of the regional change feature sequence of the intelligent sensing node in the spatial variation region, generate a variation intensity value by combining the change amplitude and the change direction in the regional change feature sequence according to a preset combination rule, and classify and label the intelligent sensing node based on the variation intensity value to form a hierarchical set of variation nodes. S6. Based on the hierarchical set of mutated nodes, a corresponding sampling frequency adjustment strategy is matched for each of the intelligent sensing nodes at different levels. The sampling frequency adjustment strategy is dynamically corrected in combination with the temporal fluctuation consistency index of the spatial mutated region. The dynamically corrected sampling frequency adjustment strategy is then applied to the intelligent sensing node to generate optimized sampling data corresponding to the spatial mutated region.

5. The wetland ecosystem carbon sink data monitoring system of claim 4, wherein: In the multi-scale data fusion module, the remote sensing data of the area where the intelligent sensing node is located is obtained by performing spatial positioning retrieval in the preset remote sensing data source based on the geographical coordinates of the area where the intelligent sensing node is located. The remote sensing image data is then processed by time matching and spatial resolution unification to generate standardized remote sensing data that corresponds one-to-one with the area where the intelligent sensing node is located and the spatiotemporal optimized sampling data. The multi-scale data fusion module performs spatial alignment and feature fusion on the spatiotemporally optimized sampling data and the remote sensing data. Specifically, it includes: matching the coordinate system and spatial location of the two types of data based on the geographic coordinate information contained in the standardized remote sensing data and the location information of the intelligent sensing nodes corresponding to the spatiotemporally optimized sampling data; extracting time-series statistical features related to carbon sinks from the spatiotemporally optimized sampling data within the same matched area; and extracting vegetation index features and surface temperature features corresponding to the area where the intelligent sensing nodes are located from the standardized remote sensing data. Then, it constructs a fused feature vector based on the time-series statistical features, the vegetation index features, and the surface temperature features, and performs consistency verification and difference correction processing on the fused feature vector to generate a fused feature set.

6. The wetland ecosystem carbon sink data monitoring system of claim 5, wherein: The method for calculating the weights represented by the space is as follows: T1. Based on the fused feature set, the fused feature set is spatially partitioned according to the region where the intelligent sensing node is located, and the feature vector set corresponding to each intelligent sensing node is extracted in each spatial partition. Based on the feature vector set, the feature similarity between any two intelligent sensing nodes in the partition is calculated to generate a node similarity set. T2. Based on the node similarity set, calculate the average feature similarity between each intelligent sensing node and all other nodes in the spatial partition, and use the average feature similarity as the feature distribution consistency index of the node. Nodes with the feature distribution consistency index higher than a preset consistency threshold are marked as high consistency nodes, and nodes with the consistency index lower than the consistency threshold are marked as low consistency nodes, thus generating a node consistency classification set. T3. Based on the node consistency classification set, for the high consistency node and the low consistency node, calculate the Euclidean distance between the feature vector of the node and the regional feature center of the spatial partition, and define the Euclidean distance as the feature deviation degree of the high consistency node or the low consistency node. Subsequently, by calculating the ratio of the feature distribution consistency index to the feature deviation degree, an initial representative evaluation value is generated for each of the intelligent sensing nodes. T4. Based on the initial representative evaluation values ​​of all the intelligent sensing nodes, perform normalization processing on the nodes within the same spatial partition, and use the normalized results as the spatial representative weights of each node.

7. The wetland ecosystem carbon sink data monitoring system of claim 6, wherein: The method for generating the carbon sink parameter distribution data is as follows: R1. Based on the spatial representative weight, perform weight mapping processing on the feature vectors in the fusion feature set corresponding to each intelligent sensing node in the fusion feature set, adjust the contribution ratio of the fusion feature vectors, obtain a weighted fusion feature set, and maintain the correspondence between the intelligent sensing node and the spatial partition in the weighted fusion feature set. R2. Based on the weighted fusion feature set, perform partition aggregation processing with weight constraints in each spatial partition according to the spatial representative weight to generate regional carbon sink feature representations for each spatial partition, and retain the contribution ratio relationship of the spatial representative weight in the regional carbon sink feature representations. R3. Based on the regional carbon sink feature representation, perform a continuity check on the regional carbon sink feature representation between adjacent spatial partitions based on the consistency of feature change direction and change magnitude, connect the regional carbon sink feature representations that satisfy the continuity constraint, and perform a transition construction process based on spatial adjacency relationship on the regional carbon sink feature representations that do not satisfy the continuity constraint to generate a spatial continuous carbon sink feature field. R4. Based on the spatial continuous carbon sink feature field, perform numerical mapping processing based on spatial location according to the correspondence between the spatial continuous carbon sink feature field and spatial location to generate carbon sink parameter distribution data covering the area where the intelligent sensing node is located, and maintain the mapping relationship between the spatial representative weight and the contribution of the carbon sink parameter in the carbon sink parameter distribution data.

8. A wetland ecosystem carbon sink data monitoring system according to claim 7, characterized in that: The total carbon content and spatial distribution of the wetland ecosystem are generated through the following steps: U1. Based on the carbon sink parameter distribution data, perform cumulative calculations based on spatial grid division on the carbon sink parameter distribution data in each spatial partition to generate the partition carbon sink amount of each spatial partition. U2. Perform a summation calculation on the carbon sink of all the spatial partitions to generate the total carbon sink of the wetland ecosystem; U3. Based on the carbon sink amount of each spatial partition and its corresponding spatial location information, perform spatial interpolation and rendering processing to generate a spatial distribution map of carbon sink in the wetland ecosystem that characterizes the spatial variation gradient of carbon sink amount. The uncertainty quantification generates carbon sink assessment results with accompanying uncertainty information through the following steps: V1. Based on the spatial representative weight, calculate the local variance of the spatial representative weight within each spatial partition, and use the local variance as a first type of uncertainty index caused by the difference in node representativeness. V2. Based on the standardized remote sensing data, extract the pixel quality score and atmospheric correction residual of the remote sensing image corresponding to each spatial partition, and fuse the pixel quality score and atmospheric correction residual to generate a second type of uncertainty index caused by the quality of remote sensing data. V3. The first type of uncertainty index and the second type of uncertainty index are weighted and fused to generate a comprehensive uncertainty value for each spatial partition. Based on the comprehensive uncertainty value of all partitions, an uncertainty distribution field consistent with the spatial range of the carbon sink parameter distribution data is generated by spatial interpolation. V4. The total carbon content of the wetland ecosystem, the spatial distribution map of the carbon sink of the wetland ecosystem, and the uncertain distribution field are associated and encapsulated to generate the carbon sink assessment result with the attached uncertain information.

9. A wetland ecosystem carbon sink data monitoring system according to claim 1, characterized in that: The node adjustment strategy in the system optimization module generates a node addition / reduction adjustment and sampling strategy update scheme based on the abnormal areas and uncertainty exceeding the limit areas identified by the carbon sink assessment results. The remote sensing revisit instruction generates a remote sensing data update scheduling scheme based on the abnormal region. The node adjustment strategy and the remote sensing revisit command work together on the intelligent sensing node network and the wetland spatial heterogeneity prior map.

10. A method for monitoring carbon sequestration data in wetland ecosystems, characterized in that: Includes the following steps: W1. Deploy a network of intelligent sensing nodes in the target wetland ecosystem, and configure node deployment and sampling strategies based on the prior map of wetland spatial heterogeneity to collect soil parameters, vegetation parameters and water parameters to generate initial sampling data. W2. Based on the prior map of wetland spatial heterogeneity and the initial sampling data, identify spatial variation regions, and adjust the sampling frequency of intelligent sensing nodes in the region to generate optimized sampling data; W3. Obtain remote sensing data of the area where the intelligent sensing node is located, and spatially align and fuse it with the optimized sampling data to generate a fused feature set; W4. Calculate the spatial representative weight of each of the intelligent sensing nodes based on the fused feature set, and use the spatial representative weight to generate carbon sink parameter distribution data; W5. Calculate the total carbon content and spatial distribution of the wetland ecosystem based on the carbon sink parameter distribution data, and quantify the uncertainty of the calculation results by combining the spatial representative weights and remote sensing data to generate carbon sink assessment results. W6. Analyze the carbon sink assessment results, identify areas with abnormal spatial distribution of carbon sinks and areas with excessive uncertainty, generate node adjustment strategies and remote sensing revisit instructions to update the deployment of the intelligent sensing node network, sampling strategies and the prior map of wetland spatial heterogeneity, and realize closed-loop optimization of monitoring.

Images