Estimation method and system for coastal salt marsh vegetation carbon sink based on multi-temporal phenological characteristics

Abstract

The embodiment of the application relates to the field of artificial intelligence technology, and provides a coastal salt marsh vegetation carbon sink estimation method and system based on multi-temporal phenology characteristics. The method comprises the following steps: acquiring remote sensing images of a target coastal salt marsh area at key phenological periods; using an unsupervised classification method to extract an initial vegetation contour of a vegetation coverage range in the target area from the remote sensing images; constructing a phenological decision tree model with an NDVI threshold method as the core, and integrating a machine learning enhanced node division mechanism and a phenological period slope analysis method to strengthen the phenological decision tree model; inputting the initial vegetation contour into the phenological decision tree model, and performing multi-level classification on the vegetation types according to the initial vegetation contour by the phenological decision tree model to distinguish the vegetation coverage ranges of various types of vegetation in the target coastal salt marsh area; acquiring a vegetation carbon density parameter, combining the vegetation coverage range, and calculating the total carbon sink amount of the target coastal salt marsh area through spatial superposition to realize accurate vegetation classification and efficient carbon sink estimation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular to a coastal salt marsh vegetation carbon sink estimation method and system based on multi-temporal phenology characteristics. BACKGROUND

[0002] As an important blue carbon ecosystem, coastal salt marshes have extremely high carbon capture and storage capacity. Their carbon sink function plays a key role in addressing global climate change and maintaining ecological balance. Quantifying the carbon sink of coastal salt marsh vegetation is an important basis for ecological protection and carbon cycle research. Therefore, there is an increasing demand for precise identification of coastal salt marsh vegetation types and carbon sink estimation techniques.

[0003] Currently, the premise of coastal salt marsh vegetation carbon sink estimation is to accurately obtain the spatial distribution information of vegetation types. Existing technologies mainly rely on field investigation and remote sensing monitoring methods. Field investigation method obtains vegetation species, biomass and other data through field sampling, and then calculates carbon sink by combining carbon density parameters. However, it has limitations. Remote sensing technology has become the mainstream method for monitoring coastal salt marsh vegetation due to its wide coverage, short cycle, and convenient data acquisition. Traditional remote sensing methods are based on single phenological period images or single vegetation index for vegetation classification, and then for carbon sink estimation. However, in practical applications, there are many technical problems: first, the parameter setting is highly subjective and the classification stability is insufficient. Existing methods often rely on artificial experience to set classification thresholds or phenological period parameters, lack of independent filtering mechanism for key features, resulting in poor adaptability of the model to different regions and years, and insufficient stability of the classification results. Second, the utilization of phenological differences is not sufficient, and the precision of vegetation differentiation is low. The phenological characteristics of coastal salt marsh vegetation (such as Suaeda salsa, Phragmites australis, Spartina alterniflora, and Aeluropus japonicus) have subtle differences. However, traditional methods rely on static phenological indicators and fail to effectively quantify the dynamic changes of vegetation during the growth cycle, especially in key phenological stages such as the withering period. It is difficult to accurately distinguish similar vegetation types (such as Spartina alterniflora and Phragmites australis), resulting in large errors in identifying the coverage of vegetation. Third, the error of carbon sink estimation is obvious. The lack of precision in vegetation classification directly affects the matching accuracy of carbon density parameters, and then leads to the deviation of the total carbon sink estimation result, making it difficult to meet the actual demand of high-precision carbon sink monitoring.

[0004] In summary, related technologies in the estimation of coastal salt marsh vegetation carbon sink face the problems of high cost of field investigation, difficulty in large-scale dynamic monitoring, subjectivity of remote sensing classification, insufficient utilization of phenological differences, and low estimation accuracy. Therefore, how to achieve accurate vegetation classification and efficient carbon sink estimation has become an important research direction in the field of coastal salt marsh ecological monitoring. SUMMARY

[0005] The main purpose of the embodiments of the present application is to provide a coastal salt marsh vegetation carbon sink estimation method and system based on multi-temporal phenological characteristics, aiming to solve at least one of the technical problems of high cost of field investigation, difficulty of large-scale dynamic monitoring, strong subjectivity of remote sensing classification, insufficient use of phenological differences, and low estimation accuracy in related technologies.

[0006] In a first aspect, the embodiments of the present application provide a coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological characteristics, comprising:

[0007] Obtaining remote sensing images of a target coastal salt marsh area at key phenological periods;

[0008] Using an unsupervised classification method to process the remote sensing images and extract the initial vegetation outline of the vegetation coverage range in the target area;

[0009] Building a phenological decision tree model with NDVI threshold method as the core, integrating a machine learning enhanced node division mechanism into the phenological decision tree model to optimize the node level and judgment priority of the decision tree in the phenological decision tree model, and using a phenological period slope analysis method to strengthen the recognition ability of the phenological decision tree model for the phenological difference characteristics of each type of vegetation in the withering period;

[0010] Inputting the initial vegetation outline into the phenological decision tree model, and using the phenological decision tree model to classify the vegetation types at multiple levels according to the initial vegetation outline to distinguish the vegetation coverage ranges corresponding to alkali grass, reed, Spartina alterniflora, and Jiangdi in the target coastal salt marsh area;

[0011] Obtaining vegetation carbon density parameters corresponding to alkali grass, reed, Spartina alterniflora, and Jiangdi, combining the vegetation coverage ranges corresponding to alkali grass, reed, Spartina alterniflora, and Jiangdi, and calculating the total carbon sink of the target coastal salt marsh area through spatial superposition.

[0012] In a second aspect, the embodiments of the present application provide a coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological characteristics, comprising:

[0013] An acquisition module for obtaining remote sensing images of a target coastal salt marsh area at key phenological periods;

[0014] A classification module for using an unsupervised classification method to process the remote sensing images and extract the initial vegetation outline of the vegetation coverage range in the target area;

[0015] A construction module for building a phenological decision tree model with NDVI threshold method as the core, integrating a machine learning enhanced node division mechanism into the phenological decision tree model to optimize the node level and judgment priority of the decision tree in the phenological decision tree model, and using a phenological period slope analysis method to strengthen the recognition ability of the phenological decision tree model for the phenological difference characteristics of each type of vegetation in the withering period;

[0016] a decision module configured to input the initial vegetation profile into the phenology decision tree model, and perform multi-level classification on the vegetation types according to the initial vegetation profile by using the phenology decision tree model, so as to distinguish the vegetation coverage ranges corresponding to the Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizome Johanson.

[0017] an estimation module configured to obtain the vegetation carbon density parameters corresponding to the Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizome Johanson, combine the vegetation coverage ranges corresponding to the Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizome Johanson, and perform carbon sink total amount estimation of the target coastal salt marsh area by using spatial superposition calculation.

[0018] In a third aspect, the embodiments of the present application further provide an electronic device, which comprises a processor, a memory for storing a computer program, and the processor is configured to execute the computer program and implement the method for estimating the coastal salt marsh vegetation carbon sink based on the multi-temporal phenology characteristics according to the first aspect or any of the embodiments of the present application.

[0019] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer software program, and the computer software program is executed by a processor to implement the method for estimating the coastal salt marsh vegetation carbon sink based on the multi-temporal phenology characteristics according to the first aspect or any of the embodiments of the present application.

[0020] The embodiment of the present application provides a coastal salt marsh vegetation carbon sink estimation method and system based on multi-temporal phenological characteristics. First, remote sensing images of a target coastal salt marsh region at key phenological periods are obtained. Then, an unsupervised classification method is used to process the remote sensing images to extract the initial vegetation profile of the vegetation coverage range in the target region. A phenological decision tree model is constructed with the NDVI threshold method as the core. The node level and judgment priority of the decision tree in the phenological decision tree model are optimized by integrating a machine learning enhanced node division mechanism into the phenological decision tree model. The ability of the phenological decision tree model to identify the difference characteristics of each type of vegetation in the withering period is strengthened by using the slope analysis method of the phenological period. Then, the initial vegetation profile is input into the phenological decision tree model, and the phenological decision tree model classifies the vegetation types at multiple levels according to the initial vegetation profile to distinguish the vegetation coverage ranges corresponding to alkali grass, reed, Spartina alterniflora and Jiangdi in the target coastal salt marsh region. Finally, the vegetation carbon density parameters corresponding to alkali grass, reed, Spartina alterniflora and Jiangdi are obtained, and the carbon sink total amount estimation of the target coastal salt marsh region is completed by spatial superposition calculation combined with the vegetation coverage ranges corresponding to alkali grass, reed, Spartina alterniflora and Jiangdi. The embodiment of the present application realizes the precision and comprehensiveness of carbon sink evaluation, completes the whole process optimization from fine vegetation classification to accurate carbon sink total amount estimation, and effectively solves the technical problems of high on-site investigation cost, difficulty in large-scale dynamic monitoring, strong subjectivity of remote sensing classification, insufficient use of phenological differences and low estimation accuracy in the related art. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological characteristics is provided for the embodiment of the present application.

[0022] Figure 2 A module structure diagram of a coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological characteristics is provided for the embodiment of the present application.

[0023] Figure 3 A structure schematic diagram of a terminal device is provided for the embodiment of the present application. DETAILED DESCRIPTION

[0024] The embodiment of the present application provides a coastal salt marsh vegetation carbon sink estimation method and system based on multi-temporal phenological characteristics. The coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological characteristics can be applied to a terminal device. The terminal device can be a mobile terminal such as a mobile phone, a virtual reality device, a tablet computer, a notebook computer, a desktop computer, a wearable device or the like. The terminal device can be a server connected to a cloud service system, or a server cluster. The above connection mode can be realized by a hardware circuit or a communication module.

[0025] Some embodiments of the present application will be described in detail below with reference to the drawings. The following embodiments and features of the embodiments described below can be combined with each other in the case of no conflict. Please refer to Figure 1 , Figure 1 A flowchart of a coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological characteristics provided by an embodiment of the present application is shown.

[0026] As Figure 1 shown, the coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological characteristics includes the following steps:

[0027] Step S101, acquiring remote sensing images of the target coastal salt marsh area in the key phenological period.

[0028] In an embodiment of the present application, the target coastal salt marsh area refers to the coastal salt marsh area that needs to be estimated for vegetation carbon sink. This area can be selected by researchers or technical implementers. This area has typical coastal salt marsh ecological characteristics, and there are characteristic vegetation such as alkali bush, reed, cordgrass, and jiangdi. It is greatly affected by environmental factors such as tides and salinity. Its specific range can be determined according to research or application needs, and usually includes complete vegetation growth environment gradient and ecological system unit.

[0029] The key phenological period is a period with large spectral feature difference in the growth process of vegetation. For example, the key phenological period includes at least one of the following: June, August, and November. Specifically, June is the initial stage of the coastal salt marsh vegetation entering the vigorous growth period, and the growth trend of the vegetation gradually appears. Various types of vegetation begin to show obvious growth characteristics. August is the peak period of vegetation growth, and the vegetation biomass reaches a high level. Its spectral characteristics are more prominent in remote sensing images, and it is an important period for distinguishing different types of vegetation. In November, it enters the withering period of vegetation, and the withering process of different vegetation is different. This difference can be reflected in the spectral information in the remote sensing image, especially for the differentiation of cordgrass and reed, which are easily confused vegetation. Selecting these three periods as the key phenological period can comprehensively capture the phenological characteristics of vegetation at different stages in the growth cycle, and provide rich and key information support for subsequent vegetation classification.

[0030] In terms of acquiring remote sensing images of the target coastal salt marsh area in the key phenological period, it can be collected by remote sensing satellites or aerial remote sensing equipment. For example, in step S101, remote sensing image data of the target coastal salt marsh area in at least one key phenological period of June, August, and November is collected by remote sensing satellites or aerial remote sensing equipment. The remote sensing image data contains spectral information for extracting NDVI vegetation index.

[0031] The remote sensing satellite has the characteristics of wide coverage and stable data acquisition period, and can realize remote sensing monitoring of a large range of coastal salt marshes. The aerial remote sensing equipment has higher spatial resolution and can obtain more detailed remote sensing image data in a specific area. The combination of the two can meet the monitoring needs of different scales and different accuracies. The collected remote sensing image data needs to contain spectral information that can extract NDVI and other vegetation indexes. NDVI is an important indicator reflecting the growth status of vegetation, and its value is closely related to vegetation coverage and growth vigor. Remote sensing images containing such spectral information are the basis for subsequent vegetation profile extraction. Through unsupervised classification method, the approximate range of vegetation can be quickly delineated according to these spectral information. At the same time, these spectral information is also the core data for building the phenology decision tree model, which provides the original basis for the key phenological period and index combination screening mechanism of the machine learning enhanced node division mechanism in the model and the NDVI change slope calculation by the phenological period slope analysis method. In addition, in the process of vegetation classification, the spectral information in the remote sensing image can assist the model in accurately distinguishing different vegetation types, ensuring the accuracy of the recognition of the vegetation coverage range, and thus laying a reliable data foundation for the estimation of the total carbon sink.

[0032] In step S102, the remote sensing image is processed by using an unsupervised classification method to extract an initial vegetation profile of the vegetation coverage range in the target region.

[0033] In the embodiment of the present application, the initial vegetation profile is the spatial distribution boundary of the vegetation coverage range preliminarily extracted from the remote sensing image by the unsupervised classification method. The initial vegetation profile is the result of preliminarily distinguishing the vegetation and non-vegetation regions based on the spectral characteristics (such as NDVI vegetation index) of the remote sensing image.

[0034] As an optional embodiment, in step S102, for the image noise characteristics of the target coastal salt marsh region caused by tidal disturbance and water reflection, the remote sensing image is preprocessed, the salt and pepper noise and strip noise in the remote sensing image are removed by an adaptive filtering algorithm, and the radiation bias correction and geometric precise correction are completed by combining the regional terrain data and the tidal correction model, to ensure the accuracy of the image spectral information and spatial position. Then, for the preprocessed remote sensing image, the improved K-means unsupervised classification algorithm is used to perform hierarchical automatic clustering analysis on the surface features in the preprocessed remote sensing image by dynamically adjusting the clustering center and the iteration threshold, and the vegetation coverage area and the non-vegetation coverage area are distinguished according to the spectral response difference of the NDVI vegetation index. Finally, the pixel set corresponding to the vegetation coverage area is selected from the clustering result by setting the vegetation spectral feature threshold interval, the discrete noise pixels are removed by morphological filtering, and the spatial distribution range of the pixel set is delineated as the initial vegetation profile of the vegetation coverage range in the target coastal salt marsh region.

[0035] Specifically, the extraction process of the initial vegetation profile in step S102 is carried out around image denoising and accurate clustering under tidal disturbance. In view of the problems of alternating appearance of water body reflection and mud flat exposure due to tidal rise and fall in the target coastal salt marsh area, and the problems of salt and pepper noise and strip noise caused by sensor noise, the remote sensing image is first subjected to targeted pretreatment. The image is scanned pixel by pixel through an adaptive filtering algorithm. For salt and pepper noise with sudden change in gray value, a median filtering window dynamic adjustment strategy is adopted. The window size is adaptively switched between 3x3 and 5x5 pixels according to the noise density. For regularly distributed strip noise, the image is converted to the frequency domain through Fourier transform, the spectral peak corresponding to the strip noise is identified and filtered, and the spectral details of the vegetation are effectively preserved. In the pretreatment stage, the regional 1:5000 terrain data and the tidal correction model are combined. According to the tidal level data at the image acquisition time, the radiation deviation of each pixel is corrected, the spectral reflectance deviation caused by the difference in tidal submergence depth is eliminated, and the geometric precision correction is completed through the ground control points, so that the image spatial position error is controlled within 1 pixel, and the dual accuracy of spectral information and spatial position in subsequent analysis is ensured.

[0036] After the pretreatment, the improved K-means unsupervised classification algorithm is used to cluster the image features. The algorithm first determines the optimal number of clusters as 6 classes based on the statistical distribution characteristics of the image NDVI vegetation index, covering high-coverage vegetation, low-coverage vegetation, water body, mud flat, tidal ditch and artificial features, etc. In the clustering process, a dynamic adjustment mechanism is introduced. The initial clustering center is automatically generated according to the NDVI mean value and the band reflectance mean value. After each iteration, the intra-class variance and inter-class distance of each class are calculated. When the intra-class variance of a class exceeds the preset threshold (1.2 times the standard deviation of spectral reflectance), the clustering center is automatically split and the pixels are re-assigned. When the inter-class distance of two classes is less than the merging threshold, they are merged into one class. The dynamic adjustment of clustering center and iteration threshold realizes hierarchical automatic clustering analysis. In the clustering process, the spectral response difference of NDVI vegetation index is mainly relied on. The high NDVI value (≥0.3) area corresponds to the area with vigorous photosynthesis of vegetation, and the low NDVI value (<0.1) area corresponds to the non-vegetation area, so as to distinguish the preliminary clustering results of vegetation coverage area and non-vegetation coverage area.

[0037] Finally, the vegetation coverage area is selected from the clustering results, and the precise extraction is realized by setting the threshold interval of the spectral characteristics of vegetation. According to the spectral characteristics of salt marsh vegetation, the cluster marked as vegetation coverage class is obtained by setting NDVI≥0.2, red band reflectivity≤0.25 and near-infrared band reflectivity≥0.4. The pixel set corresponding to these classes is extracted. In order to eliminate the discrete noise pixels generated in the clustering process, morphological filtering is used for post-processing. First, the isolated noise points are removed by erosion operation with a 3×3 structure element, and then the complete form of the vegetation area is restored by inflation operation. Finally, the spatial distribution range of the pixel set after filtering is determined as the initial vegetation contour of the vegetation coverage range in the target coastal salt marsh area. The contour can clearly distinguish the boundary between vegetation and non-vegetation areas, and provide a reliable basis for the fine classification of the subsequent phenology decision tree model.

[0038] In addition to the improved K-means unsupervised classification algorithm in the above embodiment, step S102 can also be implemented by using other unsupervised classification methods. In actual application, the selection can be based on the scene characteristics and actual image processing requirements.

[0039] For the coastal salt marsh image with large spectral fluctuation caused by tidal interference, when the ISODATA algorithm is used to extract the initial vegetation contour, first, the remote sensing image is subjected to adaptive filtering and denoising to remove salt and strip noise. Combined with regional terrain data and tidal correction model, the radiation deviation correction and geometric fine correction are completed to retain the spectral information used to extract the NDVI vegetation index. The pre-processed image data is input, the initial cluster number is set to 6 to cover vegetation, water, and light beach, the maximum iteration number is 50, the intra-class standard deviation threshold is 0.05, and the minimum intra-class pixel number is 30. The mean value of the spectral characteristics of the pixels is calculated by iteration, and the class merging and splitting operations are automatically performed. For example, when the number of pixels in a class is less than the minimum threshold, it is merged into the adjacent class, and when the intra-class standard deviation exceeds the threshold, it is split into two classes, and the cluster number is dynamically optimized. The NDVI mean value of each cluster is calculated, the clusters with NDVI greater than 0.15 are selected as vegetation candidate classes, and the pixel sets of adjacent vegetation candidate classes are merged to obtain the preliminary vegetation coverage area. The 3×3 morphological filtering is used to remove isolated noise pixels with an area less than 5 pixels, and the output of the smoothed boundary is a continuous vegetation coverage range, which is the initial vegetation contour.

[0040] For the scene with high real-time requirement, the initial vegetation outline is quickly generated by the improved K-means with dynamic threshold optimization. The pre-processing procedure is simplified to only keep the radiometric correction and geometric precise correction. The NDVI index is calculated and used as the core feature. The elbow method is used to determine the optimal number of clusters as 5. The cluster centers are dynamically initialized based on the NDVI distribution (the vegetation center is set as NDVI of 0.4, and the non-vegetation center is set according to the spectral distribution of water and mudflat). The similarity between pixels and cluster centers is calculated by Euclidean distance. The iteration threshold is dynamically adjusted (the iteration stops when the intra-class variance change rate is less than 1%). The vegetation and non-vegetation clusters are quickly divided. The cluster pixels with NDVI greater than 0.15 are directly selected, and after merging, small noise is removed by fast morphological filtering (1x3 window). The vegetation coverage boundary is generated at the fastest speed, and the initial vegetation outline meeting the real-time requirement is output.

[0041] For the complex spectral scene of alkali bush mixed with mudflat in high salinity area, the spectral clustering is used to improve the discrimination of spatial-spectral joint features. First, the NDVI vegetation index is calculated for the pre-processed image, and the 3-dimensional feature vector (NDVI, red band reflectance, near-infrared band reflectance) is constructed by combining the red and near-infrared band reflectance to enhance the spectral difference between vegetation and non-vegetation. The similarity between pixels is calculated based on the Gaussian kernel function, with the spatial distance weight accounting for 30% and the spectral distance weight accounting for 70%. The similarity matrix containing spatial adjacency relationship is constructed. The similarity matrix is decomposed, and the first 5 largest eigenvalues are selected to form the low-dimensional embedding space. The embedding vectors are clustered by K-means, and the number of clusters is set to 5. The NDVI mean value and spatial continuity of each cluster are calculated. The cluster with NDVI greater than 0.2 and spatial connectivity greater than 80% is marked as vegetation class and its pixel set is extracted. Small holes inside the vegetation area are filled by morphological closing operation, and the boundary is smoothed by erosion-dilation operation. The initial vegetation outline is output.

[0042] For the scene of vegetation phenology dynamic change of multi-temporal images, in combination with superpixels and density peak clustering to extract the initial vegetation profile, SLIC algorithm is used to generate superpixels for multi-temporal preprocessed images (such as June, August and November), the size of superpixels is set to 20x20 pixels, and the compactness parameter is set to 10 to ensure that the spectral and spatial features in each superpixel are consistent. The average NDVI of each superpixel in the key phenological period, the NDVI change slope (6-8 months, 8-11 months) are calculated, and a 5-dimensional time series feature vector (3 NDVI in the period + 2 slope) is constructed. The local density of the superpixel feature vector (the cutoff distance is set to 1.5 times the standard deviation of the feature space) and the distance peak value are calculated, and the clustering center is automatically identified, that is, the superpixel with high local density and far distance from other peak values is selected. According to the time series features of the clustering center, the superpixel set with NDVI average not less than 0.18 and NDVI in the growth period (August) greater than that in the withering period (November) is screened and marked as vegetation superpixel. The pixels corresponding to the vegetation superpixels are merged, the superpixel edge sawtooth is removed through boundary smoothing processing, and the initial vegetation profile consistent across time is output.

[0043] For high-dimensional data such as hyperspectral images, the DEC algorithm is used to automatically learn the optimal features to realize vegetation profile extraction. First, the hyperspectral image is radiometrically normalized, and 15 sensitive bands (such as red, near-infrared, and short-wave infrared bands) related to vegetation growth are selected to retain the core spectral information after dimensionality reduction. A neural network including an encoder (for example, the encoder is composed of an input layer, a 32-dimensional hidden layer, and a 10-dimensional embedding layer) and a decoder is constructed, and the hyperspectral pixel spectrum is input. The network is pre-trained through the reconstruction loss function to learn the deep feature representation of the data. The 10-dimensional embedding features output by the encoder are subjected to K-means clustering (the initial cluster number is 4) to generate pseudo-labels, and the clustering loss is calculated combined with the pseudo-labels to fine-tune the autoencoder parameters to enhance the discriminability of the features. The average NDVI value (calculated based on the red and near-infrared bands) of each cluster is calculated, and the cluster with NDVI not less than 0.2 is divided into the vegetation class and the corresponding pixel set is extracted. Morphological filtering is used to remove noise pixels, and spatial connectivity analysis is used to fill in the gaps in the vegetation area, and the high-resolution initial vegetation profile is output.

[0044] In step S103, an NDVI threshold method is used as the core to construct a phenology decision tree model, a machine learning enhanced node division mechanism is integrated into the phenology decision tree model to optimize the node level and judgment priority of the decision tree in the phenology decision tree model, and a phenological period slope analysis method is used to strengthen the recognition ability of the phenology decision tree model for the difference characteristics of each type of vegetation in the withering period.

[0045] In the embodiments of the present application, the phenology decision tree model is a decision tree model optimized for coastal salt marsh vegetation classification, which takes vegetation phenology features as the core discriminant basis and integrates multi-source data and algorithm mechanisms to realize accurate classification.

[0046] Based on the NDVI threshold method, a decision framework was constructed, combined with the growth characteristics of coastal salt marsh vegetation (Suaeda salsa, Phragmites australis, Spartina alterniflora, and Scirpus mariqueter), and dynamic NDVI threshold intervals were set for different key phenological periods. In the growth peak period (August), the NDVI threshold values of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Scirpus mariqueter were set to 0.6-0.8, 0.4-0.6, 0.2-0.4, respectively. In the withering period (November), the corresponding adjustments were 0.3-0.5, 0.2-0.4, and 0.1-0.3, respectively, forming a basic decision node system and achieving preliminary differentiation of vegetation types. The random forest algorithm was embedded as a node optimization mechanism, and the input was a combination of NDVI, EVI2, and MSAVI in June, August, and November. Through feature importance sorting, the core classification features were autonomously mined, such as the contribution of August EVI2 to Spartina alterniflora classification reaching 45%, and the contribution of June NDVI to Phragmites australis classification accounting for 38%. According to the weight order, the node level was dynamically adjusted, with high-contribution feature nodes placed in the upper layer for priority judgment, such as using "August EVI2 > 0.5" as the first-level discrimination node for Spartina alterniflora, to optimize the decision logic priority. The 8-11 month withering period NDVI change slope was introduced as a dynamic verification index, and the difference in decay rate of different vegetation was used to improve the classification accuracy. The absolute value of the withering period slope of Spartina alterniflora was larger (-0.02 to -0.05 / month), and the threshold was set to ≤-0.03 / month. The slope of Phragmites australis was moderate (-0.01 to -0.03 / month), forming a differentiated threshold system. The slope threshold was embedded in the decision tree as a key verification node, and when the initial classification result conflicted with the slope feature (such as classified as Phragmites australis but the slope ≤-0.03 / month), the secondary discrimination mechanism was triggered to reclassify, achieving secondary verification of the feature optimization result.

[0047] In this way, the NDVI threshold method ensures the intuitiveness and interpretability of the classification logic, the random forest enhancement mechanism improves the scientificity of feature mining and node optimization, and adapts to the spectral complexity of salt marsh vegetation. The analysis of the slope in the phenological period strengthens the sensitivity to the decay characteristics of the withering period vegetation, effectively solving the confusion problem of similar vegetation types. Through the integration of multiple mechanisms, the model can accurately distinguish the spatial distribution of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Scirpus mariqueter, providing high-quality vegetation classification data support for coastal salt marsh vegetation carbon sink estimation, and providing a referenceable technical framework for fine classification of vegetation in other ecological monitoring scenarios.

[0048] Step S103 focuses on the construction of the NDVI threshold method-based phenology decision tree model. By integrating machine learning enhancement mechanisms and slope analysis of phenological periods, the differences in the withering period of salt marsh vegetation are accurately identified. Specifically, first, the NDVI vegetation index extracted from the preprocessed remote sensing image is used as the core parameter to construct the model framework. Considering the growth characteristics of coastal salt marsh vegetation, dynamic initial NDVI threshold intervals are set for the phenological differences of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Scirpus mariqueter. In the growing season (August), the NDVI threshold of Spartina alterniflora is set to 0.6-0.8, that of Phragmites australis to 0.4-0.6, and that of Suaeda salsa to 0.2-0.4. In the withering season (November), the NDVI threshold of Spartina alterniflora is lowered to 0.3-0.5, that of Phragmites australis to 0.2-0.4, and that of Suaeda salsa remains 0.1-0.3, forming a basic decision node system adapted to the salt marsh environment.

[0049] On this basis, the random forest algorithm is integrated as a machine learning enhancement node division mechanism. The input is a combination of vegetation index data at key phenological periods (June, August, and November) after radiation and geometric correction, including NDVI, EVI2, and modified soil-adjusted vegetation index (MSAVI). The feature importance sorting algorithm is used to calculate the weights of the input features, and the core classification features are autonomously mined: the contribution weight of August EVI2 to Spartina alterniflora classification accounts for 45%, the contribution of June NDVI to Phragmites australis classification accounts for 38%, and the contribution of November NDVI to Suaeda salsa classification accounts for 32%. According to the weight order, the decision tree node level is dynamically adjusted, and the nodes corresponding to high-contribution features are placed in the upper layer of the decision tree for priority judgment. For example, August EVI2>0.5 is used as the first-level discrimination node for Spartina alterniflora, and June NDVI in the range of 0.3-0.5 is used as the second-level discrimination node for Phragmites australis, optimizing the node judgment priority.

[0050] Meanwhile, the slope analysis method of phenological periods is used to strengthen the identification of withering period vegetation differences. Based on the NDVI data at key phenological periods, the NDVI change slope of the growth stage from June to August and the withering stage from August to November is calculated through linear regression algorithm, with the slope value of the withering stage as the core dynamic index. By comparing the slope distribution characteristics of the vegetation in the sample library, it is found that the absolute value of the slope of Spartina alterniflora in the withering period is larger (-0.02 to -0.05 / month), the absolute value of the slope of Phragmites australis is medium (-0.01 to -0.03 / month), and the slope of Suaeda salsa is relatively flat (-0.005 to -0.02 / month). Accordingly, differentiated adaptive thresholds are set, with the slope threshold of Spartina alterniflora set to ≤-0.03 / month, that of Phragmites australis set to -0.01 to -0.03 / month, and that of Suaeda salsa set to -0.005 to -0.02 / month, embedded in the decision tree as a key verification node. When a pixel is classified as Phragmites australis by the initial NDVI threshold but the slope in the withering period is ≤-0.03 / month, a secondary discrimination mechanism is triggered, and it is reclassified as Spartina alterniflora, forming a secondary check on the feature optimization results, thereby improving the identification accuracy of withering period vegetation types.

[0051] Through the above steps, the constructed phenology decision tree model has the intuitiveness of NDVI threshold, the feature optimization capability of machine learning, and the dynamic discrimination advantage of phenological slope, can effectively distinguish the spectral differences of different vegetation types at key phenological periods, especially enhances the sensitivity recognition of the decay period vegetation attenuation characteristics, and lays a model foundation for subsequent precise vegetation classification.

[0052] As an optional embodiment, in step S103, first, taking the NDVI vegetation index extracted from the pre-processed remote sensing image as the core parameter, combining the dynamic initial NDVI threshold interval set according to the growth characteristics of coastal salt marsh vegetation, the model framework of the phenology decision tree model suitable for salt marsh environment is constructed. Secondly, the random forest algorithm is embedded in the model framework as a machine learning enhanced node division mechanism, the vegetation index combination data of each key phenological period after radiation bias correction and geometric precision correction are input, the vegetation index combination data includes NDVI and EVI2, the weight relationship of different key phenological periods and index combination to vegetation classification is autonomously mined through the feature importance sorting algorithm, the core features with the highest contribution to the classification of Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus are identified, and the node level order and judgment priority of the decision tree are dynamically adjusted according to the weight order of the core features. Thirdly, based on the NDVI data of the key phenological period, the linear regression algorithm is used to calculate the NDVI change slope of June-August and August-November, the slope value of August-November is taken as the core dynamic index of the phenological slope analysis method, the adaptive threshold is set by comparing the NDVI decay rate difference of Spartina alterniflora and Phragmites australis in the decay period, the adaptive threshold is embedded into the decision tree as a key verification node, a secondary check of the feature optimization result is formed, the sensitivity recognition accuracy of the phenology decision tree model to the phenological difference of the decay period vegetation is strengthened, and the phenology decision tree model is constructed.

[0053] Specifically, the process of constructing the phenology decision tree model starts from the basic framework construction, taking the NDVI vegetation index extracted from the pre-processed remote sensing image as the core parameter, combining the growth characteristics of the vegetation in the coastal salt marsh area, setting the initial NDVI threshold interval varying with the phenological period, and building the model basic framework adapting to the salt marsh environment, providing the initial discrimination basis for the subsequent vegetation classification. On this basis, in order to optimize the node division logic of the model, the random forest algorithm is embedded in the constructed model framework as a machine learning enhanced node division mechanism, and the combination data of the key phenological period vegetation index are input, which includes NDVI and EVI2. Through the feature importance sorting algorithm, the weight relationship of different key phenological periods and index combinations in vegetation classification is analyzed, the core features with the highest contribution to the classification of alkali grass, reed, Spartina alterniflora and Jiangdi are identified, and the hierarchical order and judgment priority of each node in the decision tree are dynamically adjusted according to the weight sorting results of the core features, so that the model can preferentially rely on more discriminant features for discrimination. In order to further strengthen the model's ability to identify the differences in the withering period of vegetation, based on the NDVI data of the key phenological period, the linear regression algorithm is used to calculate the NDVI change slope of the growth stage from June to August and the withering stage from August to November, and the slope value from August to November is used as the core dynamic index of the phenological slope analysis method. By comparing the NDVI decay rate difference of Spartina alterniflora and reed in the withering period, the corresponding adaptive threshold is set, which is embedded in the decision tree as a key verification node, and the preliminary classification results obtained by core feature optimization are subjected to secondary verification, so as to strengthen the sensitivity and recognition accuracy of the phenology decision tree model to the withering period vegetation phenology difference, and finally complete the construction of the whole phenology decision tree model.

[0054] Exemplarily, based on the NDVI data of the key phenological period in the above steps, the linear regression algorithm is used to calculate the NDVI change slope of June-August and August-November, and the slope value of August-November is used as the core dynamic index of the phenological slope analysis method. By comparing the NDVI decay rate difference of Spartina alterniflora and reed in the withering period, the adaptive threshold is set, which is embedded in the decision tree as a key verification node, forming a secondary verification of the feature optimization results, including:

[0055] The NDVI data of the key phenological period after pretreatment is time series smoothed to eliminate short-term fluctuations. The weighted linear regression algorithm is used to calculate the NDVI change slope of the growth period from June to August and the withering period from August to November. The weight of the NDVI data in November is the highest in the calculation of the withering period slope to highlight the late decay characteristics. Then, based on the distribution characteristics of the withering period slope of Spartina alterniflora and Phragmites australis in the historical vegetation classification sample library, the slope interval is divided by a dynamic clustering algorithm, and different adaptive thresholds are set for different vegetation types. For example, the withering period slope threshold of Spartina alterniflora is set to -0.02 / month to -0.05 / month, and that of Phragmites australis is set to -0.01 / month to -0.03 / month, forming a threshold system adapted to different vegetation types. Finally, the threshold is embedded in the decision tree model as an independent verification node to cross-check the preliminary classification results after feature optimization. When the sample slope value exceeds the threshold interval of the corresponding vegetation type, the secondary discrimination mechanism is triggered. Based on the verified withering period slope value, the vegetation carbon density parameters are dynamically allocated to build a correlation model between the slope and the biomass decay. For herbaceous vegetation such as Spartina alterniflora and Phragmites australis, the calculation method is to multiply the basic carbon density value by 1.2 times the normalized slope value. The greater the absolute value of the slope, the higher the decay coefficient and the lower the carbon density value. At the same time, an abnormal slope warning mechanism is established. When the withering period slope of a certain area deviates from the mean value of the same type of vegetation by more than 2 times the standard deviation, it is marked as a suspicious area to be verified in the field.

[0056] The above steps revolve around the calculation and application of the slope of the NDVI change around the phenological period. The core principle is to capture the spectral dynamic characteristics of vegetation at different growth stages, and to strengthen the accurate differentiation of withered vegetation types. Specifically, first, the NDVI data after preprocessing of the key phenological period is subjected to time series smoothing processing, so as to eliminate the interference brought by short-term environmental fluctuations, and make the data more reflect the true trend of vegetation growth. When calculating the slope, the weighted linear regression algorithm is used to process the data of the growth peak period and the withering period respectively. Especially in the calculation of the slope in the withering period, by configuring higher weight for the end period data, the spectral attenuation characteristics of the withering stage of vegetation are highlighted, and the slope value is more in line with the actual change rule of the withering process of vegetation. Then, relying on the spectral slope distribution characteristics accumulated in the historical vegetation classification sample library, the dynamic clustering algorithm is used to divide the slope interval, and the adaptive threshold is set according to the growth characteristics of different vegetation types, forming a threshold system suitable for the growth law of each type of vegetation, and providing clear basis for subsequent classification verification. Finally, these thresholds are embedded into the decision tree model as independent verification nodes to cross-check the preliminary classification results. When the sample slope value exceeds the threshold interval of the corresponding type, the secondary discrimination mechanism is triggered to correct the classification results. At the same time, based on the verified slope value, a correlation model with biomass decay is established, so that the allocation of vegetation carbon density parameters is more in line with the actual growth state of vegetation, and through the establishment of a slope anomaly early warning mechanism, the area with abnormal growth state is marked in time, providing guidance for on-site verification. Thus, through the above examples, not only the recognition sensitivity of the decision tree model to the phenological differences of withered vegetation is improved, and the classification confusion of similar vegetation types is reduced, but also the dynamic optimization of vegetation carbon density parameters is realized, providing more accurate basic data for carbon sink estimation, and at the same time, the reliability and verifiability of the model results are enhanced through the abnormal early warning mechanism.

[0057] As an optional embodiment, after step S103, the tidal table data of the target coastal salt marsh area and the real-time monitoring tidal level data of the hydrological station, and the salinity monitoring data of the target coastal salt marsh area can also be preprocessed to remove outliers and fill in missing values. For example, abnormal records caused by instrument failure or extreme weather are identified and removed by an outlier detection algorithm, and for missing values that occur during data collection, a time series interpolation method is used to fill in the missing values to ensure the continuity and integrity of the data. Further, based on the preprocessed tidal table data and tidal level data, the correlation between tidal level and vegetation type distribution is analyzed, and the tidal level division threshold is determined based on the growth characteristics of Suaeda salsa and Spartina alterniflora, and a tidal level threshold node is added to the constructed phenology decision tree model. Among them, the tidal level threshold node is set to classify Suaeda salsa first in the low tidal level bare area, and Spartina alterniflora first in the high tidal level area. Specifically, based on the preprocessed tidal and tidal level data, the distribution rule of vegetation types in different tidal level intervals is analyzed, and combined with the growth characteristics that Suaeda salsa prefers to grow in the middle and low parts of the intertidal zone and is more densely distributed in the low tidal level bare area, while Spartina alterniflora is more suitable for higher tidal level environment, the tidal level division threshold is determined, and a tidal level threshold node is added to the constructed phenology decision tree model. When the model discriminates the vegetation type of a certain area, if the area belongs to the low tidal level bare area, it is preferentially inclined to be classified as Suaeda salsa, and if it is in the high tidal level area, it is preferentially considered to be classified as Spartina alterniflora, and the rationality of the classification logic is improved by integrating the ecological correlation between tidal level and vegetation distribution. On this basis, according to the influence relationship of tidal height and salinity on the estimation of net exchange of the ecological system, combined with the difference analysis of the measured data of vegetation distribution and the preliminary classification results of the model under different salinity gradients, the salinity correction factor and the applicable salinity interval for Suaeda salsa, Phragmites australis, Spartina alterniflora, and Rhizoglyphus japonicus are determined according to their respective salinity ranges, forming a ternary correspondence table containing salinity interval, vegetation type, and correction factor. For example, the classification result of Suaeda salsa in a high salinity environment is positively corrected, and the classification weight of Phragmites australis is optimized in a moderate salinity interval. Finally, the salinity correction factor is integrated into the classification result output layer of the phenology decision tree model to optimize the classification accuracy of the phenology decision tree model. For example, after the model completes the preliminary classification, the corresponding correction factor is called according to the salinity data of the area to fine-tune the classification result, further reducing the classification deviation caused by salinity differences, and finally through multi-dimensional optimization of tidal and salinity data, the classification accuracy of the phenology decision tree model is more in line with the actual distribution of vegetation in the complex environment of the coastal salt marsh.

[0058] It is worth mentioning that in the above embodiment, the acquired tidal and tidal level data are first systematically arranged, abnormal fluctuation data caused by instrument failure or extreme weather are excluded through an outlier detection algorithm, and missing values occurring in the monitoring process are filled in by using a time series interpolation method to ensure the continuity and reliability of the data. On this basis, combined with the topographic and geomorphic features of the target coastal salt marsh area, the tidal level data are divided into multiple continuous intervals according to the spatial distribution, the distribution frequency of alkali grass, Spartina alterniflora and other vegetation types in each interval in the historical vegetation survey data is counted, and the internal correlation between different tidal level environments and vegetation community distribution is analyzed. According to the growth characteristics of the vegetation found in the ecological investigation, alkali grass as a typical intertidal vegetation, its growth depends on a certain length of low tidal level bare environment, and it is most densely distributed in the bare area at the low part of the intertidal zone, while Spartina alterniflora has stronger flooding resistance and tends to form a dominant community in areas with relatively high tidal level and high flooding frequency. Based on the difference in ecological adaptability, by statistical analysis of the dominance proportion of the two types of vegetation in different tidal level intervals, the key threshold of tidal level division is determined, that is, when the tidal level is lower than a certain value, the area is defined as a low tidal level bare area, and when the tidal level is higher than another value, it is divided into a high tidal level area. This tidal level threshold node is added to the constructed phenology decision tree model as an important supplementary link of the model classification logic. When the model discriminates the vegetation type of a certain area, the tidal level attribute data of the area are called first, if the data show that the area belongs to a low tidal level bare area, the model will preferentially classify it as alkali grass based on the NDVI threshold and the analysis of the phenology slope, because the environmental conditions are highly matched with the growth requirements of alkali grass. If the area is in a high tidal level area, the flooding resistance of Spartina alterniflora is combined to preferentially determine Spartina alterniflora in the classification decision. The above embodiment fully integrates the natural ecological correlation between tidal level and vegetation distribution, when the model appears ambiguity in the NDVI feature or the phenology slope feature, the tidal level threshold node can provide clear ecological logic support, reduce the misclassification caused by similar spectral features, make the classification decision more in line with the natural distribution law of coastal salt marsh vegetation, and further improve the accuracy and rationality of the whole phenology decision tree model in identifying vegetation types.

[0059] As an optional embodiment, after integrating the salinity correction factor into the classification result output layer of the phenology decision tree model in the above steps, a three-dimensional decision rule including the three dimensions of tidal level, salinity, and vegetation type can be constructed based on the added tidal level threshold node, the determined salinity correction factor, and the original vegetation classification features in the phenology decision tree model, wherein the original vegetation classification features include one of the following: NDVI threshold, phenological period slope, and the three-dimensional decision rule is used to represent the corresponding relationship between the tidal level interval, the salinity range, and the classification results of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Rhizoglyphus japonicus. Further, a typical vegetation distribution area of the target coastal salt marsh area is selected as a verification data set, the three-dimensional decision rule is applied to the model classification process, and the classification accuracy is calculated by comparing the measured vegetation type in the verification data set with the output result of the phenology decision tree model. Finally, according to the accuracy calculation result, the division threshold of the tidal level threshold node and the value and applicable salinity range of the salinity correction factor are iteratively optimized, if the vegetation classification error in a certain tidal level interval exceeds a preset value, the tidal level threshold of the target coastal salt marsh area is adjusted, and if the classification deviation is large in a certain salinity range, the salinity correction factor of the corresponding vegetation type is corrected, until the model classification accuracy meets the preset requirement, so as to optimize the classification accuracy of the phenology decision tree model.

[0060] In the above embodiments, the classification accuracy of the phenology decision tree model is improved by constructing a multi-dimensional decision rule and iteratively optimizing parameters. After integrating the salinity correction factor into the model output layer, a three-dimensional decision rule including the three dimensions of tidal level, salinity, and vegetation type is constructed based on the added tidal level threshold node, the determined salinity correction factor, and the original vegetation classification features such as NDVI threshold and phenological period slope in the model, so as to clearly define the classification relationship of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Rhizoglyphus japonicus in different tidal level intervals and salinity ranges. For example, in a region with medium tidal level and low salinity, the rule will preferentially correspond to Phragmites australis; in a region with high tidal level and medium salinity, it will tend to match Spartina alterniflora. Further, a typical region with known vegetation distribution in the target region is selected as a verification data set, such as a Suaeda salsa concentration area in a low tidal level bare area and a Spartina alterniflora growth area in a high tidal level area, and the three-dimensional decision rule is applied to the model classification process. By comparing the observed vegetation type in the verification data set with the model output result, the overall classification accuracy and the confusion matrix of each type of vegetation are calculated. If it is found that part of the Phragmites australis in a certain high tidal level interval is misclassified as Spartina alterniflora, and the error exceeds a preset value, it indicates that the tidal level threshold division in this interval is unreasonable, and the upper limit of the tidal level threshold needs to be adjusted to narrow the range of the high tidal level interval. If the classification deviation of Suaeda salsa is large in a high salinity region, the salinity correction factor of Suaeda salsa in this salinity range is corrected to enhance its classification weight in a high salinity environment. After multiple rounds of parameter iteration and adjustment, the model classification accuracy reaches the preset standard, thereby realizing the optimization of the classification accuracy of the phenology decision tree model.

[0061] It can be understood that the three-dimensional decision rule is a multi-dimensional classification judgment system constructed in the optimization process of the phenology decision tree model. The core is to form a precise classification judgment standard by integrating the correlation of the three key elements of tidal level, salinity and vegetation type. It is based on the original NDVI threshold and phenological period slope in the model, and adds the tidal level threshold node and the salinity correction factor to form a three-dimensional corresponding relationship framework including tidal level interval, salinity range and vegetation type, and to determine the priority classification logic of various types of vegetation under different environmental conditions. In practical application, the three-dimensional decision rule will set different judgment logic according to the ecological adaptability characteristics of vegetation. For example, according to the growth characteristics of Suaeda salsa that prefers low tidal level and high salinity, the rule will prefer Suaeda salsa in the area with low tidal level and high salinity. For Phragmites australis which is more suitable for low and medium salinity and medium tidal level environment, the rule will set it as the priority classification result in the area with medium tidal level and low salinity. Spartina alterniflora is more suitable for high tidal level environment, so when the tidal level is high and the salinity is in the medium range, the rule will tend to classify the vegetation in this area as Spartina alterniflora. This rule system is not fixed, but is closely related to the ecological habits of vegetation, forming a dynamic adaptive classification guide.

[0062] For example, when the model classifies the vegetation in a certain area, the three-dimensional decision rule will first determine the tidal level interval according to the tidal level data of the area, then match the corresponding salinity range according to the salinity monitoring data, and finally output the most possible vegetation type according to the preset three-dimensional correspondence. Through this multi-dimensional cross verification, the three-dimensional decision rule can effectively overcome the limitations of single feature classification, reduce the classification deviation caused by environmental factors, provide more comprehensive and more ecological classification basis for the model, and thus improve the accuracy of the phenology decision tree model in distinguishing the vegetation types in coastal salt marshes.

[0063] Step S104, input the initial vegetation profile into the phenology decision tree model, and classify the vegetation type according to the initial vegetation profile through the phenology decision tree model to distinguish the vegetation coverage range corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Jiangdi in the target coastal salt marsh area.

[0064] As an optional embodiment, in step S104, the initial vegetation profile is converted into a model input format at a pixel scale, combined with the key phenological NDVI data and vegetation index combination data obtained through preprocessing, and input into the phenological decision tree model. Further, the multi-level classification process of the model is started, the first-level classification is based on dynamic initial NDVI threshold interval for rough classification, and the interference pixels that do not meet the basic vegetation spectral characteristics are removed, and the vegetation pixel set containing Suaeda salsa, Phragmites australis, Spartina alterniflora, and Rhizoglyphus japonicus is selected as the first-level classification result. The second-level classification calls the machine learning enhanced node division mechanism, and according to the weight order of the core features, the key phenological period and index combination are selected through the fusion of the dynamic weight distribution random forest optimization algorithm, and the growth characteristics of the coastal salt marsh vegetation in different phenological periods are combined to assign dynamic weights to the vegetation index in the key phenological period; the key phenological period and index combination selected are used for dynamic classification of the first-level classification result to preferentially distinguish the vegetation types whose NDVI characteristics difference meets the set condition, and Spartina alterniflora in the high NDVI value interval and Suaeda salsa in the low NDVI value interval are obtained as the second-level classification result. The third-level classification enables the phenological period slope verification mechanism, and for the easily confused vegetation types in the second-level classification result, the comparison between the 8-11 month withering period NDVI change slope and the adaptive threshold is used for discrimination, the pixels with the slope value meeting the Spartina alterniflora threshold interval are marked as Spartina alterniflora, and the pixels with the slope value meeting the Phragmites australis threshold interval are marked as Phragmites australis, and the third-level classification result is obtained. Then, the three-dimensional decision rule of the tidal level threshold node and the salinity correction factor is fused to spatially verify the classification results of each level, the vegetation pixels in the low tidal level exposed area are combined with the salinity correction factor to optimize the classification boundary of Suaeda salsa, and the pixels in the high tidal level area strengthen the classification weight of Spartina alterniflora. Finally, the isolated noise pixels in the classification results of each level are removed through morphological post-processing, the vegetation coverage range boundary is smoothed, and the corresponding vegetation coverage range vector data of Suaeda salsa, Phragmites australis, Spartina alterniflora, and Rhizoglyphus japonicus after classification are output.

[0065] Exemplarily, in the optional embodiment of step S104, the vegetation type multi-level classification process realizes accurate differentiation through multi-level feature screening and cross-validation. First, the initial vegetation profile is converted into an input format recognizable by the model at the pixel scale, and the pre-processed key phenological NDVI data and NDVI, EVI2 and other vegetation index combination data are imported synchronously to provide multi-dimensional basic features for model classification. The multi-level classification process starts with the first level of rough division, and based on the dynamic initial NDVI threshold interval, non-vegetation pixels such as water and light beach are removed, and only the pixel set meeting the vegetation spectral characteristics is retained. For example, pixels with growth peak NDVI < 0.2 are determined as non-vegetation interference items, and vegetation pixels possibly containing alkali reed, reed, Spartina alterniflora and Jiangni are screened out to form the first level classification result. The second level classification calls the machine learning enhanced node division mechanism, and assigns dynamic weights to vegetation indexes at different phenological periods through random forest algorithm. For example, the weight of EVI2 for Spartina alterniflora classification is increased in August, and the weight of NDVI for reed is increased in June, and the key feature combination is screened out by using the weight order. According to these features, the first level result is further classified, and types with obvious spectral difference are preferentially distinguished, such as pixels in the high NDVI value interval are preliminarily marked as Spartina alterniflora, and pixels in the medium and low NDVI value interval are marked as alkali reed, to obtain the second level classification result. The third level classification enables the phenological period slope verification for the types that are easily confused in the second level result (such as reed and Spartina alterniflora), and through comparison of the NDVI change slope in the withering period from August to November with the adaptive threshold, the pixels meeting the Spartina alterniflora decay characteristics are confirmed, and the pixels meeting the reed characteristics are re-marked to reduce type confusion.

[0066] Subsequently, the three-dimensional decision rule is fused for spatial verification. The pixels in the low tide bare area combine with the salinity correction factor to refine the Suaeda boundary, and the pixels in the high tide area pass through the tide threshold to strengthen the Spartina classification weight, to ensure that the classification result is consistent with the ecological adaptability characteristics. Exemplarily, in the spatial verification stage of multi-level classification of vegetation, the core logic of fusing the three-dimensional decision rule is to bind the vegetation classification result with the ecological correlation depth of environmental factors such as tide level and salinity, and to optimize the classification boundary through targeted adjustment. For the pixels in the low tide bare area, first, determine the low tide interval to which it belongs according to the tide data. Due to the short submerging time and high exposure frequency, this area naturally adapts to the growth habit of Suaeda. At this time, combined with the salinity monitoring data, the salinity correction factor is called. If a pixel is in a high-salinity environment, and there is a blurred boundary between Suaeda and other vegetation in the first and second classifications, the classification signal of Suaeda is strengthened by increasing the weight of the salinity correction factor. For example, in the area where the salinity exceeds a certain threshold, the originally blurred boundary pixel is automatically classified into the Suaeda category, so as to refine the spatial distribution boundary of Suaeda and ensure that its coverage range matches the typical habitat characteristics of “low tide plus high salinity”. For the pixels in the high tide area, the classification weight of Spartina is strengthened based on the setting of the tide threshold node. The high tide area has a longer submerging time, and the water condition is more suitable for the growth of Spartina. When the model has a confusion between Spartina and reed in the second or third classification of the vegetation type of a pixel, the weight adjustment mechanism is triggered by calling the tide attribute label to automatically increase the classification priority of Spartina in this area. For example, if the NDVI characteristics and slope value of a pixel meet part of the conditions of Spartina and reed, the system will preferentially determine it as Spartina according to its attribute of being in the high tide area, avoiding classification deviation caused by similar spectral characteristics. Through this spatial verification combined with specific habitat conditions, the final vegetation classification result not only conforms to the spectral feature law, but also is highly consistent with the ecological adaptability characteristics of Suaeda, Spartina and other vegetation, further improving the accuracy and rationality of the classification boundary.

[0067] Finally, the isolated noise pixels are removed and the boundary is smoothed through morphological post-processing, and the coverage range vector data of each of the three types of vegetation is output, realizing the step-by-step optimization from rough division to refinement.

[0068] Step S105, obtain the vegetation carbon density parameters corresponding to Suaeda, reed, Spartina and Jiangni, combine the vegetation coverage ranges corresponding to Suaeda, reed, Spartina and Jiangni, and calculate the total carbon sink amount of the target coastal salt marsh area through spatial superposition.

[0069] As an optional embodiment, in step S105, the sampling points are arranged in the target coastal salt marsh area according to a first preset size of a unit grid, and the Suaeda, Phragmites australis, Spartina alterniflora and Rhizoglyphus are sampled in the field respectively, and the sampling quantity of each type of vegetation is not less than a set quantity. The carbon content of each type of vegetation is determined by collecting the vegetation samples, and a basic carbon density original data set containing the coordinates of the sampling points and the carbon content values is generated. Further, the basic carbon density original data set is spatially matched with the regional scale vegetation carbon density literature data, the Kriging interpolation method is used to spatially interpolate and calibrate the field sampling data, the interpolation parameters are adjusted through cross-validation to expand the coverage range of the sampling point data, and the calibrated basic carbon density parameters are adapted to the growth environment characteristics of the target area vegetation in the spatial distribution. For example, in the target coastal salt marsh area, the sampling points are arranged according to a 100m x 100m grid, and the Suaeda, Phragmites australis, Spartina alterniflora and Rhizoglyphus are sampled respectively, and at least 30 samples of each type of vegetation are sampled. For example, in a Spartina alterniflora growing area, plant samples at different positions are collected, and the carbon content is determined in the laboratory, thereby generating basic data containing the coordinates of each sampling point and the corresponding carbon content, such as the carbon content of the Spartina alterniflora corresponding to the sampling point coordinates (X1, Y1) is a value. Then, the field sampling basic data and the existing vegetation carbon density literature data in the region are matched, and the sampling data is expanded by the Kriging interpolation method. Like taking the carbon content of the known sampling points as a reference, the carbon density of the unsampled area is calculated by interpolation, and the adjusted carbon density distribution is adjusted through cross-validation to be consistent with the local vegetation growth environment, such as the carbon density parameters in the water sufficient area are adapted to the characteristics of more vigorous vegetation.

[0070] Then, based on the interpolated basic carbon density parameters, a multi-dimensional correction matrix of NDVI change slope in the withering period from August to November, salinity monitoring data, and tidal level data is established. Among them, for the NDVI change slope in the withering period from August to November in the dynamic characteristics of the phenology, the absolute value of the slope is set to increase by 0.01 / month, and the carbon density correction coefficient is set to decrease by 0.02, which is related to the slope label of the pixel; for the salinity in the environmental gradient characteristics, the salinity is set to increase by 5‰, and the correction coefficient of the reed carbon density is set to decrease by 0.05, which is mapped to the salinity interval label; for the tidal level, the correction coefficient of the Spartina alterniflora in the high tidal level area is set, and the low tidal level exposure time of the Suaeda salsa is set to have a floating interval of the reference value, and the reference value is set to increase by 0.1 for every 2 hours of exposure time, which is bound to the tidal level attribute label. Exemplarily, based on the interpolated basic carbon density parameters, a multi-dimensional correction matrix is established. Assuming that the absolute value of the NDVI change slope of the Spartina alterniflora in a certain area increases by 0.01 / month, the corresponding carbon density correction coefficient decreases by 0.02, and is related to the slope label of the pixel in the area; if the salinity in a certain area increases by 5‰, the carbon density correction coefficient of the reed decreases by 0.05, which corresponds to the salinity interval label; the correction coefficient of the Spartina alterniflora in the high tidal level area is set, and the reference value of the Suaeda salsa in the low tidal level area increases by 0.1 for every 2 hours of exposure time, which is bound to the tidal level attribute label.

[0071] The multi-dimensional correction matrix is a parameter adjustment tool integrating multiple influencing factors such as phenology dynamics and environmental gradient, which is used to accurately correct the basic carbon density parameters of different vegetation types. By establishing the correlation between the phenological characteristics, environmental factors and correction coefficients, the carbon density is finely calibrated. Specifically, this matrix covers the correction logic of three key dimensions: first, the dynamic characteristics of the phenological period, taking the NDVI change slope in the withering period from August to November as the core index, the growth and decline rate of the vegetation is bound to the carbon density correction coefficient, the greater the absolute value of the slope, the faster the vegetation decays, and the correction coefficient decreases; second, the environmental salinity gradient, according to the influence of different salinity intervals on vegetation growth, a special correction rule is set for vegetation sensitive to salinity, and the correction coefficient decreases when the salinity increases; third, the tidal level environmental characteristics, according to the difference in tidal level adaptability of Spartina alterniflora and Suaeda salsa, the Spartina alterniflora in the high tidal level area is set to have a tidal level correction coefficient, and the Suaeda salsa is associated with the low tidal level exposure time and the reference value floating rule, the longer the exposure time, the more the correction coefficient increases. Through this multi-dimensional parameter correlation, the multi-dimensional correction matrix can closely combine the basic carbon density parameters with the actual growth state of the vegetation and the environmental adaptability, making the carbon density assignment more suitable for the complex and variable ecological environment of the coastal salt marsh, and providing accurate parameter support for subsequent pixel-level carbon density calculation and total carbon sink estimation.

[0072] Further, a differentiated parameter calling model is constructed based on the multi-dimensional correction matrix, wherein the differentiated parameter calling model comprises a Spartina alterniflora model, a reed model, and a Suaeda model; the Spartina alterniflora model takes the basic carbon density value as input, and outputs the result of weighted calculation of the standardized slope value taken from the pixel slope label and the tidal level correction coefficient taken from the tidal level attribute label; the reed model multiplies the correction coefficient corresponding to the salinity interval label by the basic value; and the Suaeda model determines the reference value floating multiple according to the low-tidal-level exposure duration label, to form a three-dimensional carbon density parameter library containing data correlation rules. The vegetation coverage range vector data corresponding to the Suaeda, reed, Spartina alterniflora, and Jiangni respectively are converted into pixel grids of a second preset size, the phenology slope label, salinity interval label, and tidal level attribute label of the pixel are matched with the three-dimensional carbon density parameter library through spatial coordinate correlation, the corresponding model is called to complete pixel-level carbon density assignment, and a carbon density grid data set is generated. Based on the carbon density grid data set, a carbon sink amount is calculated by using a hierarchical weighted summation algorithm, wherein the pixel weight is determined by weighted fusion of the classification confidence and the August NDVI value. In the example, when the differentiated model is constructed, the Spartina alterniflora model multiplies the basic carbon density value by (1 plus 0.1 times the standardized slope value) and then by the tidal level correction coefficient, for example, the basic value is b, the standardized slope value is c, and the tidal level correction coefficient is 1.1, and the result is b x (1+0.1c) x 1.1; the reed model multiplies the basic value by the correction coefficient corresponding to the salinity interval, for example, if the salinity is in the interval of ≤10‰, the correction coefficient is 1.05, and the basic value is d, and the result is d x 1.05; and the Suaeda model determines the reference value floating multiple according to the low-tidal-level exposure duration, for example, if the exposure duration is 10 hours, the reference value is increased by 0.1 x ((10-8) / 2) = 0.1, to form the three-dimensional carbon density parameter library. The coverage range vector data of the three types of vegetation are converted into 30m x 30m pixel grids, each type of label and parameter library of the pixel is matched through coordinate matching, each pixel is assigned a carbon density, and a carbon density grid data set is generated.

[0073] Finally, a sampling area with a preset proportion of the total area is selected, the measured value of the soil profile carbon storage of the sampling area is compared with the estimated value of the corresponding area model, a regional error correction coefficient is calculated and fed back to the carbon density grid data set for correction, and the estimated results including the total carbon sink amount of the target area, the carbon sink contribution rate of each vegetation type, and the carbon density spatial distribution heat map are output. Continuing the above example, the carbon sink amount is calculated by using the hierarchical weighted summation algorithm based on the carbon density grid data set of the foregoing example, and the pixel weight is combined with the classification confidence and the August NDVI value, for example, the pixel weight is higher when the classification confidence is high and the August NDVI value is large. Finally, a region with a total area of 5% is selected as a sampling area, the measured value of the soil profile carbon storage of the region is compared with the estimated value of the model, the error correction coefficient is calculated, the carbon density grid data set is corrected by using the error correction coefficient, and finally the results including the total carbon sink amount, the carbon sink contribution rate of each vegetation, and the carbon density heat map are output.

[0074] Further optionally, in the process of estimating the carbon sink of coastal salt marsh vegetation, the precision of the model and the integrity of the evaluation can be improved through multi-source data fusion and technical linkage. The observation data of the eddy correlation method of the flux tower are introduced, and the observation results of the net exchange of the ecosystem are used to calibrate the vegetation carbon sink model, so that the model output is more in line with the actual carbon exchange process of the ecological system. For example, the real-time captured net exchange data of the ecosystem by the flux tower can reflect the carbon exchange dynamics between the ecosystem and the atmosphere. These observation results are compared and analyzed with the simulation output of the vegetation carbon sink model. By adjusting the carbon allocation parameters and the phenology response coefficient in the model, the carbon exchange process output by the model is consistent with the diurnal and seasonal variation law of the actual ecological system.

[0075] At the same time, the underground biomass sampling work can also be combined to supplement the blind area of remote sensing technology in underground carbon storage monitoring through laboratory analysis of the carbon content of root samples, and to perfect the full-factor coverage of carbon sink evaluation. For example, in view of the limitation that remote sensing technology cannot penetrate the surface to monitor underground carbon storage, underground biomass is sampled, root samples at different depths are collected in typical vegetation communities, and root carbon density data are obtained through laboratory carbon content analysis. These measured data are used as reference values to supplement the model to correct the deviation of the underground carbon storage inverted by remote sensing, and to realize full-factor carbon sink evaluation coverage from the ground to the ground. On this basis, machine learning enhanced nodes are applied to the assimilation process of flux data, and the idea of flux data correcting the light use efficiency model is used to dynamically optimize the classification rules of the decision tree with observation data, so that the vegetation type division and carbon sink parameter assignment are more suitable for the actual dynamic changes of the ecological system. Specifically, the flux observation data are input into the decision tree model as dynamic constraint conditions, the judgment threshold of the vegetation classification node and the assignment rule of the carbon sink parameter are adjusted in real time through the feature importance reordering algorithm, so that the vegetation type division result is more in line with the actual community distribution, and the carbon density parameter is accurately optimized with the dynamic changes of the ecological system.

[0076] To build a more comprehensive carbon sink evaluation system, further optionally, add the sediment deposition rate inversion link, use the suspended sediment concentration data obtained by remote sensing technology to invert the regional deposition condition, and at the same time, through laboratory calibration of the organic carbon content of soil samples, establish a soil carbon burial evaluation module. By combining the vegetation carbon storage and the soil carbon burial, a double-channel evaluation framework including vegetation carbon sink and soil carbon burial is formed, which more completely reflects the carbon sink function of the coastal salt marsh ecological system.

[0077] Exemplarily, through multi-period observation of the target coastal salt marsh area by remote sensing technology, the suspended sediment concentration data of different time periods are obtained, which can reflect the distribution and migration of sediment in the water body. Based on the spatial distribution characteristics and variation law of suspended sediment concentration, combined with the regional hydrodynamic model, the sediment deposition rate of the whole region is obtained. The principle is that the suspended sediment settles under the action of water flow, and the deposition rate is related to the suspended sediment concentration, water flow velocity and other factors. By capturing this correlation through remote sensing data and converting it into a quantitative deposition rate index, the accumulation of sediment in the salt marsh area is determined. At the same time, soil sample collection work is carried out, representative soil profiles are selected in different vegetation coverage areas and terrain positions of the salt marsh, soil samples are collected layer by layer and soil organic carbon content is calibrated in the laboratory to determine the storage of organic carbon in different depth of soil, and an evaluation module of soil carbon burial is established. The core principle is that the organic carbon in the soil will be gradually buried and accumulated with the deposition of sediment, and by calibrating the organic carbon content of soil in different areas, the potential and current situation of soil carbon burial can be quantified.

[0078] Further, the vegetation carbon storage calculated by vegetation classification and carbon density parameter is integrated with the soil carbon burial amount analyzed by sediment deposition rate and soil organic carbon content to form a double-channel evaluation framework including vegetation carbon sink and soil carbon burial. Vegetation carbon sink reflects the process of carbon fixation by photosynthesis of vegetation, while soil carbon burial embodies the long-term storage of carbon carried by sediment deposition in soil, both of which constitute an important part of the carbon sink of coastal salt marsh ecosystem. This double-channel framework can more comprehensively cover the carbon cycle process of the ecosystem, making up for the limitations of relying only on vegetation carbon storage evaluation, thus more completely reflecting the overall function of coastal salt marsh ecosystem in carbon sequestration and sink enhancement.

[0079] Further, laser radar data is compatible to improve the accuracy of vegetation canopy structure analysis, especially for low coverage woody vegetation, to solve the underestimation problem of optical remote sensing in carbon storage evaluation of this type of vegetation. Combined with the demand of blue carbon trading scene, on the basis of outputting the total carbon sink, a carbon sink spatial distribution heat map is generated, and the uncertainty interval of the evaluation result is determined to provide accurate spatial data support for carbon credit accounting, and enhance the practicality and reliability of carbon sink estimation results in actual application scenarios.

[0080] From the data source, the remote sensing image of the key phenological period is obtained in the embodiments of the present application, the spectral differences of the vegetation in different growth stages are captured, and a high-quality data foundation is laid for subsequent analysis. The unsupervised classification method combined with targeted pretreatment effectively eliminates the influence of noise such as tidal interference, and the extracted initial vegetation profile accurately distinguishes the vegetation and non-vegetation areas, ensuring the reliability of the classification starting point. The construction of the phenological decision tree model is the core innovation point in the embodiments of the present application. Based on the NDVI threshold method as the basic framework, after integrating the machine learning enhanced node division mechanism, the key classification features can be autonomously mined and the node priority can be optimized, which improves the accuracy of the vegetation type differentiation. The introduction of the slope analysis method of the phenological period further strengthens the recognition ability of the decay period vegetation attenuation difference, greatly reduces the confusion rate of similar vegetation types, and makes the classification results of alkali grass, reed, Spartina alterniflora and Jiangdi more accurate. In the carbon sink estimation link, the accurate vegetation coverage range is combined with the corresponding carbon density parameters through spatial superposition calculation, and a double-channel evaluation framework is constructed by combining the sediment deposition inversion and soil organic carbon calibration, which not only covers the vegetation carbon sink, but also includes the soil carbon burial, and comprehensively covers the key processes of the ecological system carbon cycle. Overall, the embodiments of the present application realize the whole process optimization from fine classification of vegetation to accurate estimation of total carbon sink, effectively make up for the shortcomings of traditional methods in dynamic feature capture, classification accuracy and evaluation integrity, and provide reliable technical support for scientific evaluation of the carbon sink function of the coastal salt marsh ecosystem.

[0081] Please refer to Figure 2 , Figure 2 A coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological characteristics is provided in the embodiments of the present application, and the system comprises: an acquisition module configured to acquire remote sensing images of a target coastal salt marsh area at key phenological periods;

[0082] A classification module is configured to process the remote sensing images by using an unsupervised classification method to extract an initial vegetation profile of the vegetation coverage range in the target area;

[0083] A construction module is configured to construct a phenological decision tree model based on the NDVI threshold method, integrate a machine learning enhanced node division mechanism into the phenological decision tree model to optimize the node level and judgment priority of the decision tree in the phenological decision tree model, and use a slope analysis method of the phenological period to strengthen the recognition ability of the phenological decision tree model to the phenological difference characteristics of each type of vegetation in the decay period;

[0084] A decision module is configured to input the initial vegetation profile into the phenological decision tree model, and perform multi-level classification on the vegetation types according to the initial vegetation profile by using the phenological decision tree model, so as to distinguish the vegetation coverage ranges corresponding to alkali grass, reed, Spartina alterniflora and Jiangdi in the target coastal salt marsh area;

[0085] An estimation module is configured to obtain the vegetation carbon density parameters of Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus, combine the vegetation coverage ranges of Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus, and calculate the total carbon sink amount of the target coastal salt marsh area through spatial superposition.

[0086] In some embodiments, the coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological features can be applied to a terminal device. It should be noted that, for the convenience and brevity of description, the specific working process of the above-described coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological features can refer to the corresponding process in the foregoing embodiment of the coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological features, and will not be described here.

[0087] Please refer to Figure 3 , Figure 3 The embodiment of the present application provides a structural schematic block diagram of a terminal device.

[0088] As Figure 3 shown, the terminal device 300 includes a processor 301 and a memory 302, and the processor 301 and the memory 302 are connected through a bus 303, such as an I2C bus. Specifically, the processor 301 is configured to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit, and the processor 301 can also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the embodiment of the present application, and does not constitute a limitation on the terminal device to which the embodiment of the present application is applied. Specifically, the server can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. Among them, the processor is configured to run the computer program stored in the memory, and implement any one of the coastal salt marsh vegetation carbon sink estimation methods based on multi-temporal phenological features provided by the embodiments of the present application when executing the computer program. It should be noted that, for the convenience and brevity of description, the specific working process of the above-described terminal device can refer to the foregoing embodiment of the coastal salt marsh vegetation carbon sink estimation method based on multi-temporal phenological features, and will not be described here.

[0089] The embodiment of the present application also provides a storage medium for computer readable storage, and the storage medium stores one or more programs, which can be executed by one or more processors to implement the steps of any one of the coastal salt marsh vegetation carbon sink estimation methods based on multi-temporal phenological features provided by the embodiments of the present application.

Claims

1. A method for estimating coastal salt marsh vegetation carbon sink based on multi-temporal phenological features, characterized in that, The method comprises: acquiring remote sensing images of a target coastal salt marsh area at key phenological periods; processing the remote sensing images using an unsupervised classification method to extract an initial vegetation profile of the vegetation coverage range in the target area; constructing a phenological decision tree model with NDVI threshold method as the core, incorporating a machine learning enhanced node division mechanism into the phenological decision tree model to optimize the node level and judgment priority of the decision tree in the phenological decision tree model, and using a phenological period slope analysis method to strengthen the recognition ability of the phenological decision tree model for the difference characteristics of each type of vegetation in the withering period, including: taking the NDVI vegetation index extracted from the preprocessed remote sensing images as the core parameter, combining the growth characteristics of coastal salt marsh vegetation to set a dynamic initial NDVI threshold interval, and constructing a model framework of a phenological decision tree model adapted to the salt marsh environment; embedding a random forest algorithm as a machine learning enhanced node division mechanism in the model framework, inputting a combination of vegetation index data at each key phenological period after radiation bias correction and geometric precision correction, and the combination of vegetation index data including NDVI and EVI2, autonomously mining the weight relationship of different key phenological periods and index combinations for vegetation classification through a feature importance sorting algorithm, identifying the core features with the highest classification contribution degree among alkali grass, reed, common reed, and Jiangdi, and dynamically adjusting the node level order and judgment priority of the decision tree according to the weight order of the core features; based on the NDVI data of the key phenological period, using a linear regression algorithm to calculate the NDVI change slope from June to August and from August to November, taking the slope value from August to November as the core dynamic index of the phenological period slope analysis method, setting an adaptive threshold by comparing the NDVI decay rate difference between common reed and reed in the withering period, embedding the adaptive threshold into the decision tree as a key verification node to form a secondary check of the feature optimization result, and strengthening the sensitivity recognition accuracy of the phenological decision tree model for the withering period vegetation phenological difference, to complete the construction of the phenological decision tree model; inputting the initial vegetation profile into the phenological decision tree model, and classifying the vegetation types according to the initial vegetation profile through the phenological decision tree model to distinguish the vegetation coverage range corresponding to alkali grass, reed, common reed, and Jiangdi in the target coastal salt marsh area; acquiring the vegetation carbon density parameters corresponding to alkali grass, reed, common reed, and Jiangdi, combining the vegetation coverage range corresponding to alkali grass, reed, common reed, and Jiangdi, and calculating the total carbon sink amount of the target coastal salt marsh area through spatial superposition.

2. The method of claim 1, wherein, The key phenological periods include at least one of June, August, and November; and the acquisition of the remote sensing images of the target coastal salt marsh area at the key phenological periods comprises: acquiring remote sensing image data of the target coastal salt marsh area at at least one key phenological period in June, August, and November through a remote sensing satellite or an aerial remote sensing device, and the remote sensing image data includes spectral information for extracting the NDVI vegetation index.

3. The method of claim 1, wherein, The processing of the remote sensing images using an unsupervised classification method to extract an initial vegetation profile of the vegetation coverage range in the target area comprises: The remote sensing image is preprocessed in view of the image noise characteristics of the target coastal salt marsh area caused by tidal disturbance and water reflection, the salt-and-pepper noise and strip noise in the remote sensing image are removed through an adaptive filtering algorithm, and radiation bias correction and geometric fine correction are completed by combining regional terrain data and a tidal correction model, so as to ensure the accuracy of image spectral information and spatial position; For the preprocessed remote sensing image, an improved K-means unsupervised classification algorithm is used to perform hierarchical automatic clustering analysis on the ground object types in the preprocessed remote sensing image by dynamically adjusting the clustering center and the iteration threshold, and the vegetation coverage area and the non-vegetation coverage area are distinguished according to the spectral response difference of the NDVI vegetation index; From the clustering results, the vegetation coverage area corresponding to the pixel set is screened out by setting the vegetation spectral feature threshold interval, and the morphological filtering is used to remove the discrete noise pixels, and the spatial distribution range of the pixel set is determined as the initial vegetation contour of the vegetation coverage range in the target coastal salt marsh area.

4. The method of claim 1, wherein, The NDVI threshold method is used as the core to construct a phenology decision tree model, the node level and judgment priority of the decision tree in the phenology decision tree model are optimized by integrating a machine learning enhanced node division mechanism into the phenology decision tree model, and the ability of the phenology decision tree model to identify the difference characteristics of each type of vegetation in the withering period is strengthened by using the phenology period slope analysis method, and the method further comprises: The tidal table data of the target coastal salt marsh area, the real-time monitored tidal level data of the hydrological station, and the salinity monitoring data of the target coastal salt marsh area are preprocessed to remove outliers and fill in missing values; Based on the preprocessed tidal table data and the tidal level data, the correlation between the tidal level and the vegetation type distribution is analyzed, the tidal level division threshold is determined according to the growth characteristics of the Suaeda salsa and the Spartina alterniflora, and the tidal level threshold node is added to the constructed phenology decision tree model; wherein the tidal level threshold node is set to preferentially classify Suaeda salsa in the low tidal level exposed area and preferentially classify Spartina alterniflora in the high tidal level area; According to the influence relationship of the tidal height and the salinity on the net exchange amount estimation of the ecological system, and combining the difference analysis of the vegetation distribution measured data and the model preliminary classification results under different salinity gradients, the corresponding vegetation salinity correction factors and applicable salinity intervals of Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus are determined, and a three-dimensional corresponding table containing the salinity interval, the vegetation type and the correction factor is formed; the salinity correction factor is integrated into the classification result output layer of the phenology decision tree model, so as to optimize the classification accuracy of the phenology decision tree model.

5. The method of claim 4, wherein, After the salinity correction factor is integrated into the classification result output layer of the phenology decision tree model, the method further comprises: Based on the added tidal level threshold node, the determined salinity correction factor and the original vegetation classification features in the phenology decision tree model, a three-dimensional decision rule containing three dimensions of tidal level, salinity and vegetation type is constructed, wherein the original vegetation classification features include one of the following: NDVI threshold, phenology period slope, and the three-dimensional decision rule is used to represent the corresponding relationship between the tidal level interval, the salinity range and the classification results of Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus. The typical vegetation distribution area of the target coastal salt marsh area is selected as a verification data set, the three-dimensional decision rule is applied to the model classification process, the measured vegetation type in the verification data set is compared with the output result of the phenological decision tree model, and the classification accuracy is calculated; According to the accuracy calculation result, the division threshold of the tidal level threshold node and the numerical value and applicable salinity interval of the salinity correction factor are iteratively optimized. If the vegetation classification error in a certain tidal level interval exceeds a preset value, the tidal level threshold of the target coastal salt marsh area is adjusted. If the classification deviation is large in a certain salinity range, the salinity correction factor of the corresponding vegetation type is corrected. Until the model classification accuracy meets the preset requirement, the classification accuracy of the phenological decision tree model is optimized.

6. The method of claim 1, wherein, The NDVI data based on the key phenological period is calculated by using a linear regression algorithm. The NDVI change slope of 6-8 months and 8-11 months is calculated. The slope value of 8-11 months is used as the core dynamic index of the phenological period slope analysis method. By comparing the difference between the NDVI decay rates of Spartina alterniflora and Phragmites australis in the withering period, an adaptive threshold is set. The adaptive threshold is embedded in the decision tree as a key verification node to form a secondary check on the feature optimization result, including: The NDVI data after preprocessing of the key phenological period is subjected to time series smoothing processing to eliminate short-term fluctuation interference. The NDVI change slope of the growth peak period of 6-8 months and the withering period of 8-11 months is calculated by using a weighted linear regression algorithm. The weight of the NDVI data of November is the highest in the calculation of the withering period slope to highlight the late decay characteristics. Based on the withering period slope distribution characteristics of Spartina alterniflora and Phragmites australis in the historical vegetation classification sample library, the slope interval is divided by using a dynamic clustering algorithm. Different adaptive thresholds are set for different vegetation types, such as the withering period slope threshold of Spartina alterniflora being-0.02 / month to-0.05 / month and the withering period slope threshold of Phragmites australis being-0.01 / month to-0.03 / month, to form a threshold system adapted to each vegetation type. The threshold is embedded in the decision tree model as an independent verification node to cross-check the preliminary classification result after feature optimization. When the sample slope value exceeds the threshold interval of the corresponding vegetation type, the secondary discrimination mechanism is triggered. Based on the checked withering period slope value, the vegetation carbon density parameter is dynamically allocated to build a slope and biomass decay correlation model. For Spartina alterniflora and Phragmites australis, the calculation method is to multiply the basic value of carbon density by the product of the normalized slope value, wherein the greater the absolute value of the slope, the higher the decay coefficient and the lower the carbon density value. A slope anomaly early warning mechanism is also established. When the withering period slope of a certain area deviates from the mean value of the same type of vegetation by more than 2 times the standard deviation, it is marked as a suspicious area to be verified in the field.

7. The method of claim 1, wherein, The initial vegetation profile is input into the phenological decision tree model, and the vegetation type is classified by the phenological decision tree model according to the initial vegetation profile to distinguish the vegetation coverage range of Suaeda salsa, Phragmites australis, Spartina alterniflora and Jiangdi in the target coastal salt marsh area, including: The initial vegetation profile is converted into a model input format on a pixel scale, combined with the pre-processed key phenological NDVI data and vegetation index combination data, and input into the phenological decision tree model; A multi-level classification process of the model is started. The first level classification is based on a dynamic initial NDVI threshold interval for rough classification, removes interference pixels that do not meet the basic vegetation spectral characteristics, and selects vegetation pixel sets containing Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus as the first level classification result. The second level classification calls a machine learning enhanced node division mechanism. According to the weight order of the core features, the key phenological period and index combination are selected by fusing a dynamic weight allocation random forest optimization algorithm. The growth characteristics of coastal salt marsh vegetation in different phenological periods are combined to assign dynamic weights to the vegetation index of the key phenological period. The key phenological period and index combination selected are used for dynamic classification of the first level classification result to preferentially distinguish vegetation types that meet the set conditions in terms of NDVI feature differences. Spartina alterniflora in the high NDVI value interval and Suaeda salsa in the low NDVI value interval are obtained as the second level classification result. The third level classification enables a phenological period slope verification mechanism. For the easily confused vegetation types in the second level classification result, the slope value is compared with the adaptive threshold value to distinguish the slope value of the NDVI change in the withering period from August to November. The pixels with a slope value within the Spartina alterniflora threshold interval are marked as Spartina alterniflora, and the pixels with a slope value within the Phragmites australis threshold interval are marked as Phragmites australis, and the third level classification result is obtained. A three-dimensional decision rule combining the tidal level threshold node and the salinity correction factor is used to spatially verify the classification results of each level. The vegetation pixels in the low-tide exposed area are combined with the salinity correction factor to optimize the Suaeda salsa classification boundary, and the classification weight of Spartina alterniflora is strengthened in the high-tide area. Isolated noise pixels in the classification results of each level are removed through morphological post-processing, and the boundaries of the vegetation coverage are smoothed. The vector data of the distinguished Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus corresponding to the vegetation coverage range are output.

8. The method of claim 1, wherein, The vegetation carbon density parameters corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus are obtained, and the vegetation coverage range corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus is combined to calculate the total carbon sink amount of the target coastal salt marsh area through spatial superposition, including: Sampling points are arranged in a unit grid of a first preset size in the target coastal salt marsh area. Suaeda salsa, Phragmites australis and Spartina alterniflora are sampled in the field, and the number of samples for each type of vegetation is not less than a set number. The basic carbon density raw data set containing the coordinates and carbon content values of the sampling points is generated by collecting vegetation samples and measuring the carbon content of each type of vegetation. The basic carbon density raw data set is spatially matched with the regional scale vegetation carbon density literature data. The spatial interpolation calibration of the field sampling data is performed using the Kriging interpolation method. The interpolation parameters are adjusted through cross-validation to expand the coverage of the sampling point data, so that the calibrated basic carbon density parameters are adapted to the growth environment characteristics of the target area vegetation in terms of spatial distribution. Based on the interpolated basic carbon density parameters, a multi-dimensional correction matrix of the NDVI change slope in the withering period from August to November, salinity monitoring data, tidal level data and correction coefficients is established. The difference parameter calling model is constructed based on a multi-dimensional correction matrix, and the difference parameter calling model includes: a Spartina alterniflora model, a reed model, and a Suaeda salsa model; the Spartina alterniflora model takes a basic carbon density value as input, and outputs a result of weighted calculation of a normalized slope value taken from a pixel slope label and a tidal level correction coefficient taken from a tidal level attribute label; the reed model multiplies a correction coefficient corresponding to a salinity interval label by a basic value; and the Suaeda salsa model determines a reference value floating multiple according to a low-tidal-level exposure time label, and forms a three-dimensional carbon density parameter library containing data correlation rules; The vector data of the vegetation coverage range of the Suaeda salsa, the reed, the Spartina alterniflora, and the Rhizome are converted into pixel grids of a second preset size, the pixel slope label, the salinity interval label, and the tidal level attribute label of the pixel are matched with the three-dimensional carbon density parameter library through spatial coordinate correlation, the corresponding model is called to complete pixel-level carbon density assignment, and a carbon density grid data set is generated; Based on the carbon density grid data set, a hierarchical weighted summation algorithm is used to calculate the carbon sink amount, wherein the pixel weight is determined by weighted fusion of the classification confidence and the August NDVI value; A sampling area with a preset proportion of the total area is selected, the measured value of the soil profile carbon storage of the sampling area is compared with the estimated value of the corresponding area model, a regional error correction coefficient is calculated and fed back to the carbon density grid data set for correction, and the estimated results including the total carbon sink amount of the target area, the carbon sink contribution rate of each vegetation type, and the carbon density spatial distribution thermal map are output.

9. A coastal salt marsh vegetation carbon sink estimation system based on multi-temporal phenological features, characterized by, The system comprises: An acquisition module configured to acquire remote sensing images of a target coastal salt marsh area at key phenological periods; A classification module configured to process the remote sensing images using an unsupervised classification method to extract initial vegetation outlines of vegetation coverage ranges in the target area; A construction module configured to construct a phenological decision tree model using an NDVI threshold method as a core, to optimize node levels and judgment priorities of decision trees in the phenological decision tree model by incorporating a machine learning enhanced node division mechanism into the phenological decision tree model, and to enhance the ability of the phenological decision tree model to identify differences in characteristics of each type of vegetation at the withering period by using a phenological period slope analysis method, including: Using NDVI vegetation indexes extracted from the remote sensing images after preprocessing as core parameters, setting a dynamic initial NDVI threshold interval in combination with growth characteristics of coastal salt marsh vegetation, and constructing a model framework of the phenological decision tree model adapted to the salt marsh environment; Embedding a random forest algorithm as a machine learning enhanced node division mechanism in the model framework, inputting combinations of vegetation indexes of each key phenological period after radiation bias correction and geometric precision correction, the combinations of vegetation indexes including NDVI and EVI2, autonomously mining weight relationships of different key phenological periods and index combinations for vegetation classification through a feature importance sorting algorithm, identifying core features with the highest classification contribution degree among the Suaeda salsa, the reed, the Spartina alterniflora, and the Rhizome, and dynamically adjusting node level orders and judgment priorities of the decision trees according to weight orders of the core features. Based on the NDVI data of the key phenological period, the linear regression algorithm is used to calculate the NDVI change slope of 6-8 months and 8-11 months, and the slope value of 8-11 months is taken as the core dynamic index of the phenological slope analysis method. By comparing the NDVI decay rate difference of Spartina alterniflora and Phragmites australis in the withering period, an adaptive threshold is set, which is embedded into the decision tree as a key verification node to form a secondary check on the feature optimization result, and the sensitivity recognition accuracy of the phenological decision tree model to the withering period vegetation phenological difference is strengthened to complete the phenological decision tree model. The decision module is used for inputting the initial vegetation profile into the phenological decision tree model, and classifying the vegetation types in multiple levels according to the initial vegetation profile through the phenological decision tree model, so as to distinguish the vegetation coverage range corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus in the target coastal salt marsh area. The estimation module is used for obtaining the vegetation carbon density parameters corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus, combining the vegetation coverage range corresponding to Suaeda salsa, Phragmites australis, Spartina alterniflora and Rhizoglyphus japonicus, and completing the carbon sink total amount estimation of the target coastal salt marsh area through spatial superposition calculation.

Images